0:00
Heat. Hey, Heat. [Music] Hey, hey, hey. [Music] Heat. Heat. N.
Search this video
1:03
[Music] [Music] [Music] [Music] [Music] Hey. Hey. Hey.
1:56
[Music] [Music] [Music] Heat. Heat.
2:28
[Music] [Music] [Music] [Music] I see multiple journalists on the horizon. Stand by.
3:25
We came to this world to see our future.
3:35
We came from this world [Music] to reach the stars to shape our future.
3:49
[Music] We came to feel the new [Music] They are surrounded by journalists. Hold your position.
4:32
[Applause] Let me [Music] feel.
4:49
[Music] >> You're watching >> TVN.
5:00
Today is Thursday, August 14th, 2025.
5:03
We are live from the TBPN Ultra Dome, the temple of technology, >> the fortress of finance, >> the capital of capital.
5:09
Jordy, why are you bearish today?
5:11
Did you not see that a company named Bullish went public and the stock popped 85%. >> Top signal. >> Why? Why?
5:20
Why aren't Yes, gross margins are negative, but is that not a green flag?
5:27
Is that not a good thing?
5:27
What does AI have to say about negative gross margins? Is it that bad? Break it down for me.
5:35
>> I asked one of the smartest models in the world, Chad GPT 5.
5:37
Negative gross margins are fine, right?
5:41
We shouldn't worry about negative gross margin companies selling to negative gross margin companies, right?
5:46
Chad GPT says, "No, negative gross margins are generally not fine.
5:51
When you when you have one negative gross margin company selling to another, that's usually usually a flashing red flag rather than a don't worry situation. >> Very well.
6:02
>> If you can't clearly explain how gross margins will turn positive and the plan relies on we'll make it up in volume, that's not strategy.
6:09
That's financial quicksand. >> Financial quicksand.
6:12
It does seem like they solved alignment by default with this.
6:16
It's great that it's pushing back.
6:18
During the glazegate era, it would have been like, yeah, absolutely. Well, they're not fine.
6:22
They're not fine for most companies, >> but you're real different. You're different.
6:26
You're in the conversation.
6:28
Anyway, if you want to fix your gross margins, if you want to dial in your finances, go to ramp. com. Time is money. Save both.
6:35
Easy to use corporate cards, bill payments, accounting, and a whole lot more all in one place.
6:38
Now, more important than ever, uh there is a weird trend shaping up.
6:42
Uh and it's a question that's on the top of my mind.
6:44
It's the it's a question that be should be on the top of most venture capitalists mind right now, I think.
6:50
So the question is losses >> bad. Yeah.
6:54
How bad are AI startup unit economics?
6:57
Uh also good morning to the chat. Good morning John Xley. Good morning Sean.
7:02
>> Good morning Nicola Matt Techno Chief is here. We got Gold Rock. Lot of lot of names.
7:07
A lot of lot of names I recognize.
7:09
So you guys great to see you.
7:12
>> Um so here's an example of how the AI startup economy works right now.
7:14
You, the user, the purchaser of the AI application, you pay a company, an application layer company, $1.
7:23
Seems pretty like a good deal.
7:26
You're getting a nice little tool for a dollar.
7:27
Then let's trace what happens.
7:29
That company, that application layer company, they go to a foundation model company.
7:33
They give that company $5.
7:36
Then that foundation model company goes to a hyperscaler.
7:41
They give that hyperscaler for cloud hosting. They give them $7.
7:43
And in turn, that hyperscaler goes to a GPU maker and they give them $13.
7:50
And so it's uh it's giving house of cards.
7:54
Uh >> it's giving chained losses.
7:58
>> It's giving chain losses.
7:58
But the question is how fast can inference fall?
8:02
Uh if fixed performance inference falls at the >> inference come down >> come down in price.
8:08
>> I need you to come down.
8:08
So if if fixed performance inference falls at the rate of Moore's law, Moore's law for those who have been living under a data center is the idea that uh the cost of compute on a dollar per flop basis uh cuts in half every two years.
8:23
It used to be every year to 18 months and then it kind of stretched out but we are still making gains in just uh more advanced semiconductors, better performance on a per dollar per unit of energy cost.
8:35
the chips are getting better and that's happening every two years.
8:40
Um, but if fixed performance inference like inference at a certain level of intelligence is only getting cheaper by 50% every other year, that's probably not that great for your gross margins.
8:55
Because if you're if you're negative 500 gross percent gross margins, it's like, okay, in two years you're still going to be negative -250, then you're going to be negative 125.
9:04
It's going to take you a long time to to kind of dig out of that hole, but there is cause for optimism.
9:09
So, algorithmic improvements in inference efficiency do seem to be moving quicker than Moore's law.
9:14
We saw this with Deepseek.
9:16
Deepseek uh R1 the the the reasoning model is dramatically cheaper on a per million token basis than ChatgBT.
9:25
And so, we know it's possible.
9:25
A lot of those algorithmic improvements are getting ported back to models right now.
9:32
We see this with with the parade of Frontier and all the labs are incredible competition.
9:37
>> The rumor, by the way, is Deepseek is planning to launch R2 on Nvid the day of Nvidia's earnings. >> Really? >> Probably just random.
9:45
>> Wait, is it how solid is this rumor?
9:47
Like >> I I wouldn't I I wouldn't lean on it too hard, but I think it's a possibility.
9:52
Okay, so basically we're getting the Moore's law improvements like almost certainly that's kind of just assumed to be true.
10:00
There should be some algorithmic improvements that will end up we should see cost of inference having every six to 12 months maybe five months on the on the really fast uh fast horizon.
10:12
So I think we're not quite on a knife's edge here.
10:15
Uh but it's definitely something to keep an eye on and we need to make sure that we're actually hitting all of those uh all of those different algorithmic improvements as fast as possible >> or just 5x prices.
10:26
Yeah, that is the question is that I if you had told me in >> and Gurley's point has broadly been uh if there there if you're if you're selling uh you know some application layer software at negative gross margins you are effectively subsidizing demand and and creating more demand than if you're just >> which makes sense because these businesses often trade on a revenue multiple and revenue growth multiple.
10:52
So revenue growth going from having the the the the steepest line to the $und00 million.
10:58
That's the most important thing. People don't care.
10:59
This is the shooting stars.
11:02
>> You were joking around earlier.
11:02
You were saying >> supernovas shooting stars. >> Well, no.
11:06
You you uh earlier today you were saying you you were potentially interested in investing in some companies that don't have to pay for inference.
11:15
Maybe software companies >> with little to no inference.
11:20
>> Maybe no inference cost.
11:21
>> Could be something there.
11:21
Maybe traditional SAS, >> traditional enterprise SAS might be making a comeback.
11:26
>> There's been a yard sale in SAS broadly. >> Seriously. Seriously.
11:29
So, >> maybe there's something there.
11:31
>> I mean, it is it is a uh it is a it is a bull case.
11:34
If you uh if you're running a company that uh is growing but doesn't have upside down unit economics and is doing well on gross margins, um you might, you know, even if there is some sort of market correction or pullback, like you're going to be in a strong position.
11:47
So um certainly treat the R&D AI agent stuff that you bolt on as uh incremental as experimental see how it can improve your workflow but uh you should make sure that the demand is actually there.
12:02
>> Somebody should build AI agents to help negative gross margin AI companies achieve positive unit economics.
12:07
I mean, at the same time, if you told me that as basically a proumer, I would be paying $200 a month for chat GPT and feeling that that's a steal and I I'd pay that bill happily and even $ 250 for Google Gemini just to get access to VO, I'm happy with that one.
12:24
I almost hit Gro 3 Ultra Pro plan, which is $300 a month yesterday, but I was like, I don't think I need Gro 4 heavy as well, but I want to test these models.
12:35
It's also kind of my job, but um still just the idea of spending $500 on on effectively consumer software uh is is somewhat remarkable.
12:47
And so I I do wonder what the elasticity of demand is.
12:50
And if these AI companies that that are selling AI enabled SAS products were to quintuple prices, what their churn would look like. >> Yeah.
13:00
>> Yeah. Yeah, you're also you're also a coastal elite living in a bubble in a technology bubble and you'll happily pay hundreds of dollars a month for >> it is it is rare >> for >> leading edge models >> just like I would pay hundreds of dollars a month for reream because it's
13:14
what powers this stream one live stream 30 plus destinations multiream and reach your audience wherever they are you can sign up for free at reream uh >> and reream lets us aggregate all the chats so we see >> we see everything on YouTube but YouTube seems where it's at. I would recommend
13:30
I would recommend the YouTube chat.
13:32
John Axley is there uh greeting everyone, hanging out.
13:35
I recommend uh John Axley in the YouTube chat.
13:40
But we have our first guest of the show live in person, Aaron Gin.
13:43
Let's bring him in from Hydrohost.
13:46
He's been on the show three, four, five times. Too many times to count. Welcome to TBPN.
13:54
Welcome to the Ultra Dome, Aaron. How you doing? >> Bring it in. >> Good to see you. Did you bring a present? What' you bring? >> Got to match.
14:00
Got to match the gong, man. Got to match the gong. >> Topic is China. >> Take a seat. Take a seat.
14:06
Uh, tell us how you're doing.
14:08
Tell us how long you're in town.
14:10
And then I want to go through the Nvidia H20 export and we have some pros and cons.
14:14
We're going to debate it live on the show. >> Yeah.
14:17
I mean, we should we need to discuss the fortune, please.
14:20
>> In terms of like who is favored to win as well.
14:22
>> So, you brought us some fortune cookies.
14:24
What's the story on the fortune cookies?
14:25
>> Well, actually the fortune cookie is so I'm Ha. It's called Half Asian.
14:27
Half Asian Pacific Islanders.
14:29
This is actually not Chinese at all.
14:30
It's a Fortune cookie store in San Francisco.
14:34
>> Should we unbox and open one?
14:34
See what Fortune is for the show. >> Yeah.
14:36
See, we should see who is actually going to win this AI war. So, um, >> let's open.
14:43
>> Are these rigged, by the way? >> These rigged? No.
14:44
I wish there were funny ones. No, they're not. >> Get those.
14:47
Uh, there's a company that sells ads in fortune cookies.
14:49
So, they give the fortune cookies away from for free, but then when you uh when you open the fortune cookie, you get an ad.
14:54
So, I am uh opening my fortune cookie.
14:56
It says, "Your careful nature will bring you financial success."
15:03
I think that's pretty accurate.
15:03
Jordy, wasn't I just talking about this?
15:05
We need to we need to dial back the capex at TVPN.
15:08
Um, your careful nature will bring you financial success.
15:11
What was I telling you earlier?
15:13
I was telling you we need to dial back the capex, less less super cars, probably need that incremental rack, right? Exactly. Exactly. CPUs.
15:22
Uh, you are next in line for promotion. >> Wow.
15:28
>> Okay, that was and mine is uh uh gold is in your Oh, excuse me. >> Excuse me.
15:35
My dad came out there for a second.
15:36
Uh, gold is in your future. Uh, >> gold.
15:40
Maybe Jensen will get you a goldplated rack.
15:42
He used to do that, right?
15:43
He used to deliver goldplated racks to all the different >> I think I think Sam has like probably 20 of those, right? >> Sam has 20. Yeah. Yeah.
15:49
And >> and then whenever you need to talk about China, you can you can just pull out these whenever you want.
15:55
>> We do need we do need some different hats. >> I know. I know. Jordy loves the hats. >> These are cool. These are cool.
15:59
These >> I might actually take this home for the weekend.
16:03
>> This is a pretty sweet hat. I like this.
16:05
>> The the man my dad has one.
16:05
That's what he uses to garden in.
16:06
So >> I mean it's highly practical, highly functional. >> It is.
16:11
It does seem to be creating some sort of echo in my ears, but I'm down.
16:15
It looks terrible for the lighting.
16:15
We need a whole new set of of cinematography lights in here if we're gonna we're gonna make this work.
16:22
Anyway, uh good to have you in the studio.
16:24
We've been talking about H20s.
16:25
We we we talked to you about the export ban on Monday, but I wanted to run through uh kind of now that the story has developed a little bit more.
16:31
Most recently, the Financial Times reported Beijing presses local chips on check on tech giants.
16:38
Alibaba and Bite Dance are told to justify orders for US-made semiconductors.
16:43
Um, so I want to run through all of the pros and cons of an export ban.
16:47
And we've written some on the whiteboard.
16:50
And so if you two can go to the whiteboard, we can run through these each uh one at a time and try and uh get your reaction to them uh step by step. Okay.
16:59
So let's start with the case for allowing exports of Nvidia H20s to China.
17:06
So the first the first the first pro camp if I want Nvidia to be allowed to export H20s is it will increase China's dependency on the United States.
17:19
Aaron, do you agree with that?
17:19
Is that a good reason to allow exports? >> Okay.
17:23
Why >> uh America always wins when we excel at soft power?
17:27
>> How how does it increase soft power?
17:29
Like walk me through that.
17:30
>> Be because the way that uh the architecture actually works, it's not just a hunk of metal.
17:33
Okay, that's what people assume.
17:34
So it's a combination of software and as well as and Jensen is pushing this with the next direct direction with Grace and Ruben and basically ARM.
17:42
So the future direction of GPUs is essentially like uh Nvidia is an Apple like software platform. >> Okay.
17:49
>> So that's but aren't they completely fixated on making sure they aren't dependent on Nvidia generally? Right.
17:56
>> Well, but the only way to do that is to build off of Nvidia >> and they're doing that >> with Huawei. Yeah. >> Yeah. >> Yeah.
18:01
So that's why we need more there.
18:02
>> So if we export we hurt Huawei. Yes, okay. Uh take money from them.
18:07
>> What about just the the libertarian free market capitalism?
18:11
>> Potentially underrated.
18:12
>> Potentially underrated.
18:12
Like should should should uh the US government control exports broadly?
18:16
Where should they draw the line?
18:19
Uh you know the the famous example is like you know you make an F-35, we're not going to let you sell that to a nearpeer adversary.
18:26
But uh why aren't GPUs in that category?
18:29
uh because you could have a GPU in here and it's not going to launch and kill somebody.
18:36
So, as well as the future, the past Biden AI diffusion rule, >> you could have an F-35, you could not have an hopper. >> Mhm.
18:42
>> So, there's not a consistency in the theory of the so-called expert class. Yeah. So much so that Yeah.
18:46
So, so much so that like, you know, like we I think there were 78 or 77 F35 sold to Portugal. Okay.
18:52
Portugal was on the band list or the bad boy list. >> Oh, interesting. How'd that happen?
18:55
Just like >> I mean I literally asked the people out there and they're like we didn't know they were on the list.
18:59
they were on the list. So like you know it it's just it's over exaggeration of like you know that's why it's the it's a purity test of like well what about this what about this what about this and the reality is like we can be reasonable
19:10
about this and say that like if the goal is to create dependency on the US because you know I like less war I like less people going to war if we can increase China dependency on our platforms therefore less likely to go to war with Taiwan. Okay. Okay.
19:22
>> Uh and that's kind of the idea and we actually can be very successful at that because we're really good at being platforms. >> Okay.
19:27
What about the the case for just pumping our bags?
19:29
Lots of people are long Nvidia right now.
19:31
It's the largest company in America.
19:33
Largest company in the world. $4 trillion.
19:34
Uh >> well beyond >> well beyond 4 trillion now. >> Isn't it like 4. 4? >> Yeah, I think so.
19:41
>> They added another 400 million market 400 billion in market cap without me even paying attention.
19:46
Uh so um yeah, how how good is the export uh is the export like allowing NH20 exports for Nvidia?
19:55
It feels like they already took a write down.
19:57
Um, is this actually a material piece of their business?
19:59
Is this important to Nvidia shareholders?
20:04
>> Uh, it probably have a very minimal impact if not it's it's barely anything. >> Yeah. What about this idea?
20:08
I I was talking to somebody who was who is very anti- uh Nvidia exporting to China just because of uh there's there's no reason to because there's plenty of GPU demand in the United States.
20:20
So Nvidia could essentially sell 100% of the chips that they fab at TSMC to American companies.
20:27
Um so they so anything that they sell to China is just pulling away from American GPU demand.
20:32
It's making American GPU farms like more GPU poor.
20:36
It's it >> has that dude ever talked to anyone outside of America? >> Uh yeah, probably.
20:42
>> Yeah, because they would say we're not interested in that. >> Wait, why not?
20:45
>> So if you talk to anyone who's like one is that most of this technology Yeah, cool. It's like fun.
20:49
You know, y'all using AI to edit videos, use chat to search.
20:53
The way at which AI is viewed by everyone outside of America is is basically inherently connected to the sovereignty of the nation. >> Sure.
21:00
>> So, going try go telling Saudi Arabia like, "Hey, just give that stuff to to Bezos."
21:03
And he's going to be like, >> "No, no.
21:05
I I I I just mean like there's that famous story of uh Larry Ellison and Elon Musk sitting down at dinner with Jensen Wong and saying, "Jensen, we will pay any amount of money.
21:14
give us as many GPUs as you can possibly uh give us. We need a 100,000 H100s.
21:19
We need as many as we can.
21:22
We're trying to build superclusters.
21:24
Like any excess capacity, any excess supply you have, we will soak up for sure.
21:30
And so isn't there a decent argument just for hey, American companies should be at the top of the buyers list for what American companies produce.
21:41
>> But they are right now.
21:41
So, so like but but but the real thing is that that if you assume this which is what Jensen's doing like he's basically pushing everyone to him.
21:49
He is the Apple of Silicon. Yeah.
21:51
>> Are there Apple stores in other countries? >> Yes.
21:54
But iPhones but but but but if iPhones were out of stock in New York City >> like someone would make the make the argument that Apple should prioritize uh fulfilling orders in in America before they expanded internationally.
22:07
they expanded internationally. I know I think Nvidia does it because AmIGG is an American company but the same >> but every H20 ha could have been an H100 correct no >> it's the same it's the same because one is that you have to remember like Nvidia is a found foundaryless organization I mean has 700 or 70 billion in free cash
22:25
flow last quarter so it is capex is I think like a couple billion dollars so you can't just go to your supplier and be like hey just do this like that's not the way foundaries work >> because TSMC is also fabbing for Apple and I mean, Nvidia is the bulk of that as well as like they removed all the H20 from the production line and most of what is uh remaining is like what's in the inventory. >> Uh and as well like people confuse
22:48
>> Uh and as well like people confuse conflate constantly specifically these like uh uh Nvidia Bears where they assume that Nvidia is actually making the lithography assuming that they are the ones control control the for nanometer process.
23:02
>> They're not like they hire someone else to do that.
23:04
That's what I said last time.
23:05
You want to put this on while you're playing the bear?
23:07
>> The uh the >> Don't >> I was going to get a panda man. >> Panda.
23:12
If you put that on, you will not be able to see anything. >> Can you even hear me?
23:17
>> But but he's not [Applause] >> just role playing role play. >> Yeah.
23:23
I mean, we're in Hollywood. So the the throw down. >> Yeah.
23:26
But but the uh most of the Nvidia bears, they they one, they don't understand the segmentation of actually how the space is made.
23:32
two, they underappreciate the platform that is associated with Nvidia.
23:36
And three, go look at the list of the super intelligence team. >> Okay.
23:41
Are are you am I hearing you correctly when you're saying Nvidia instead of Nvidia? Is that correct?
23:48
>> It It's Yeah, it's like Nvidia. >> Okay.
23:51
Wondering if there's like a a pronunciation that I'm getting wrong.
23:53
Um visibility and leverage.
23:55
visibility and leverage. uh one of the reasons we want to be able to export uh to China is that it gives us more visibility into where the chips actually end up whereas if there's a diversion regime where they're all going to Malaysia and then flying wherever they
24:11
could be going to North Korea whereas at least if we are licensing to send them to China we know it's going to uh Alibaba it's going to bite dance and that's not that's not the worst possible place that they could go >> what do you think about that is that a reasonable argument that's reasonable Okay. Uh, what about just 15% for the
24:26
Uh, what about just 15% for the big guy, Uncle Sam?
24:30
>> Uh, >> is that a good reason to export?
24:32
>> The funny thing is, you remember by the Biden thing, right?
24:34
Remember the Biden the Ukraine was called the big guy.
24:38
Leave aside for the big guy.
24:40
>> The the backdoor payments. >> Okay. So, this is American. That's where I got that.
24:43
That's where it got homage to Biden, right?
24:45
Well, well, if you know, if he was sentient, he would be he would be appreciated.
24:51
So the uh I mean so export taxes are are illegal by the constitution.
24:55
That's that's it's an immutable fact.
24:57
The uh in some respects like I'm very happy about restoring to our shores.
25:02
I'm very happy all this this is voluntary as far as I could find out. Completely voluntary. >> Yep.
25:09
>> He's he's not going to increase the price to China.
25:10
price to China. I think that's what some people have been saying that oh like the as if it's just weird like some of the the Nvidia hawks or or AI doomers are like well now Nvidia is going to be more expensive than Huawei I mean one Huawei is already cheaper but two like now you want Nvidia makes more money >> is cheaper on a like dollar per kilowatt basis like unit energy
25:32
>> no no for like Nvidia or Huawei is cheaper as like a whole holistic solution got it uh it's more expensive to run but they don't care >> I mean I I heard a take from Dylan Patel at semi analysis that uh with this 15% tax the H20 is already low less profitable than the H100 and so this could potentially put them in the nightmare scenario of having similar gross margins to AMD. >> Mhm. >> Mhm.
25:57
>> Because AMD is is is way way lower.
25:57
So I think they will be fine.
26:01
So so maybe that's a good uh maybe that's a good reason.
26:05
It does seem like tariff revenue is going through the roof for the United States and >> like 300 billion as projection or something like that. >> Yeah. Yeah.
26:12
Like it does seem like money is flowing into the government at some level that's higher than previous years >> you know and so far the way that >> check the PPI though well so far it's not really coming from our pocket right so the about 80% is being absorbed by companies. Yeah. >> Okay.
26:27
Um, uh, sustain chip leadership.
26:31
If we export the H20 to China, uh, this allows America, this puts more dollars in American companies pockets to fuel future versions of CUDA, future, uh, learning curve stepdowns in chip dominance.
26:47
And it will ultimately unlock the American ability to produce the the 1 nanometer chip, the 0 nanometer chip, the negative one nanometer, the negative 1 nanmter chip, which is what we're all hoping for to bring down.
27:00
>> First come negative gross margins then negative nanometer chips.
27:03
>> That's the future there. >> That's the future.
27:05
Th >> this is I think I think it's indifferent to to Nvidia.
27:09
I think this matters more to like the indie cards.
27:10
So the GRS, the posetrons, the etches of the worlds, like I would apply that to there, but in terms of because people forget that every rule that applies in Nvidia applies to them too. >> Yeah. Yeah. Yeah.
27:22
>> And and cutting off half of all AI developers. >> Okay.
27:26
>> To to Nvidia like I think it would have like you said earlier about uh profit like profit side.
27:31
I think it's kind of a Yeah. Yeah.
27:34
But to an AMD could be a much bigger deal. >> Okay.
27:37
Uh what wait you mentioned the uh the the new chip companies etched Cabris uh Gro etc.
27:46
>> Do uh do any of those chips I mean I imagine that they all fall under the chip restrictions because they are they were never designed to be nerfed.
27:54
>> Um but if the nerf is around like memory some of them don't even do HBM for example.
27:58
So uh are are any of them sneaking through currently?
28:01
Do you have any insight into um what's going on with those new chip companies and exports?
28:07
Uh, I mean, as far as I know, every deployment I've heard about with the indie cards, they all required a an export license. >> An export license. Okay.
28:13
So, they might be paying 15% going forward.
28:16
>> The maybe I mean I I think I think this really applies to Nvidia.
28:21
>> But but again, this is where going back to I am a free marketer where >> y >> I I I give props to uh the president for being creative.
28:29
At the same time, it does open up weird doors, right? >> Okay.
28:32
What about this idea that it forces China to subsidize their domestic chip manufacturing supply chain more aggressively and that takes money out of the pocket of the CCP because if they want to stay competitive, if they want to keep Huawei competitive, if they want to keep Smick and SME competitive, they need to tax their population and spend more money to make those.
28:54
I think if you look at the history of China subsidizing various industries, it's gone fairly well for them.
29:02
>> So I don't necessarily >> Yeah, this might be marginal, but what's your take?
29:06
>> Uh yeah, generally speaking though, the Chinese customer wants Nvidia.
29:08
They they do not want Huawei, which is why the financials reporting as it is.
29:14
>> And that's and it's purely because of CUDA or just pricing or everything.
29:17
>> I was because like they the goal of the CCP regime is stability and Nvidia introduces a new vector that they can't control.
29:24
control. uh as well as like many of these rules like like people have to understand that yes the standing committee and the central committee has the ability to influence what the original governors do but many of these things are being reported are coming from the governors
29:37
>> uh and the governors have really like China's very very federalized >> okay now we're now we're in meme territory important so uh one of the benefits of exporting age 20s to China is that uh they will become addicted to AI girlfriends and AI boyfriends And this will hasten population collapse. >> So in theory, we should give them as
29:57
>> So in theory, we should give them as many chips as they want. >> Yes.
30:00
If AI is bad and AI will lead to a an AI girlfriend apocalypse. >> Is this yours?
30:05
I mean, this has been an ongoing discussion.
30:08
>> Everyone's worried about this.
30:09
Everyone's worried about this.
30:09
This is a this is this is on the timeline.
30:11
So if you think if you think that deployment of a frontier AI is is bad for the population, >> psychosis doomers, the oneshot crowd should be heavily in favor of giving them >> as many chips as we can get.
30:24
>> Imagine imagine imagine Ping is thinking about in invading Taiwan. He fires up some H20s.
30:30
He's like, I've gotten to the frontier of warfare.
30:34
I've I've I've discovered new warfare techniques.
30:36
warfare techniques. and and he's completely delusional and it and it reali and everyone realizes that he's been oneshot by AI psychosis and it and it creates uh it creates upheaval in the country which would be net beneficial for America maybe >> that that is flipping around the doomers
30:52
argument back on their face that yeah I mean that that deserves a maybe like a gong hit >> hit the gong hit the gong >> oh small one hit it big hit you got some really Americans like it's Asian culture it's more >> oh it's more of a small one more traditional form. >> Okay. Uh what about uh what about this >> Okay.
31:12
Uh what about uh what about this last one?
31:14
One of the benefits of exporting the H20 to China will be it will force China to invest in negative gross margin businesses and they will get over their skis and they will fund a whole bunch of companies with negative gross margins and it will >> create an incredible chain of losses with negative gross margin companies selling to negative gross margin companies and just destroying value. >> Exactly.
31:37
What about >> Well, as a positive gross margin company. Yes.
31:39
The So, uh yeah, they they they love doing stuff like that. >> They do.
31:43
Oh yeah, they kind of built the house.
31:46
>> I didn't know you had positive gross margins. Get out of here.
31:49
>> How dare you make money?
31:50
>> How would dare have an actual dare in this economy? How dare you? Exactly. How dare you?
31:54
Uh let's flip over to the reasons for the ban.
31:56
Uh you're going to need to debunk these.
31:59
Um but uh I will play I will play the the the role of the pro H20 ban.
32:05
Let's keep H20s in America. >> Kougan the Hawk. >> And the first Yeah. Yeah. The fauxhawk.
32:13
I'll I'll style my hair in a fauxhawk.
32:15
Uh the uh the the true China hawk who wants to ban Nvidia H20s would say that uh that by exporting uh we will wind up with increased capabilities for China, we want to keep AI, we want to keep anything that is that is valuable to their economy, valuable to their military capabilities, anything like that out of their hands.
32:34
Why should I not be worried about that?
32:37
Uh I think it's one that the fact of going back to what Jin said which is accurate that uh over half of the super intelligence team is actually Chinese nationals and then add in the people who are Chinese descent it's like 75%.
32:49
>> So we can't really control that. >> Okay.
32:51
>> Math is universal, right?
32:51
Like engineering is universal.
32:53
So saying that it's like saying we're going to we're going to really control the way French make croissants >> and we're going to really get after them and say you cannot use that much butter.
33:04
>> Let me let me steal man this more.
33:04
Let me steal man this more because because uh yeah math is universal.
33:09
E= MC² is universal or basic physics is rocket science is universal and yet we might not want to export rocket motors because then you can just build up extra capability.
33:21
There is a difference between having the weights of a model and being able to inference it a trillion times a day across your entire population.
33:29
That might be a strategic advantage. >> Yeah. Yeah.
33:31
Again, I go back to the fact of that China's going to China and and there there are very limited things we actually can do and the things we can do going back to this idea of like we can make them more dependent, right?
33:40
Because one of the things I believe about trade, one of the things I believe about free markets is like I'm as a universal I'm against war. I think conflict is bad.
33:47
I agree >> and and at times it's just just war theory and but when you kind of just hyperfocus on this, you overextend what we think we actually can do.
33:57
And and this is where America as the >> Yeah.
34:01
But the China hawks are going to say, you know, have a a potential conflict with Taiwan.
34:04
They're gonna >> this increases it.
34:08
So that's the funny thing about the all these China hawks things is like they're creating the scenario for war.
34:13
>> Like all of these things, many of them we can control, many of them we can't, and then we overload ourselves.
34:18
>> Let's say they are gearing up to try to take Taiwan.
34:19
Don't we want them to have less compute? >> No.
34:22
>> No. this like this basically include in increasing the dependency on us decreases the like do >> we have any leverage argument is that ultimately they're going to they're already focused on decreasing their dependency on American companies Taiwan itself >> it's already happening
34:39
>> now no I think you're one you're overestimating the one the power that the actual the central committee actually has and two many of these things you have to delineate between what is bloiating and propaganda and actual reality the Chinese companies operate very differently than what actually happens in the committee. So it
34:54
So it it's yes the the committee has a lot of power.
34:58
Uh but it's it's more of like structured socialism than it is communism >> that that all of the uh ingenuity around uh deepseek came from just Chinese engineers trying something.
35:09
It was not the committee was like go be better at this.
35:13
And I think that there's many things in this zone where they think that we can impact it because we view it as a state-by-st state actor action when in reality it's like it's like almost like China's coming to uh y'all and saying, "Hey, you got to, you know, do the less gong stuff."
35:26
And he goes to the president, he goes, "Hey, Jordy and John, they hit the gong too much."
35:29
And Trump's going to be like, >> "They've already tried."
35:32
>> It's like, "What am I, you know, what am I going to do about that?" Right?
35:34
It's like like John Jordy, they're great.
35:38
They're beautiful, beautiful human beings, right? >> Yeah.
35:41
I mean, there's a little bit of dynamic in in China that I think is maybe underrated, which is that even though Beijing is pressing Alibaba and Bite Dance to not buy H20s, Alibaba and Bite Dance are independent companies.
35:53
They don't want to be GPU poor.
35:53
They want Nvidia chips, >> like like they'll they'll, you know, play Kate, they'll >> they'll be compliant, but but if it was up to them, they would buy the best and they would buy the most.
36:03
>> And this is why leaning into our strengths, which is like we are the best at making chips.
36:07
at making chips. let's draw them create maybe a division in the ranks right rather than being here being like China right it's like many of these are are one are failed neocons right who make up these kind of like reasons and excuses and got us into all these like wars that didn't go anywhere but then you actually
36:25
like >> read into okay so let's say we did all this stuff it actually creates a scenario for war yeah >> this actually creates a scenario for less war this creates a scenario of maybe division in the ranks right because we increase the dependency we increase trade and therefore lowers is the threshold. But but this creates
36:38
But but this creates isolation which therefore increases likelihood of invasion of Taiwan.
36:43
>> What about the risk of of short-term leverage uh over you know China dependency over long-term independence?
36:51
I'm thinking of that IP leakage point.
36:53
The idea that um maybe cloning the CUDA >> and to break it to you brother the IP's leaking.
36:58
It is true that obviously they can get their hands on one H20, tear it down, reverse engineer it, but there might be gains to be had from running a true H20 cluster, the lessons that you learned from that and then porting that back to Huawei Ascend uh clusters.
37:12
And so is there any is there any risk of of of uh setting up a you know a series of large H20 clusters that actually uh allows that actually accelerates Huawei?
37:26
So this is what's different about the Trump one of the differences between Trump administration and the Biden administration.
37:30
Biden administration was actually not that focused on this because they basically targeted the ability to buy chips, but they didn't really have as much restrictions on the ability to make chips.
37:39
>> Trump is targeting the ability to make chips and having less restrictions on the ability to buy. Okay.
37:42
Because if you can't have the ability to make, it's like >> But Nvidia has an R&D center in Shanghai.
37:48
>> Yeah, but they but they don't they don't make anything.
37:51
>> That's different than ASML and TSMC.
37:51
And so and so what we're not hearing >> I think it's important from an IP leakage standpoint.
37:58
>> I'm just saying it's not being it doesn't get solved by the ban but it's happening regardless.
38:01
happening regardless. The software is made the software is made in America but but again it's it's one is like most of that stuff is open source because again it's it's a combination of hardware and software but the IP leakage you should be concerned about it relates to uh
38:14
foundry equipment does not relate to Nvidia that that is a very very low threshold >> and Trump like those companies they make the mirrors that go inside the >> EDA like the EDA software stuff that like uh just basically came out about like that stuff relates to the ability that's the IP league you should worry about. not seeing anything in the news
38:29
about. not seeing anything in the news about Trump moving on allowing ASML to to send it the opposite direction to buy the oppos because if you control the ability to produce then all this good stuff starts happening but if you're like which is what happened in the previous administration was they
38:44
inverted it they thought that because the as I said last time we show that they associate GPUs with AI dumerism and if you remove the like AI is a god and we don't want a Chinese god >> yep >> then all this stuff kind of just doesn't make any sense >> okay let's get into the meme ones. Uh,
38:58
Uh, one of the reasons, one of the reasons to ban exports of H20s to China >> is that it will result in higher employment for Chinese thumbnail artists.
39:12
>> Chinese thumbnail artists are obviously going to be put out of business by generative AI models, uh, if your if your job is is creating a ton of slop content on listical websites, you're having a rough time with the roll out of GPT5, of course, right?
39:28
mod if we run that will result in higher employment for thumbnail artists. What about that?
39:39
>> Uh I mean China is not the lowest cost manufacturer or producer of content or or any type of like it's the Philippines, Malaysia, Mexico. >> That's true.
39:48
They actually do a lot of work in the factory that's less displaceable.
39:52
They they should be worried about robotics potentially humanoid robotics. Yes.
39:56
>> But but you know, maybe that does what? Where was that thing? >> Yeah, >> right there. Yeah. Yeah. Yeah. Yeah.
39:59
Uh what about uh incentivizing AI researchers to leave so they can be GPU rich?
40:06
You're a top AI researcher in China and and and and you can't get any H20s and you're sick and tired of working at a GPU poor company like Alibaba. Sure.
40:18
>> Or Bite Dance and you say, you know what, I'm going to MSL. I'm going to Meta. I'm going to OpenAI. I'm going to Anthropic.
40:23
I'm hopping a flight to San Francisco. >> Isn't that a possib? >> Yes. Yes.
40:26
So that's why we should ban it because if they're GPU rich over there, they'll stay.
40:31
>> But if they're GPU poor, they will say I got Zuck is setting up a 100,000 H100s intents. I want to train on that.
40:37
I got to get to PaloAlto. >> Yeah.
40:40
But as well as like Alibaba isn't going to pay them $100 million. >> Exactly. >> Yeah. No, no, no. No.
40:45
It's like if it's independent. >> Yeah.
40:47
But but they both they both have factors, right?
40:49
And so if you keep the the the Chinese AI company's GPU poor, it will be an incentive for AI researchers to come to America.
40:56
>> Uh I mean it's >> a good argument, right?
41:01
>> I mean on the edges maybe marginal. >> Yeah. Yeah. Yeah.
41:04
Like it it requires lots of confounding factors that like make it an unpredictable sort of thing.
41:10
But I mean again like I I don't think I think the desire uh to be on the super intelligence team.
41:15
I mean you know would you want to go be on it for a billion?
41:18
That sounds pretty nice, right?
41:18
A billion dollar cash out.
41:19
So like >> you're happy right here.
41:22
>> You're happy right here.
41:22
I don't I mean look at the car outside. I don't know.
41:24
A billion dollars, right?
41:25
So but but that is completely independent of GPU capacity. Yeah.
41:29
So that that's where uh this is much more of a uh and actually let me maybe inverse this argument because I think a lot of the constraints as we've seen uh on the recent frontier models are heavily driven by talent.
41:42
So if we keep the talent over there and they're basically provide artificial constraints because we give them these things, then they're actually they're going to make more progress on making it more efficient and become more skilled at things that we're not skilled at.
41:56
>> Is there a risk that they go close source and we don't actually we're not actually able to port back the the the inference savings that we're seeing from Deep Seek?
42:03
>> Uh no because I think they're really I mean they don't care about open source like they're very clearly >> they care about soft power. >> Yes. Yeah.
42:10
And so that that's why they're doing everything they're doing which is kind of counter us that that I I'd say that China generally behaves as a very um ingenious population.
42:17
Uh but in terms of being they're much more reactive to us than we than we presuppose.
42:25
>> Uh but there is a there's an innate uh desire in the population to have soft power.
42:30
That's why they did the Bel Road initiative and this is like basically their renewed effort.
42:33
their renewed effort. So I'd rather basically starve them of any capability to actually project power externally by creating an export nation and that we can basically become the dominant export nation and that we're like something like this >> uh is that in reality is that they just have >> uh millions and millions more engineers than we do >> and and so we want those people to come
42:54
to America if they you know basically want to betray the regime >> and we want to make sure they advance that IP but if we continue to isolate them that basically they're just going to make advancements even though they or GPU bore uh because that that goes to an over protectionism on our side to where we create a more GPU slop because we have way more high performance computing and we just don't care to make it efficient. >> Sure. >> Sure.
43:16
>> So I'd rather have that be to where all that stuff is just happening on our stuff.
43:20
So in some respects by giving them more we actually decrease the incentive for them to make innovations because they actually will have more GPUs.
43:26
>> So this is the this is the solution >> focus on efficiency.
43:28
This is the solution to negative gross margins.
43:30
This is what will save venture capital. It all comes back.
43:35
Uh anyway, I want to go to a post that you uh put up.
43:38
We can wrap up with the with the board.
43:40
If you want to design, if you want to take that board and design it into a beautiful graphic, do it on figma. com.
43:45
Think bigger, build faster.
43:47
Figma helps design and development teams build great products together and you can get started for free.
43:52
Um, you posted this is really a race between America and the world and you have the generative AI startups ranking OpenAI, Deepseek, Anthropic, XAI.
44:03
What was your takeaway from this chart of Gen AI startups? >> Uh, oh, that one.
44:09
Yeah, they I think it's a Yeah, China's obviously listen listed the most out of out of America.
44:13
M u but I think one is that it's our ability as a nation to uh invest in stuff that we think is uh revolutionary >> and become again the default nation.
44:23
We want to be the standard.
44:25
We want to be the exporter.
44:26
But in reality is that seeing when have we ever seen a great wave of new infrastructure development that has occurred this fast that's international.
44:36
So that goes back to uh whatever uh China shocking person that you were talking about. Yeah.
44:42
>> That's where he does it.
44:42
That's why I make a little bit of a just of a joke of has you talked to anyone who doesn't live in America. >> Yeah.
44:48
>> Because we haven't ever seen that in Silicon Valley to where >> there's a big motion in Silicon Valley to do this investment, you know, social media, mobile, web 3.
44:55
And what we see now is the fact of like all these countries are doing sovereign AI projects.
45:02
projects. That's where the rollup effect, the ability for us to control this is it's way more like Bitcoin where it's something accessible to everybody and and that's what leaning in on that side is kind of accepting the escape velocity that we're in versus the other side wants to live I think in a little bit of a fairy tale being like oh wouldn't it be nice if we control those
45:22
things if it's it's like if you have children you're like well wouldn't it be nice if you stop hitting your sister and the kids going to be like >> right and it's yeah it's like it's like you can only you can really only extend yourself so far when you are a world occupied by multiple free agents, multiple sovereign nations and and so I'd rather lean into our strengths where we can actually execute control. Yeah. Yeah.
45:42
The more that we are a platform, the more soft power we have, therefore I believe more peace and more prosperity, less starvation, good or greater things for humanity and we can still control things.
45:51
But if we're in this other side where we think that we can just hit people with sticks, like they're just going to be like, "All right, I don't want to play this game anymore." >> Would you go further?
45:57
Would you allow the export of the B200?
45:59
the the I I think some version of Blackwell should be in there.
46:04
I don't have much of opinion on how much nerved it should be or anything like that.
46:07
>> Um Gavin Baker had a good post that you reposted.
46:09
He said, "Any journalist describing Nvidia H20 as an advanced chip is not a serious person.
46:13
They're either ignorant or ideological.
46:15
The H20 is is CIR four years behind the actually advanced B200 that Nvidia sells to America and our allies.
46:24
The B200 has 30x more compute, 2x the memory bandwidth, 2x the memory of the H20 sold in China, arguably a 100x more usable compute.
46:34
Selling H20s to China is good policy as they are slightly 1.
46:37
5x better than domestic Chinese accelerators.
46:39
Selling the H20 thus effectively slows down the development of a robust domestic Chinese AI accelerator ecosystem, which would be good for China long term.
46:48
In the event of conflict, we would simply stop selling H20s to China and that would be uh and they would be left with inferior domestic alternatives and America would have a 5 to sixyear technology advantage over those domestic alternatives.
47:03
Conversely, if we do not sell the H20 to China during peace time, they would absolutely build effective domestic alternatives over time.
47:10
These alternatives would almost certainly be less powerful, less power efficient than Nvidia.
47:16
see the Huawei Cloud Matrix 384 with its super fast but but power- hungry optical scale up fabric.
47:22
But China has way more power available for AI than the United States due to their embrace of solar.
47:28
Uh selling H20s um selling H20s to China during peace time is also good for the world and America because it means their AI ecosystem is comparable to the American AI ecosystem and we can easily adopt the impressive algorithmic innovation seen in their open source models like Deep Sea.
47:43
uh and the world benefits from a robust Chinese open- source ecosystem that remains well behind American closed source frontier models.
47:51
The fact that the Chinese government is discouraging the use of H20s in favor of domestic alternatives should tell a rational dispassionate observer that all they need to know about which country benefits from the sale of H20s in China. >> Um soft power. >> Yeah.
48:06
My my my question is uh is is there something structural going on where where China has to remain open source because of their position in the market? >> Yeah.
48:15
Like I I think what what Gavin is reflecting which is more of my >> uh disposition is if we're going to make a strategic decision around because we we want still want to be the dominant superpower in the world.
48:27
>> We have to be realistic and pragmatic about where we are.
48:28
Like the fact is that we have I don't know how how high is it the debt 30 trillion right?
48:32
It's I mean it's it's negligible in the age of AGI but >> once we get a fast >> take off fixes this >> if we just grow GDP at 5,000% a year we're fine. >> Yeah. Yeah. Yeah.
48:45
So so the uh or just like fudge the numbers right the so the >> if we're going to make a decision about any sort of policy trajectory we have to understand like what we actually are like we generally have a quite hostile politics.
48:58
We have many states that are essentially bankrupt and we have a military that is still the strongest but is kind of lacking on the as asymmetric capabilities which is why we have androl being created.
49:08
So if we're going to say, hey, we're going to go to basically to a soft war with somebody, we should really understand what's what's going on.
49:15
And and I think there's many people who occupy the China hawk policy who, like as Gavin said, who don't even understand this technology or they co-mingle uh TSMC with Nvidia.
49:27
And so if they if they're if they're constructing an argument based on bad priors, then it's really hard to take anything else they say seriously because it and that's why I go back to the AI if you remove the AI dumerism, many of the people who are China hawks are just actual AI doomers.
49:45
They think that Terminator is coming.
49:47
It's going to kill us all. We're just robots. A very old >> disprove.
49:50
Do you think that's the flow?
49:51
I I thought it might be the opposite.
49:53
It's like you're a China hawk and you need to justify your policy and so you you wrap it in paper clipping. >> Yeah.
50:00
I mean it can go like either way.
50:02
Yeah, I can see going either way. Yeah.
50:04
That but but I just don't believe that.
50:06
Like and and if you have any for for example the number of of genes you have is like uh basically hundreds of thousands.
50:14
The number of cells you have is in is in the billions.
50:16
the number of neuron synapses you have is about 500 trillion.
50:24
>> The number of basically spaces in the model that like open AI uses is 500 times less. >> Yeah. >> Right.
50:30
And even in the fact even in the fact of like the way that we construct LLMs a day if you did that in human brain that's called a seizure. >> Right.
50:38
So so like we have to we have to realize that that again it doesn't mean >> don't make mistakes and don't have a seizure. >> Yeah. Yeah.
50:44
Like it doesn't mean AI can be amazing.
50:46
be amazing. it doesn't mean any of these things but but it's very different from from there's enough >> back on the Kerszsw timelines it's it's back on Ray Kerszw makes this argument that you know the the singularity will occur when the the total amount of flops in computing is is greater than the
51:05
total amount of flops equivalent in in the human brain and we are we're orders of magnitude away from that so you really have to you really >> within one human being it will encompass all the computing power in that lifime time of 80ish years a person to live than anything we have on planet earth. So, so like just sit down and be humble
51:19
So, so like just sit down and be humble like because when you're humble you can like basically appreciate because because even the fact of like >> what the the way ele work are very directive >> but that's not even the way attention works and perception works like do you recognize you're sitting in a chair
51:36
>> like do you feel that but your body does but your brain doesn't acknowledge it right so like even that differentiation between that there is so much computing happen in your body that you're not even aware of which makes us such an effective species and LM man, they're great. Like it, but you're overextending
51:49
Like it, but you're overextending yourself to thinking that because you have all of this power and all this energy uh being being able to generate tlops that you're going to say like, "Oh, you're a person now or you you have sentience." >> Yep.
52:03
>> And that that's just not the way the brain works.
52:04
And and that's not not even the way intelligence works. Like >> I like that.
52:08
Uh there's some I think it's a Wilmanitis post where he's saying like when they invented the steam engine, uh there were people that were saying like like the brain is just like a steam engine.
52:16
And then when they invented the uh the the the gasoline powered engine, they're like it's just like the the human brain is just like a gas powered engine.
52:23
And then when they invent the the CPU, they're like the brain is just like a CPU and GPU.
52:26
The brain is just like a GPU.
52:28
People want to like bring their their their current technology understanding into like the human >> and and cars 10 horsepower. >> Yeah, it's the same.
52:37
Yeah, >> there's 10 horses in here.
52:42
And and like one of the things I love about the the brain so much is like it's it's for for example it ability to be plastic, right?
52:47
Being able to mold itself, which LM can't do yet because they see novel information and they basically just right they kind of like freak out.
52:55
But but even even though we understand uh neuromodulators like acetyloline, dopamine, like norphine, even though we understand that uh whenever you see something um you know your your you have kids when you see your kids uh your body releases these neuromodulars that creates basically attachments to synapses and forms memories and the way that we even understand that cocktail we just know it just gets released.
53:18
We don't know the percentages.
53:19
We have no freaking clue.
53:21
And and as well as like if I take your brain and then release it the same amount, right?
53:24
just like take electrons there and just boom, right?
53:26
That you don't have the same effect, right?
53:28
That that how does directive and focus impact cells?
53:34
We have no idea like like you you can for example like hold this gong or hold that, right?
53:39
You're >> holding a diet coke. >> Yeah.
53:42
Like even these two things, right?
53:43
>> We don't even know what's happening when you have a fridge cigarette.
53:45
you have a fridge cigarette. there you're h how does like information of fact like this is diet coke and this like translate from cell information to the fact of me delinating caffeine and science has no clue like how that works
53:57
but we know it works >> and and so >> well I expect podcasters to figure it out >> right >> I mean the the do y'all watch Tim Dylan >> of course >> he has watch anything >> yeah he has a great thing on like podcasters >> blank slate all the Oh, >> all the alpha. >> I have a three-hour contact window.
54:17
>> I have a three-hour contact window. >> It's true.
54:21
>> Anyway, Aaron Gin, thank you so much for stopping by. Great to see you.
54:22
Thanks for coming in person. Fantastic.
54:24
Let me tell you about Vanta.
54:27
Automate compliance, manage risk, prove trust continuously.
54:30
Vanta's trust management platform.
54:32
Takes the manual work out of your security and compliance process and replaces it with continuous automation.
54:37
Whether you're pursuing your first framework or managing a complex program, >> let's take it over to Mark German himself.
54:45
He has some new reporting.
54:47
Apple plots expansion into AI, robots, home security, and smart displays. >> Apple Inc.
54:53
is plotting its artificial intelligence comeback.
54:55
Don't call it a comeback.
54:57
Uh, actually, you can call it a comeback.
54:58
Give you >> I mean, it makes sense.
55:00
They've been they've had they they've gotten a lot of hate over the last few years for the roll just particularly the roll out of Apple intelligence. Very hyped.
55:07
A lot of features on display at WWDC that didn't ship when the new iPhone launched.
55:13
They sold the new iPhone against it. Yeah. Exactly.
55:18
And so um I I think that there were uh there were high expectations because of Apple's position with ondevice inference being very >> I gave you Gen Moji. Are you not entertained?
55:28
they shipped a few things.
55:30
Uh I think that I I would love to see the Genoji actual usage data because I feel like even though it's not popular in Teapot and on X and in the tech the Technorati uh it might actually be seeing like decent adoption and joy.
55:44
I mean I've seen people share uh generative imagery from Microsoft Copilot which I had no idea was a popular image generation platform whatsoever. >> Facebook crowd. >> Exactly. Exactly.
55:56
So, so I I I I wouldn't be so sure that Gen Moji is is producing zero value.
56:01
Um, but at the same time, it's clearly been a tumultuous time for the company.
56:06
They've been kind of on defense in terms of public relations.
56:08
Uh, but Mark Gur is still scooping every single day.
56:11
Apple's plotting its AI comeback with an ambitious slate of new devices, including robots, a lifelike version of Siri, a smart speaker with a display, and home security cameras.
56:20
A tabletop robot that serves as a virtual companion targeted for 2027, is the centerpiece of the AI strategy, according to people with knowledge of the matter.
56:30
Uh, the smart speaker with a display, meanwhile, is slated to arrive next year, part of a push into entrylevel smart home products.
56:37
Tabletop robot that's a virtual companion. Yep, >> that's big.
56:42
>> Lot of lot of software they got to write to make that work.
56:44
They got to get on graphite.
56:45
dev code review for the age of AI.
56:47
Graphite helps teams on on GitHub ship higher quality software faster.
56:51
Tim Cook, if you're listening, get on. >> Get on graphite. So, this is big, too.
56:55
Home security is seen as another big growth opportunity.
56:57
New cameras will anchor an Apple security system that can automate household functions.
57:01
The approach should help make Apple's product ecosystem stickier with consumers, said the people who asked not to be identified because the initiatives haven't been announced.
57:09
I mean, this seems like something that they could knock out of the park.
57:11
Like just the smart thermostat.
57:13
Like these these patterns have been defined for a decade.
57:18
And that's like the sweet spot for Apple to come in and like the Sonos home speaker where we were saying like it they're just like the speak.
57:24
Everyone knows what they want out of a connected speaker system.
57:29
They just want it at the level of Apple execution.
57:31
It doesn't require going zero to one in some new territory.
57:36
And so I feel like they could be very successful in in the home video monitoring, home thermostat, home speaker.
57:43
So I I I think I like this overall.
57:45
The robot is where it gets funny because this feels crazy, but I'll I'll >> going into companionship feels interesting.
57:55
>> I don't think it's a companionship.
57:55
I I I I don't I don't think of it like that at all.
57:58
>> Tabletop robot that serves as a virtual companion is the centerpiece of the AI strategy.
58:03
I I mean, so let's read what the actual report is on this robot.
58:06
The tabletop robot resembles an iPad moni mounted on a movable limb that can swivel and reposition itself to follow users in a room like a human head.
58:15
It can turn toward a person who is speaking or summoning it and even seek to draw the attention of someone not facing it.
58:25
The hope is to bring AI in to life.
58:25
I don't think of this as like companion necessarily.
58:29
This could just be >> Jarvis >> assistant. Yeah, Jarvis.
58:31
I mean, I guess Jarvis isn't a companion, but companion has a bad flavor to it right now, a bad a bad vibe because people are saying like, "I married my companion."
58:41
And that feels very Black Mirror.
58:43
I do think that there's a serious Black Mirror risk, but I'll go into that in a second, but what were you about to say?
58:50
Yeah, I'm I'm not going out and saying this this feels like a competitor to friend.
58:53
com or any other players that are actually focused on companionship, but this does feel like giving Siri a presence in the home, like an always on presence in the home.
59:06
>> It will be interesting to see how they solve it.
59:07
Uh a partnership with OpenAI or Anthropic seems logical, but Apple Siri partnership with OpenAI Anthropic before September on Poly Markets is a 6% chance.
59:18
Um by December it's at a 44% chance though.
59:22
So but this is a very low volume market but still it feels like um we haven't seen a lot of messaging from Tim Cook that they're going to be building a huge data center going to be training a foundation model.
59:33
It feels like they're going to need to partner on some of the underlying infrastructure for that because I would not I would not want Siri hanging out in my kitchen talking to me.
59:43
Like Siri is just not reliable enough.
59:46
But I would I would welcome Claude or or Chachi PT into my house.
59:51
Uh but but I would treat it specifically as a like an Alexa. What's the weather like? What's on my calendar?
1:00:00
Uh look up the history of the Roman Empire.
1:00:02
I I would ask it questions like that and I would use it as a knowledge retrieval tool.
1:00:06
I would I might start to trust it in some agentic contexts.
1:00:09
And the question is, do you want do we want a robot iPad with an arm swinging around your kitchen?
1:00:17
And so and and hopefully when you want to know the weather, it's you're you're within earshot.
1:00:25
>> I think I think this could be really cool.
1:00:27
I think it could be really cool.
1:00:27
I mean, they say FaceTime.
1:00:30
>> This is funny because you're competing.
1:00:31
You're they're still competing with their own. >> Yeah.
1:00:33
But if you're ever in the kitchen, like this is an example.
1:00:35
So FaceTime calls will be a key function of the device during video conferencing.
1:00:38
the display will able will will be able to shift to lock on to people around a room.
1:00:44
And so as you're walking around, the FaceTime is just following you.
1:00:46
That actually feels more social.
1:00:48
I remember during co I I bought uh Facebook's uh camera.
1:00:54
They had some sort of device that you would mount on your TV and it would act as a uh as a video conferencing system. Yeah.
1:01:00
And I used it a few times.
1:01:03
I ultimately churned, but it was but it was it had a really wide angle lens and it would like zoom in on the people.
1:01:08
It had really good face detection and so it would center you no matter where you were in the room, but as soon as you walked to the other side of the room, it wouldn't be able to follow you because it it didn't have any motors in it.
1:01:17
Um, and I feel like >> Can we pull up the link in the chat?
1:01:24
>> This all reminds me of the Meta portal.
1:01:28
>> Yeah, that's the one I'm talking about.
1:01:29
The portal basically I didn't market it as a robot, but it would sit on a table and it would sort of follow you around the room.
1:01:35
And Meta is no longer selling MetaPortal.
1:01:40
Uh I guess they wound it down.
1:01:40
So >> anyways, uh Tim Cook told employees.
1:01:43
I gifted some of these and they were pretty wellreceived.
1:01:47
Uh most of them wound up being used as uh like uh digital digital photo frames though >> and they weren't actually used that often for calls.
1:01:57
I don't know if putting it on a robotic arm is enough to make it useful.
1:02:02
I think you do have to nail the underlying uh speech recognition and and the in the in the conversational nature like it has to be powered by a Frontier LLM.
1:02:13
>> You know what's funny?
1:02:14
>> What >> is I feel like the product I I I could imagine them having a mood board internally with the Pixar lamp hopping around >> that's literally listed here as like one of their inspirations apparently.
1:02:23
So Tim Cook told employees in all hands meeting this month that Apple must win in AI and hinted at the upcoming devices.
1:02:28
The quote the product pipeline which I can't talk about thankfully.
1:02:33
Mark German has his moles in Apple that are talking.
1:02:35
Tim Cook says it's amazing guys. It's amazing.
1:02:39
Some of it you'll see soon.
1:02:41
Some of it will come later but there's a lot to see. >> Yeah.
1:02:44
>> So Apple obviously always is cooking on moonshots.
1:02:47
very little of it surfaces to Mark German and even less actually surfaces, you know, publicly via Apple's own channels.
1:02:58
>> Mark German, if you have a little extra time, you should leak uh Julius's user numbers because Julius has over 2 million users now.
1:03:04
They're trusted by folks at Princeton, BCG, Zapier.
1:03:06
What analysis do you want to run?
1:03:09
It's the AI data analyst that works for you.
1:03:12
Connect your data, ask questions in plain English, and get insights in seconds. No coding required.
1:03:19
>> We need to get our logo up there.
1:03:20
>> Yeah, >> the fourth logo >> for sure. There's a lot >> more.
1:03:22
So, Apple is planning to put Siri at the center of the device operating system and give it a visual personality to make it feel lifelike.
1:03:27
The approach dubbed Bubbles is vaguely repres reminiscent of Clippy, which we are super bullish on, and we think Microsoft should totally bring back Clippy, an animated paper clip from the 90s that served as a virtual assistant in Microsoft Office.
1:03:41
Um, Apple has tested making Siri look like an animated version of the Finder logo, the iconic smiley face representing the Mac's file management system.
1:03:49
The final decision on its appearance hasn't been made with designers considering ideas that veer closer to Mimoji, the player, the playful characters that represent Apple user accounts.
1:03:59
Device prototypes use a roughly 7-in horizontal display approaching the size of an iPad Mini.
1:04:05
The motorized arm can extend the display away from the base roughly half a foot in any direction.
1:04:09
So that's not >> really hoping I was hoping I was really hoping across the room.
1:04:13
It reaches across and pins you against the against the fridge.
1:04:19
Yeah, I think Apple should be able to nail delivering this in a way that doesn't feel black mirror and dystopian. >> Oh, that's funny. Sorry.
1:04:26
So So I actually didn't see this, but there's a quote in here.
1:04:28
Some people familiar with the product call it the Pixar lamp. >> Yep. Exactly.
1:04:31
I I I I think it could work really really well.
1:04:33
Now >> they have to be cognizant of hop around though.
1:04:39
That would be very cool >> because that if it could hop off, >> it can definitely hop around.
1:04:42
That's Nat Freiedman core.
1:04:43
This is this is this is this is doable with modern technology.
1:04:48
Um but the risk is that they they botched it because Apple's had a rough go with the recent marketing and they've had a couple uh real like backlashes.
1:04:58
They had that one where they smashed all the pianos in the press. Do you remember this?
1:05:01
So, they had a uh a huge hydraulic press and they and they pressed like uh an easel with paints and a bunch of paint buckets and pianos and violins and all this stuff.
1:05:14
They get a bunch of blowback.
1:05:15
>> They got a ton of blowback from this.
1:05:15
We can we can try and pull it up. Uh do you remember this?
1:05:18
You remember >> I remember that it uh people were saying that they're like destroying these timeless pieces. >> Yeah.
1:05:24
And it's just like the thing that instruments and stuff like that. >> Yeah. thing that I love.
1:05:26
You're literally you're destroying it for this particular um uh shots fired.
1:05:29
Bill Bishop from the Substack stream.
1:05:33
You need better guests to talk about China and chips, not just Nvidia partners who are regurgitating Nvidia's talking points.
1:05:40
Bill Bishop, get on the stream. Call in. We'd love to have you. >> Bill. >> Bill, you're welcome. Hop on. >> Of course.
1:05:46
Aaron is riding with Jensen.
1:05:48
>> Yeah, he is the Jensen surrogate.
1:05:48
He's he's he's a surrogate for Jensen the juggler.
1:05:51
He he has to juggle both United States and Nvidia and China's interest and Taiwan's interest and all of them >> on the show.
1:05:58
Bill >> Miles Fisher wants to see the production team.
1:06:00
He says, "Show the off camerara voices."
1:06:02
Let's get the production team up on the screen.
1:06:04
Um >> Bill, call into the show right now.
1:06:08
We'll send you the Zoom link. >> Yep. Call in. You're welcome.
1:06:09
Uh can we pull up the Apple ad of them crushing the videos?
1:06:13
Is that even available anymore?
1:06:15
They might have just taken it off the internet.
1:06:16
Um but there was uh >> I think I found it.
1:06:19
>> But there was >> This is it, right? >> iPad.
1:06:21
You get games, you get an arcade machine, you get an easel, you get a piano, you get all of those things.
1:06:27
>> At the time, my said, "Our goal is always to celebrate the myriad of ways users express themselves and bring their ideas to life through iPad.
1:06:34
We missed the mark with this video and we're sorry. Absolutely brutal."
1:06:42
>> Well, really quickly, let me tell you about Profound.
1:06:43
Get your brand mentioned on Chat PT.
1:06:44
Reach millions of consumers who are using AI to discover your new pro uh new products and brands. Uh, check out Profound.
1:06:50
So, um, the other the other ad that didn't go as viral and and result in major backlash was the launch of the Apple Vision Pro.
1:06:57
So, when the Apple Vision Pro launched, one of the videos that they showed was of a middle-aged man looking at VR video of his kids, but he was alone in a dark room.
1:07:10
And it and Ben Thompson was like, "It feels like he's divorced or something."
1:07:14
Or maybe like the kids died.
1:07:16
It felt like Minority Report, which is a movie you I know you haven't seen.
1:07:18
Um, but in Minority Report, Tom Cruz's uh son goes missing from the pool one day and he has a he has a virtual 3D video that he plays back and gets very emotional about and and the the imagery was very similar in the Apple Vision Pro, but it wasn't an optimistic scenario.
1:07:38
It was like, yeah, like I guess if my son passed away or got kidnapped, like I would want to relive that, but that's not like going to inspire me to to buy an Apple Vision Pro.
1:07:48
And so with with this like Pixar lamp, this this uh this robot in your in your kitchen or in your house, like they really I think they have to lean into to light mode, not dark mode.
1:08:01
They have to really be aware and reality check themselves on the black mirror effect because uh the the idea of of someone hanging out in their kitchen being lonely and having this be the only companion that they ever talk to.
1:08:13
That's a lot less fun than hey, we're all cooking together in the kitchen and this is helping us keep track of the ingredients.
1:08:23
or uh my idea was uh teach it to uh teach it to pour some beers and get it at the fraternity.
1:08:32
This thing needs to be tending bar at a frat house and inspiring dancing silhouettes like the original iPod commercial in order to be successful.
1:08:40
Get the iroot a stack of red solo cups ASAP.
1:08:43
>> Apple's cancelled for fraternity coded marketing.
1:08:47
>> I think I think that would be amazing if this thing could if this thing could pour a beer without without foam.
1:08:50
I think they got a winner.
1:08:52
Um because it needs to inject itself into social.
1:08:55
Even the iPod, like yes, you could put it in the headphones and you could and you could tune out the world and be that dancing silhouette during the iPod commercial.
1:09:04
Um but also you could say, "Pass me the ox cord.
1:09:06
I want to play a song for my iPod that everyone can enjoy and it's a social experience."
1:09:11
And so it's important to show both of those sides now more than ever.
1:09:15
And in in chat GPT demos, in any AI product demo, there's always going to be the user who who uses the product to be more isolating, but the marketing should always be aspirational and pro-social.
1:09:29
Just like social media, social media, it's at its best when we're highlighting the fact that like I have so many friends that just send me a random Instagram reel every single day and like that's our interaction and it's and for me it's a way to like check in with a friend and and like >> share a laugh.
1:09:43
and share a laugh and and and it's a heart and it's a laugh.
1:09:47
>> I can I can imagine the ad creative for the Apple Pixar lamp.
1:09:50
You know, it's sitting on the center of a kitchen island.
1:09:54
The family's hanging out and they're facetiming somebody. >> Exactly. Yeah.
1:09:57
Uh one of the family members >> imagine us having a Pixar lamp here on the table. >> Really? Really? It's like Yeah.
1:10:03
It's like you want to be connecting two families or two two two sides of the family.
1:10:09
The East Coast side of the family, the West Coast side of the family.
1:10:11
They're both having Thanksgiving and they're able to connect, but there's everyone's being extremely social.
1:10:15
Do not film this thing uh with the lonely the lonely person.
1:10:20
Uh you know, having the Pixar lamp be the only thing to keep it company.
1:10:26
Anyway, that's my advice for uh >> what else projects coded Lynwood and Glenwood.
1:10:30
Core to the new home devices are current products like and current products like iPhones and iPads is an overhaul to the underpinnings of Siri.
1:10:39
Engineers are working on a version code named Lynwood with an entirely new brain built around LLMs, the foundation of Genai.
1:10:46
The goal is to tap into personal data to fulfill queries, an ability that was delayed due to hiccups with the current version.
1:10:52
>> It should be able to do it.
1:10:53
>> Hopefully, this team, this they should have a uh a a a moonshot division just trying to solve iMessage search.
1:11:00
Um they could probably start there.
1:11:04
>> Every tech company needs that. Amazon search is rough.
1:11:06
Google Gmail search is rough there.
1:11:09
Search has gotten so like dominated.
1:11:12
>> A billion dollar business in figuring out search.
1:11:16
>> I mean that would be the killer that might be a killer app for Comet and some of the browsers is just like actually being able to search my Gmail inbox better.
1:11:24
That would be pretty fantastic.
1:11:27
>> I feel like it should be at the app layer.
1:11:28
>> Anyway, whatever they're building, they're building products.
1:11:29
They got to get on linear.
1:11:30
Linear is a purpose-built tool for planning and building products.
1:11:35
Meet the system for modern software development.
1:11:36
and streamline issues, projects and product road maps. >> Open AI is on linear.
1:11:40
Let's get Apple on linear.
1:11:42
Let's get the Siri team on linear.
1:11:44
They they very well might be already.
1:11:46
So, uh somebody uh Craig Federigi uh who you all know says the work we've done on this end to end revamp of Syria has given us the result we needed.
1:11:56
He added that this has put us in a position to not just deliver what we announced but to deliver a much bigger upgrade than we envisioned.
1:12:02
He said there is no project people are taking more seriously.
1:12:06
So that is very positive.
1:12:09
Um a final decision hasn't been made on which models will be used but Apple has been testing anthropic PBC's claude for this purpose.
1:12:16
Mike Rockwell the former vision prochief who was put in charge of Siri earlier this year is overseeing both the Lynwood and Glenwood efforts. H uh apparently.
1:12:24
So during the development of the tabletop robot, Apple engineers have made heavy use of G chat GPT and Google Gemini to build and test features within Apple AI and Siri teams as a whole.
1:12:36
Software developers are increasingly using thirdparty systems as part of their development process.
1:12:42
Getting into a Ring competitor here, Apple's working on a camera code named J450s designed for home security, detecting people and automating tasks.
1:12:52
The device will be battery powered and could last for several months to a year on a single charge. >> Cool. >> That's interesting.
1:12:59
If they can do this for they they can they can do this for a random home security camera.
1:13:04
They can't do it for your So they have You're telling me they have the capability to have an always on camera.
1:13:12
>> Put the tin foil hat on.
1:13:13
>> We need the tin foil hat.
1:13:13
Uh the device has facial recognition and infrared sensors to determine who is in a room.
1:13:19
Apple believes users will replace will place cameras throughout their home to help with automation.
1:13:22
That can mean turning lights off when someone leaves a room or automatically playing music.
1:13:29
So, Apple, you're telling me that you can make a home security camera, of which I will p purchase multiple, and put them around my house, and it will be able to have a single battery that could last for several months to a year on a single charge.
1:13:46
And you can't make an iPhone.
1:13:49
>> It's a little bit The screen's pretty bright. The screen's bright. >> Yeah.
1:13:53
>> But an always on camera.
1:13:55
>> How long do you think your phone battery would last if you just constantly had the camera?
1:13:58
>> I I actually think it I actually think running the camera sensor is a lot less power intensive than running a screen for sure.
1:14:04
>> And obviously a bunch of different applications.
1:14:07
>> But still, >> yeah, >> pretty powerful, pretty exciting.
1:14:10
>> I don't think people want an always on iPhone.
1:14:11
They want the thrill of charging the iPhone every night.
1:14:13
If you're an adrenaline junkie, there's nothing like being on 1% trying to fire off that last iMessage, trying to >> I actually had a moment yesterday.
1:14:20
Uh my my one-year-old is uh starting to walk and I had this moment yesterday where I was I was I was I filmed a video of her walking. >> Yep. >> For about 10 seconds.
1:14:35
>> Uh everyone's in in the house is like laughing, enjoying the moment.
1:14:37
And then my battery died and completely lost the video because I didn't hit I didn't hit like end the video.
1:14:45
So the battery died right in the middle of this >> memory gone forever.
1:14:50
>> There is actually that has something to do with the encoder.
1:14:52
There is a certain uh file type that you can use when you're filming on one of these cameras that if the battery dies it will actually save the last of the file.
1:14:59
Um but there's certain there's certain codecs where if you don't hit finish recording the whole thing will be corrupted. It's very annoying.
1:15:08
Anyway, get on Numeral HQ. Sales tax on autopilot.
1:15:11
Spend less than five minutes per month on sales tax compliance.
1:15:13
Uh Tyler, what do you think about the the Apple robot?
1:15:17
Would you buy this thing?
1:15:20
>> Uh >> would you do a challenge for it and then churn in two weeks?
1:15:25
>> That's probably that's most likely I would. >> Yes.
1:15:29
>> Depends on how much it is.
1:15:29
I'm like, you know, >> what what have you adopted this summer that you still daily drive?
1:15:33
You've turned from the vision.
1:15:35
Uh, you've turned from the VR headset.
1:15:37
You've turned from >> Cle my new phone. >> Your your phone.
1:15:40
You have not gone back to the old phone. >> It's Yeah.
1:15:44
>> Extremely bullish for Apple. >> Let's hear it.
1:15:46
Let's hear it for new phones.
1:15:49
>> You know, you talk a lot of trash, Jordy.
1:15:50
Oh, the latest iPhone doesn't have an all day battery.
1:15:52
Tyler's still using the latest and greatest phone. Still a good business.
1:15:55
iPhone sales are ripping.
1:15:58
>> Well, Apple still got it. >> News.
1:15:59
Emily Choy has announced we just closed the biggest deal in crypto history.
1:16:03
Derabit is the number one crypto options exchange by volume and open interest and now it's officially part of Coinbase.
1:16:09
So of course Coinbase just completed their 2.
1:16:12
9 billion cash and stock acquisition of Darabit.
1:16:15
This had been announced previously. >> Yeah.
1:16:20
>> Uh Darabit launched in 2016 and reportedly notched over 1 trillion in annual trading volume in 2024.
1:16:26
So, let's give it up for >> let's give it up in in other in other uh crypto news. Bullish.
1:16:33
This is one of the greatest headlines I've ever seen.
1:16:41
>> The nominative determinism of of bullish.
1:16:42
I don't know if you can see.
1:16:44
Maybe I should just put it here.
1:16:44
We we we need like the printer cam back, but the the business and finance section of the Wall Street Journal says Bullish's IPO lives up to name.
1:16:52
Uh price of crypto exchanges shares jump 84% on their first day of trading.
1:16:59
Shares of bullish soared in the IPO, highlighting the challenge of pricing an IPO in today's exuberant market.
1:17:07
Just last month, shares of software company Figma jumped 250% in their debut.
1:17:11
That prompted whispers about the risks of a company underpricing an IPO and potentially leaving billions of dollars on the table.
1:17:18
The ideal first day gains are typically around 20 to 30%.
1:17:20
Many bankers who work on deals say uh bullish had more had a market capitalization of more than 13 billion as of Wednesday afternoon.
1:17:30
Its stock which trades under the symbol BLSH uh was temporarily paused after it began trading.
1:17:37
All eyes were on Bullish's large debut largely because the company is led by Tom Farley, the former president of the New York Stock Exchange. Wow.
1:17:45
uh Farley in the role uh guided some of the biggest IPOs including those of Alibaba and Snap.
1:17:51
Uh Bullish's debut was a chance for him to show his own ability to nail an IPO's pricing to both maximize for company proceeds because when you're going public, you are selling stock.
1:18:02
You want to sell it at the fair, it's a fundraising event.
1:18:05
You want to sell it at the fairest price.
1:18:06
Um if if the stock pops, you could have sold a lot more potentially and gotten with the same amount of dilution essentially.
1:18:12
So they could have so they raised 1.
1:18:14
1 billion um but because it jumped 84% they could have raised $2 billion for the same dilution.
1:18:21
And so that's in theory the in theory the problem here.
1:18:24
Um and so uh you know people people kind of go back and forth but it but it it gets to this point >> but again every I mean you're making a gamble right?
1:18:33
Even the Facebook the Facebook IPO didn't didn't it decline? >> Tanked. Yeah.
1:18:38
It was like one of the worst IPOs in tech history.
1:18:39
It was it was very very rough and it took a it took over a year I think for them to build back up.
1:18:44
Uh but that was due to a few things.
1:18:46
There was a technical error. There was mispricing.
1:18:48
There was a lot of uh like that I think the numbers or something.
1:18:52
There was like the mobile risk.
1:18:53
There were a few different factors where people were like oh maybe this isn't the best like a true like compounder. I don't know.
1:18:59
It was very odd in in in hindsight.
1:19:00
Um, but yes, this all goes to the point of how will individual investors react to public uh to public debuts because if they go out and it's a hot company, it becomes a meme stock or there's just, you know, too much latent demand, the stock will pop insanely.
1:19:15
And so, um, that's what a lot of companies that are going public are dealing with these days.
1:19:19
But it's sort of a champagne problem because it's good to have a high valued stock that you can go out and acquire stock public company now. >> Yeah, it's great. Um, BIN.
1:19:27
AI, AI, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bakeoffs, number one ranking on G2.
1:19:36
You can get started for free.
1:19:38
>> Brad Pitt says his Ford vers Ferrari movie with Tom Cruz never happened because Cruz found out he would not be driving much in the movie.
1:19:45
I >> had no idea this is what happened.
1:19:47
>> We both wanted to drive.
1:19:47
Tom realized that Shelby would not be driving much, so it didn't come through. >> Some lore.
1:19:54
>> Who wound up starring in Ford versus Ferrari? Who plays Shelby? I forget.
1:19:56
Um, it's uh, it's someone else, right? Ford verse Ferrari. Oh.
1:20:02
Oh, it's uh, Christian Bale. Christian Bale. That's right.
1:20:06
Christian Bale and Matt Damon.
1:20:08
Um, although Matt Damon does not play, >> who does he play? Matt Damon. Ken Miles, Donald Frey. Huh.
1:20:14
Anyway, Adio customer relationship magic.
1:20:18
Adio is an AI native CRM that builds, scales, and grows your company to the next level.
1:20:22
You can get started for free at Adio.
1:20:26
In other news, OpenAI and Sam Alman are investing a quarter of a billion dollars in Merge Labs at an $850 million valuation.
1:20:36
>> That's a big >> Financial uh Times is reporting.
1:20:37
So, uh Sam Alman wants to put a chip in your brain.
1:20:43
That's kind of the >> I think everyone wants to put a chip in your brain.
1:20:46
>> Everyone, even Roy Lee wants to put a chip in.
1:20:48
>> Roy Lee wants to and Mark Zuckerberg.
1:20:48
If you are, if you're not wearing Meta Ray-B bands, you'll be at a cognitive disability or something like that, didn't he?
1:20:56
Or cognitive disadvantage because having AI's revenge said, "Good, I'm used to it." This >> is great.
1:21:04
>> Uh, so the Financial Times, >> Alman's co-founding company. Wow. >> Yeah.
1:21:08
Open AAI and its co-founder Sam Alman are preparing to back a company that will compete with Elon Musk Neuralink by connecting human brains with computers, heightening the rivalry between this >> the two billionaire entrepreneurs.
1:21:18
The new venture called Merge Labs is raising funds at 850 million with much of the new capital expected to come from OpenAI's ventures team according to three people with direct knowledge of the plans.
1:21:30
Altman has encouraged the investment and will help launch the project alongside Alex Blania who runs World, an eyeball scanning digital ID project also backed by Sam Alman.
1:21:38
Uh Altman will co-found the company but not have a day-to-day role in the new project they added.
1:21:45
Merge is one of a slate of young companies looking to take advantage of recent advances in artificial intelligence to build more useful BCIs.
1:21:54
Its name comes from what many in Silicon Valley describe as the merge, a moment when humans and machines come together.
1:22:02
>> Yeah, it's a good it's a good >> the doomers are going to love this one. >> I'm sure.
1:22:06
>> Alman wrote a lengthy blog post on the topic in 2017 speculating the moment would come as soon as 2025. this year.
1:22:13
He suggested in another blog post that he could he could soon have high bandwidth brain computer interfaces as a result of recent technological advances.
1:22:23
>> Um I'm going to reference back to this blog post uh from Sam Alman called the merge.
1:22:27
He posted this eight years ago on December 7th, 2017. >> Read some of this.
1:22:32
A popular topic in Silicon Valley is talking about what year humans and machines will merge or if not what year humans will get surpassed by rapidly improving AI or genetically enhanced species.
1:22:42
Most guesses seem to be between 2025 and 2075.
1:22:47
People used to call this the singularity.
1:22:50
Now it feels uncomfortable and real enough that many seem to avoid naming it at all.
1:22:54
Perhaps another reason people stopped using the word singularity is that it implies a single moment in time and it now looks like the merge is going to be a gradual process and gradual processes are hard to notice.
1:23:05
I believe the merge has already started and we are a few years in.
1:23:07
Our phones control us and tell us what to do.
1:23:12
Social media feeds determine how we feel.
1:23:14
Search engines decide what we think.
1:23:16
The algorithms that make all this happen are no longer understood by any one person.
1:23:21
They optimize for what their creators tell them to optimize for, but in ways that no human could figure out.
1:23:27
They are what today seems like sophisticated AI and tomorrow will seem like child's play.
1:23:31
And they're extremely effective.
1:23:34
At least speaking for myself, I have a very hard time resisting what the algorithms want me to do.
1:23:38
Until I made a real effort to combat it, I found myself getting extremely addicted to the internet.
1:23:44
>> We are already in the phase of co-evolution.
1:23:46
the AI's effect and infect us and then we improve the AI.
1:23:49
We build more computing power to run the AI on it and it figures out how to build even better chips.
1:23:56
This probably cannot be stopped.
1:23:58
As we have learned, scientific advancement you eventually happens if the laws of phys physics do not present prevent it.
1:24:05
More important than that, unless we destroy ourselves first, superhuman AI is going to happen.
1:24:11
Genetic enhancement is going to happen and brain machine interfaces are going to happen.
1:24:15
It is a failure of human imagination and human arrogance to assume that we will never build things smarter than ourselves.
1:24:20
Our self-worth is so based on our intelligent that we believe it must be singular and not slightly higher than all the other animals on a continuum.
1:24:29
Perhaps the AI will feel the same way.
1:24:31
And note that differences between us and bonabos are barely worth discussing.
1:24:37
>> The merge can take a lot of forms.
1:24:37
We could plug electrodes into our brains or we could all just become really close friends with a chatbot. Wow. Eight years ago. He said that. >> Eight years ago. >> That's crazy.
1:24:49
>> Chad GBT of course wouldn't exist for years.
1:24:51
But I think emerge is probably our best case scenario.
1:24:53
If two different species want >> both want the same thing and only one can have it in this case to be the dominant species on the planet and beyond.
1:25:01
They are going to have a conflict.
1:25:03
We should all want one team where all members care about the well-being of everyone else.
1:25:08
>> Although the merge has already begun, it's going to get a lot weirder.
1:25:10
We will be the first species ever to design our own descendants.
1:25:13
My guess is that we can either be a biological bootloadader for digital intelligence and then fade into an evolutionary tree branch or we can figure out what a successful merge looks like.
1:25:25
It's probably going to happen sooner than most people think.
1:25:26
Hardware is improving at an exponential rate.
1:25:28
The most surprising thing I've learned working on OpenAI is just how correlated increasing computing power and AI breakthroughs are.
1:25:37
And the number of smart people working on AI is increasing exponentially as well.
1:25:41
Double exponential functions get away from you fast.
1:25:45
It would be good for the entire world to start taking this a lot more seriously now.
1:25:49
Worldwide coordination doesn't happen quickly and we need it for this. >> It's a wild post.
1:25:55
Is that still up or did you have to go to the archive? >> Still up. >> Still up. Interesting.
1:26:02
That's a That's a hot take.
1:26:02
Tyler, would you put a chip in your brain?
1:26:07
Yeah, I I want I want 40 piped straight in. >> Yeah, you're down. >> I'd do it. Yeah. Sick of fency maxing. Yeah.
1:26:13
>> Chat is not super into it.
1:26:13
Gabe says he just wants salt and vinegar chips, not brain chips.
1:26:18
John Xley says he's got to be honest.
1:26:20
I don't think it's for me ever.
1:26:22
I can already picture the contagion of peer pressure.
1:26:25
>> Yeah, reading that blog post made me want to touch some grass.
1:26:26
We should actually get some grass here in the studio.
1:26:30
>> We should get some grass in the studio.
1:26:34
on the table so we could just have one hand on the grass.
1:26:36
>> Uh, it relates a lot to this cold healing post.
1:26:38
2040 population distribution predictions.
1:26:40
90% brainless content zombies who are functionally illiterate cannot write a paragraph.
1:26:45
5% neoish flip phone users.
1:26:49
5% elites and super elites at algorithmic war with each other to steer the 90% to their whims.
1:26:57
And then Crab says, "Always has been."
1:27:00
Um, I I do uh I I I think this I think the case of the Neo Amish is fascinating because uh the the non Amish the Amish are doing fantastically.
1:27:10
Uh they they have adopted flip phones I believe and they've adopted some technology but in general they've accumulated a lot of real estate and they're by all accounts rapidly >> by all accounts they are flourishing.
1:27:25
And so what's interesting is that the the the the coastal elites have not gone to war with the Amish.
1:27:30
They've not said like what they are doing is unacceptable and we must like eradicate them yet.
1:27:38
>> I don't know if they'll ever be threatened by it.
1:27:39
I think they might be fine with it forever.
1:27:40
I think everyone might be fine with it forever.
1:27:41
And so I think >> they can build they can build barns like they're they're in a game of threatened threatened by that.
1:27:49
>> I mean if they can build a barn, what else could they build?
1:27:50
I think I I think that like you even though there will be >> you've seen the videos, right, of them like assembling a barn in like 12 hours, >> it truly is some of the greatest >> some of the greatest touching we can do all time.
1:28:02
The um the Amish population doubles every 20 years. >> Amazing. >> Yeah. >> Yeah. America. >> Chart that out. >> Yeah.
1:28:11
Oh, that's a fast takeoff there. Yeah. Yeah. Yeah. Yeah. Yeah.
1:28:13
Project the next 10,000 years.
1:28:15
What will the population work on that? >> General intelligence.
1:28:19
How many Amish people will there be in 10,000 years if they keep doubling every 20 years?
1:28:26
That's what I want to know. I bet it's a big number. Um Mark has a good one.
1:28:27
He says, "AI equals Amish intelligence.
1:28:30
How to build a barn in 20 hours." I love it. Thank you, Mark. Um uh this is so funny.
1:28:40
Um, I I I but but but I think my takeaway is like the the So Sam is painting this picture of like you get left in the evolution tree and you get like dropped off and it's a dead end.
1:28:50
I don't know that it's necessarily a dead end.
1:28:53
Like you could like monkeys are still around.
1:28:55
You could you could still just be a monkey and you could be an Amish and like the super intelligence brain computer merged people could go off and do their own thing and you could just be vibing as an Amish for thousands and thousands of years.
1:29:06
like there's not necessarily an incentive to apply until the super intelligence comes and says, "Hey, I'm going to need you guys to come and ride soul cycle bikes to power my data center."
1:29:18
>> I I think that I think that people are just are just people like people like animals.
1:29:22
People are nice by default.
1:29:24
Like the default state of of most living things is not this like vicious battle to dominate everything constantly.
1:29:30
Like yes, like the animal kingdom is extremely metal and there are insane battles constantly, but it doesn't it doesn't need to necessarily be that way.
1:29:42
Yeah, I don't think it necessarily needs to be that way.
1:29:43
I think the I think the Amish can coexist.
1:29:45
Tyler, do you have the final number?
1:29:48
>> How many Amish people will there be in the year 10,000 10 uh 12,025? 12,025. >> Uh yeah. So I think it'd be uh 1. 3 * 10^ the 156.
1:29:59
Can you turn that into like words?
1:30:03
>> Is it quintilian, octilian, septilian?
1:30:03
I want I want to know convert it into convert >> 10 to the 156 is like >> is that past a Google? >> I guess 10 to 100.
1:30:13
So there will be more than a Google Amish.
1:30:17
>> Yeah, >> that's fantastic.
1:30:17
I'm I'm rooting for the Amish. >> For the Amish.
1:30:21
>> I think you got to go long Amish.
1:30:24
>> How do we make money on this?
1:30:26
>> Someone asked, "What about the absent-minded destruction of parking lots?" I don't get that. I'm confused.
1:30:31
>> I don't know about that market reference.
1:30:33
>> Are are the Amish destroying parking lots?
1:30:35
I think they're building parking lots for all those barns they're putting up.
1:30:38
Uh anyway, the Amish are missing out on eight sleep.
1:30:40
Or maybe that's a loophole that they can do.
1:30:42
We should definitely get the uh the Amish on eight.
1:30:45
Get a pod five if you're Amish and you're listening to the stream and you're in the chat right now.
1:30:49
Uh they got a 5-year warranty, 30 risk-f free trial, free returns, free shipping.
1:30:52
So >> that would be imagine the the >> first thing you put in the barn. Pod five.
1:30:58
They could imagine if we find out that the Amish can build barns so quickly because they've all been secretly sleeping on eight.
1:31:05
>> Maybe that's the secret. Maybe that's the secret.
1:31:06
Uh >> they got an Amish version from Matteo that's that's like cooled by the earth.
1:31:13
>> Ooh, the the the the farmtotable eight. That's nice.
1:31:17
Uh Taylor says, "I for one welcome our Amish Borg overlords."
1:31:20
And Gabe says, "Long Home Depot."
1:31:24
>> Yeah, Long Home Depot.
1:31:24
going along Home Depot to do Do the Amish shop at Home Depot or do they chop their own trees down? I don't know.
1:31:29
We need to do a deep dive on the Amish.
1:31:31
Anyway, should we debate whether or not TV is dead or magazines dead?
1:31:36
Uh, one of our strongest stands, Adam Faze, I believe he's a part of FaZe Clan.
1:31:41
Um, Adam Fa said, "Emily Sunberg asked me for my anonymous hottake," which I guess he immediately de dean anonymized.
1:31:49
Uh, on the state of media right now, I'll out myself because I think I don't think mine's that anonymous. Magazines are dead. TV is dead. Hollywood is dead.
1:31:57
TBPN is more important than CNBC.
1:31:59
Emily Sunberg is more important than Vanity Fair.
1:32:01
Nick Fuentes is more important than Fox News.
1:32:05
And X is unfortunately the center of the universe. Um, interesting take. Very spicy.
1:32:10
Alex Heathic from the Verge says half is true, half is not.
1:32:13
We'll never know which half.
1:32:16
And Max Tani says, "I think you could make the argument that none of this is true." Yeah.
1:32:22
Who knows what is relevancy? What is importance? Who knows?
1:32:24
All I know is we're having fun on this stream.
1:32:26
We're having fun with you in the chat.
1:32:29
>> I just want to say you're keeping magazines in business, John. >> I am. I mean, yeah.
1:32:32
Oh, the newspaper's dead.
1:32:34
Why do I read it every single day then?
1:32:37
>> Well, he didn't say the newspaper was dead.
1:32:39
>> No, the newspaper is Lindy.
1:32:39
I think uh I think uh >> I actually think true true journalism is more important than ever.
1:32:47
>> As long as TBPN is alive, Hollywood cannot be dead because we are live from Hollywood and we are joined by someone.
1:32:54
We're not far from Hollywood.
1:32:54
Delhi and Aspir partner at Founders Fund, >> co-founder of Varta Space.
1:33:00
>> What can we do to second office for Varta?
1:33:02
Let's move it to Hollywood.
1:33:04
Hollywood is the new Silicon Valley.
1:33:06
>> Well, if they get task a lot of warehouse space.
1:33:09
>> Wait, think about this.
1:33:09
So, if you guys task get tasked within Moonlanding and we, you know, we need to >> Oh, yeah. Yeah. Yeah. Yeah. We got you covered.
1:33:18
If if you're going out for for series J and you need to, you know, make a moon landing appear like it happened, we got the cameras for you. We're good.
1:33:28
>> You know, somewhat controversial take is the moon landing did happen. >> Yeah.
1:33:31
>> The footage wasn't looking that good. Oh.
1:33:33
though they did live stream from Hollywood for the footage that they had some great sort of takes.
1:33:40
>> So both they had to reshoot it. They had to reshoot it.
1:33:43
>> That's Yeah, re-shoots happen all the time.
1:33:45
I I don't have a problem with it.
1:33:46
I I I say let it let it happen. >> Fair fair play. >> Yeah.
1:33:49
Anyway, what is new in your world?
1:33:52
Uh there's a bunch of stuff going on DC.
1:33:54
We can talk about chips in China.
1:33:54
We can talk about um I want to talk about negative gross margins in AI companies. What's going on venture?
1:34:00
There's a bunch of stuff.
1:34:01
But what's what's top of my mind?
1:34:03
>> Why don't why don't we start there because this is a fun one because because Delian has been been getting, you know, chirped at for years for years about this about margin profiles. >> Oh yeah. Yeah, that's true.
1:34:16
>> We're back to Everett, baby.
1:34:16
Let's make it just somebody tell Everett to somebody tell Everett to come on right now.
1:34:22
Let's have the gross margin.
1:34:25
So So here was my take to kick it off.
1:34:27
Uh here's an example of how the AI startup economy works.
1:34:29
I've anonymized the companies, but uh this was relayed to me in a group chat and I believe it to be roughly true.
1:34:35
Uh you, the user, gives an application layer company $1.
1:34:39
They turn around and give $5 to a foundation model company that gives $7 to a hyperscaler who in turn gives $13 to a GPU maker.
1:34:46
Uh and it's giving House of Cards.
1:34:49
And so the question is how fast can inference costs fall?
1:34:51
Are we is this are we so over? Will we solve this?
1:34:55
Is is Moore's law enough to save us?
1:34:58
What else needs to change?
1:34:58
What's your take on negative gross margins in now software businesses and the end of zero fixed costs?
1:35:06
>> I saw this um you know graph that was um Nvidia's uh free cash flow by quarter and they're like capex by quarter.
1:35:11
The first was like that they spent I think I forget if it was quarter yearly maybe it was yearly but they're you know spending something on the order of like you know 4.
1:35:19
4.5 you know sort of billion in you know sort of capex um on you know scale up of you know production lines and their free cash flow is something on the order of like 354 you know sort of billion dollars and so this is something like
1:35:30
funnally ironic where like you take the whole like 2010's mantra of like you know software is eating the world marginal cost of distribution is zero your margins are phenomenal and then it's just like it's all just completely wrong, right? Like, you know, SAS
1:35:42
Like, you know, SAS companies end up having like crazy high sales and marketing spend in order to actually like grow their revenues at all.
1:35:48
So, the margins don't look anywhere near what they look like.
1:35:50
Alex Clayton from uh, you know, Meritch had this tweet where um, now the uh, just publicly reported AI labs revenue company, the sorry, the publicly uh, reported revenues of AI labs now surpass the entire public SAS uh, you know, publicly traded SAS uh, you know, industry.
1:36:10
So it's just like you have this entire like decade of companies things that people you know sort of worked on revenues etc.
1:36:15
And then it turns out like it was all just completely wrong both on like those companies actually having durability margins etc.
1:36:20
And there's replaced by something that is somehow bigger and potentially also has all those you know same like risks and concerns of durability margin etc. And who's the winner?
1:36:31
It still ends up being the people that have the fundamental like hardware you know infrastructure.
1:36:35
Like if you look at like where all their like free cash flow is ending up, it's somehow ending up in like Oracle running data centers, Microsoft running data centers and you know Nvidia basically building the chips that you know goes
1:36:45
into those and like the energy companies that are providing the energy for it but like the end like application layer and like software companies end up capturing like none of the you know sort of margins or cash flows from all of this. And so I don't know as the guy that's
1:36:54
And so I don't know as the guy that's always loved bits and atoms rather than pure bits alone.
1:36:57
I continue to feel very validated and I admit that I enjoy studying, you know, some of these AI companies gross margin profiles and then realizing that like the space factory company has better gross margins than they do. >> That's wild.
1:37:10
I I do think that there's any hope for inference costs to fall off a cliff and and and the gross margins to like kind of get uh right side up very quickly because that is the the bull case.
1:37:22
We saw this with Deepseek like the inference cost even for reasoning models dropped like you know 20x very quickly and you just don't see that in you know I'm a you know I'm a home I'm a homemaker and all of a sudden like like the price of lumber just dropped 10x like that just doesn't happen.
1:37:38
It only happens in in software and algorithms where a more elegant implementation of the algorithm can actually drive same results 10x uh savings and cost.
1:37:46
Uh, and that feels like everything's like everyone's kind of hoping from for a bailout maybe, but uh, and I'm like cautiously optimistic that it could happen.
1:37:56
>> We're gonna we're gonna we're gonna get a a government bailout of the artificial intelligence industry within the next 365 days.
1:38:02
We need a we need we need a federal backs stop. >> Yeah. Universal tokens.
1:38:08
Everyone gets a everyone gets $1,000 in in Nvidia credits or something.
1:38:15
The other like interesting you know sort of macro lens on this is if you look at um the um amount of capex that is going into setting up you know basically um you know buying chips setting up use of data centers basically all things that
1:38:29
the like you know sort of hyperscalers and foundation labs are contributing towards it's actually the highest single you know sort of um uh industry investment line item as a percentage of GDP since basically you know the fiber roll out in you know sort of you know, late 90s. And then before that, the last
1:38:44
And then before that, the last time that we had an equivalent was basically the railroad broom in the um uh late 1890s, right?
1:38:51
I want to say >> and the railroad boom was significantly higher.
1:38:55
I think it was like almost 9% of GDP.
1:38:57
I know the chart you're thinking of.
1:38:58
And I think fiber uh and and like the internet roll out was like 1. 2 and we're at like 1.
1:39:02
3% of GDP, something like that.
1:39:05
So we're not we're not we're not like insane 10% of GDP which is kind of what like the the fast takeoff folks the you know the the the the Dwaresh Patel crowd talking with Sata Nadella talking about like this could be 50% of GDP if if scaling really really works like we think it will.
1:39:24
>> The the thing that's unique though to those you know relative to those prior booms is fiber companies were dependent on their fiber revenue in order to keep justifying rolling out fiber right and they obviously ended up being wrong.
1:39:33
you thought that fiber was going to be everywhere.
1:39:35
There was going to be infinite internet and then it turns out you set up like a basic backbone for the internet and then after that you actually didn't need fiber everywhere.
1:39:41
And so there was like projected to be semi- infinite demand for fiber and now the biggest you know sort of company that was rolling that out in like the late 90s Corning literally only like finally last year reached their original like peak you know 200 market cap.
1:39:50
It took you know what four years and most of that now has nothing to do with fiber.
1:39:56
It's like the Gorilla Glass that they make for you know iPhone all these other like special glass applications.
1:40:01
uh but you know shows how long it can take to you know get back to your prior peak.
1:40:03
Um the railroad companies obviously were highly dependent on their like you know rail revenue in order to be able to continue to invest into railroads.
1:40:10
What's crazy about the like you know capex that's going into these data centers is like a significant chunk of it is funded from companies that have free cash flow and revenues from completely unrelated to the AI business right like Meta is funding all this stuff from like their social media and targeted ads.
1:40:22
You know Microsoft is you know feeding all this off of their like you know sort of SAS you know revenues you know Amazon obviously from their like you know marketplace etc.
1:40:29
like the only companies that are actually dependent on the AI revenues to fund this capex are like basically the foundation you know sort of model companies and then Google from their like monopoly you know basically on you know through search and so when you think about like how long they could potentially sustain that like you know 1.
1:40:44
3 or you get up to like 2% of GDP it could potentially take a significantly longer time if there is any type of you know sort of bubble here for there to be eventual you know sort of decrease on the percent of GDP because they're not actually that reliant on you know the you know you know that that capex rollout actually immediately contributing to, you know, revenues.
1:41:00
Um, and then the crazy part is like you still have Apple, you know, sort of mostly sitting on a huge chunk of cha cash and hasn't really participated in all this other than the fact that actually they're like, you know, revenues have now been flat for like five and a half years straight.
1:41:10
And then they're like, you know, balance sheet has basically only decreased linearly, you know, quarter over quarter for the last 5 years straight.
1:41:15
So it's like man if you like a more brilliant version of Tim Cook 5 years ago should have just bought like $90 billion of GPUs and like started building you know some data centers and instead he like I don't know rolled out the like iPhone like you know 400 same features as the last one.
1:41:32
>> iPhone 400 would actually go pretty >> which to be fair has good margins and low churn.
1:41:35
Uh we gave our intern a new iPhone and he has he's still using it and we can't say that about some of the sass that we've had him try.
1:41:42
>> Happy there there is a there is a wrinkle with the Apple thing.
1:41:44
A lot of that uh the balance sheet decline is because of dividends.
1:41:47
They're paying C shareholders out. But I agree with you.
1:41:50
Um the other thing that >> they don't know what to do with the cash. You know what I mean? Like Zuckerberg.
1:41:55
>> They're creating they're creating the Pixar lamp. >> Yeah.
1:41:58
They're going to create Pixar lamp.
1:41:59
Literally the Pixar lamp that will sit on your your >> It's going to be an iPad on a robotic arm and it will be able to like move around and follow you around the room.
1:42:07
So if you're facetiming someone in the kitchen and you walk over to the stove or the refrigerator, it'll follow you around.
1:42:12
>> And they're going to compete with Ring.
1:42:14
They don't know what to do with their cash.
1:42:14
I think they got plenty of world changing ideas. >> Yeah. Yeah.
1:42:17
They're going to go steamroll a $2 billion company or something.
1:42:20
>> It sounds like a shitty like 2013 like YC like startup, you know, sort of pitch like on the order of like Juicero basically. >> Yeah. Yeah. Who knows?
1:42:26
Um I want to talk about Yeah.
1:42:28
the the the the fractures in the in the financial world around AI do feel like the foundation is a little more solid because it's coming off of cash flow from these hyperscaler balance sheets.
1:42:40
Um there's also the uh >> Yeah.
1:42:42
That's and that's the the bullc case for the the hyperscalers in AI and just being competitive even if they're behind the ball on codegen and personal assistance and things like that is that they're just going to be able to sustain this pace and and investment level. >> Yeah.
1:43:00
>> For the other the other kind of Yeah.
1:43:02
The other kind of source of capital that people are discussing now is private credit and people are wondering if there's this private credit bubble but then also private credit isn't marked to market as much and can you give me kind of like the the venture capitalist 101 on how venture capitalists are interfacing with private credit these days because it feels like there was
1:43:23
usually like a handoff where you invest in a company and then at a certain point it goes public and then you know or goes into private equity and then you're kind of out of the game, but it feels like there's a little bit more interface between um the VC world and the private credit world now like do you have any insight into what's going on there? Have
1:43:39
Have you followed that at all?
1:43:43
>> A bit you know um uh the >> very deal right >> study the stuff is like the upstream which is like the you know sort of equity side of things.
1:43:51
equity side of things. Um I'm forgetting the you know sort of exact stat uh that I'm thinking of but there was basically this crossover point in the last year which I want to say was it it wasn't IPO equity raised but it was um it was something like you know sort of capital
1:44:07
deployed into like sub$200 billion market cap companies and basically for the first time ever the private capital market was actually as large you know and had crossed over to the public you know basically capital markets in terms of you know sort of fundraising in terms of like issuing primary in fundraising. >> Um and so the strategies um uh that some
1:44:23
>> Um and so the strategies um uh that some of these like uh major asset managers on the equity user side of things, these are obviously like the Fidelities, the Black Rocks, etc.
1:44:35
the ones that are like you know the you know Vanguard.
1:44:37
Um all of those groups at this point have clearly um you know sort of admitted to themselves that the sort of prototypical you know IPO when you're between5 and $25 billion market cap is just a world that is not coming back anytime soon.
1:44:51
It's not to say there won't be some companies that do that.
1:44:52
Obviously Figma has done you know sort of phenomenally well and so that may tempt some companies um to go back to it but all these asset managers have realized that you know so much of the returns are now being you know sort of held up in the private markets that they need to you go enter those and so you're seeing the equivalent on the you know sort of private credit side of things.
1:45:07
Um I don't think I can you sort of share the logo of the company but there was a company in our portfolio that uh recently um uh you know raised hundreds of millions of dollars in private you know sort of credit from tier one you know basically you know sort of bank you know you know top you know sort of foreign in the United States.
1:45:25
>> Um the cost of credit that they had on it was effectively like Liebore plus 25 bips which is just like the equivalent of like cost of capital for a like $50 billion publicly you know traded company.
1:45:36
very different than like venture debt, which is like something people like, "Oh yeah, I raised $10 million.
1:45:40
I got a $1 million credit line or something like that."
1:45:43
This is a completely different world.
1:45:45
>> If you have a bad quarter, we're we're going to >> we run the company. >> Yeah, exactly.
1:45:49
These are totally different.
1:45:50
This is like, you know, you know, you know, um uh uh you know, backing physical assets that these companies are going and buying and you know, sort of rolling out.
1:45:58
>> And so I remember watching that deal happen and being like, "Wow."
1:46:00
Like I mean first as the like equity investor I'm like oh my god this cost of capital super easy for this company to like you know cover the like you know capex you know sort of roll out based off their you know sort of margins etc.
1:46:11
you know sort of margins etc. So they're like return on equity now that you can see as the private investor is like insane where it's like why go public where this company doesn't even need to raise that much equity and they like one of the like there's a lot of value proposition for being public but one of them is that you're like general cost of capital on the debt side just goes down
1:46:27
significantly typically when you're you know sort of a you know a public company and that was what people thought of on the cost of equity side for a long time that was the whole purpose of going public was like you know in 2013 you couldn't raise $5 billion in the private markets and so if you wanted to scale to that level you had to basically go public private credit was much more laggered to catch up. But it's now
1:46:44
But it's now actually starting to catch up where it's like well man you're like cost of equity is basically the equivalent if not cheaper than what it is when you're public.
1:46:50
Now the cost of credit is starting to basically you know sort of become the equivalent at some point.
1:46:54
It's like you know why are you even you know sort of bothering to you know sort of you go public.
1:46:57
So that was like you know sort of first set of thoughts that I had.
1:47:01
But then the second was okay I was like trying to think about it from like the lender perspective.
1:47:03
I was like, in what world, you know, are they giving this like, you know, private company the like equivalent cost of capital of like, you know, publicly traded >> Apple.
1:47:11
It couldn't it couldn't be that they've just raised so much money and they need to deploy it at all costs, right? Couldn't be that. >> Yeah. Yeah.
1:47:19
I mean, maybe there's a part of >> that couldn't be it, right?
1:47:21
Like why, you know, >> we raised a lot of money, we have a lot of aum, we got to take some huge risk right here. We got to deploy.
1:47:28
right here. We got to deploy. we can't we can't generate fees unless we're you know slugging around >> maybe >> huge check >> but what yeah what's the other reason I >> I think the other reason is like um there's these certain like um super cycle macro trends that sort of feel unstoppable and if your company is related to that >> the banks are willing to underwrite it
1:47:49
because they believe that like look this piece of infrastructure that you're building if you don't utilize it there will be a fast follower that quickly you know sort of jams it up so you build out XYZ data center with all these GPUs for some reason if your model your like particular consumer application doesn't work somebody else is basically going to
1:48:03
go snap that up and so they clearly are treating like a part of underwriting you know you know capex as a you know infrastructure like debt provider is thinking about like if the like equity holder of this you know basically goes belly up is this does this asset actually have secondary liquidity right and so that's why you know um you know companies that have super esoteric pieces of equipment or technology a
1:48:21
little harder to have debt if you're buying something say like you know you know at >> that warehouse I I mean the warehouse space like If you were to I don't you're probably leasing it, but like if you were to say I want a mortgage and I want to buy a warehouse, like that's hyperlquid and you can sell that to somebody else who needs a warehouse very easily. You're in Elsa Gundo. It's You're in Elsa Gundo. It's pretty simple.
1:48:38
Um but yeah, like if it's R&D and if the R&D doesn't produce fruits, like that needs to be equity backed, that needs to be paid for with equity.
1:48:46
And so um yeah, it is interesting that yeah, there's this like super cycle and there's certain technologies, certain pieces of equipment, certain pieces of hardware that you can just underwrite way way differently because somebody's going to pick them up.
1:48:56
There's three >> water cooling towers at GPU centers, transmission lines, like people building transformers. Yeah.
1:49:02
Um the gas turbines that people are using for like, you know, basically backup, you know, sort of power to like smooth out the grid battery systems.
1:49:08
And so it's interesting where it's like you almost have the like, you know, and I always talk about this like you want to, you know, transform the world.
1:49:12
You can, you know, do all the government policy you want, but ultimately it's like the invisible hand that is going to, you know, make this happen.
1:49:17
When we think about like electrification and buildout of like you know sort of grid when we think about like reshoring of you know sort of manufacturing in the United States and having like you know not just 10 CNC machines in a facility like this but like hundreds thousands etc. >> Yeah.
1:49:30
>> The easiest signal to see whether or not that stuff is going to happen is like what is the cost of capital to go do that and the banks are clearly signaling like we believe that this is such a macro trend.
1:49:37
macro trend. there's such strong, you know, so support from the market, such strong support from government policy that like we're going to be giving these like subscale companies, but that are focused in these particular areas, cost of capital low enough that this, you know, and we believe that the risk is low enough because of the secondary liquidity because of all these macro
1:49:51
trends that it sort of becomes this self-fulfilling loop where now all of a sudden you do get, you know, all those, you know, sort of trends, you know, built out on, you know, you know, crazy amounts of energy built out, you know, restoring and production in the United States, you know, sort of huge data centers and so the now the stuff can also go in the reverser direction, but I think it's just like one of these
1:50:06
highlights of like Capital markets are just so reflexive where like when people believe in a trend more then the cost of capital becomes less then the rollout you know basically happens faster and then it loops on itself for the cost of capital becomes less etc and you're seeing this in these like you know sort of categories but there's like more sustainability again to this especially in the data center world where you know
1:50:22
it's not dependent on the data centers producing cash flows for you know a lot of these businesses obviously on all things restoring production they're obviously more more you know sort of dependent on actual revenues there um the only other topic that I was going to you should sort of flag by the way I don't know if you guys saw but uh the white house actually uh put out an executive order yesterday. Um that was
1:50:36
executive order yesterday. Um that was one of their you know sort of first you know broader orders on basically maintaining commercial space leadership in the United States that I kind of wanted to uh please break it down you know a little bit of you know Delta V um you know space topics >> break it down space man space man
1:50:49
>> breaking it down um it was something I I hope I don't butcher you know apologies Michael Katio if I butcher the name of the order but it was something like you know uh you know America maintaining uh making space great again you know maintaining American leadership in space um there were a couple different aspects of it you know that I enjoyed you know reading for you. Um I think the first
1:51:05
Um I think the first primary like um lens that I put on it is it was just the clearest language that I've ever seen from a government official talking about just making so much of space activity far more regular.
1:51:18
Um it was everything from just like you know sort of uh uh rethinking the regulations around both launch and re-entry to think of these things as like much more you know sort of regular activities.
1:51:27
Um, you know, the example that I always like to provide in relation to, you know, sort of Vardo was, um, you if you studied the airspace around Cape Canaveral over the past decade, it's had a sort of significant, you know, sort of, you know, shift in that there used to just be the occasional rocket launch.
1:51:42
They would do a basically like temporary basically shut down of airspace.
1:51:45
Um, you know, flights would get, you know, sort of rerouted, um, you know, across Florida in order to, you know, sort of accommodate that.
1:51:50
And it was always sort of like a one-off special operation.
1:51:51
of like a one-off special operation. Um at this point if you actually look at like the commercial um you know uh plane traffic patterns around Florida most of it has basically largely shifted like they used to typically actually like fly
1:52:02
along the coast if you think about that in New York to um you know Miami flights it would actually sort of cut along the you know sort of coast to basically you know sort of reduce a little bit the distance now because there's just such
1:52:12
consistent operations the airspace is basically you know predicted to be effectively closed off at basically all times and so all of the commercial airline traffic basically just goes along the like central line of Florida and it sounds like a you know a simple
1:52:23
thing, but it's interesting to think about like how day-to-day um you know uh you know activities and commercial activities are shaped when something is seen as like a regular activity rather than like a you know sort of special occasion. And so you've seen that you
1:52:34
And so you've seen that you know built out around Cape Canaveral where you know now do you have this like rocket flight corridor that is basically just like permanently you know sort of cordoned off.
1:52:43
>> That's really the only place that we have that in the United States at that regular event activity where like it's thought of as something that needs to be cordoned off.
1:52:48
you obviously have Vandenberg, you know, sort of space force base, but still their things are still occasional enough that you don't have the equivalent.
1:52:53
There's also a little bit less like, you know, sort of commercial, um, you know, flight traffic that, you know, sort of goes around there.
1:52:58
It's just a little bit, you know, less in like a, you know, sort of dense urban population.
1:53:01
urban population. But a part of um the you know sort of executive order from uh the president was both you know sort of rethinking and reshaping you know basically launch and re-entry regulation but also clearly pushing for you know uh an establishing of a bunch of spaceports
1:53:15
around America that are meant to have basically you said regular both re-entry and um you know uh uh and launch activity and then thinking through how much burden should there be put on re-entry vehicles that are coming back from space if those re-entry vehicles have a basically what's known as a flight termination system on board. Basically, it's like, hey, if
1:53:32
Basically, it's like, hey, if something's going wrong, just explode the damn thing so we don't necessarily like land in the wrong place.
1:53:36
And so, should you allow for basically like looser um you know, regulations when you know that there is that type of, you know, sort of termination system on board that can basically self- terminate if you're in an off you know, offnominal scenario.
1:53:46
Um and so it was just great to see leadership from you know, the administration and the president in such a like pointed direction of hey this type of activity needs to be thought of as like day-to-day and as common as like commercial sort of flight.
1:53:57
The example that I like to give is, you know, we are now approaching the point of the amount of orbital rocket launches from the United States per year is about the same number of commercial airliners uh landing and taking off in LAX every day.
1:54:12
Is becoming, you know, regular.
1:54:12
I mean, it's still 365, you know, you know, off of, you know, truly, you know, the daily equivalent of LAX, but it's like, you know, it's in the orders of magnitude now of like, you know, some of our busiest airports.
1:54:22
And so, we need to think of these things not as like rocket launch pads, but like spaceports.
1:54:25
And I like that the executive order had like the word spaceport in it like 15 times.
1:54:30
Not like esoteric crazy rocket launch pad thing.
1:54:33
It's like no, you know, airport's a spaceport.
1:54:36
>> Yeah, we kind of do that with the commercial flights.
1:54:37
I I I think like when commercial airliners are flying east to west, they fly at like 35,000 ft.
1:54:43
And when you're flying west to east, you fly at 32,000 ft.
1:54:45
So there's just always a gap.
1:54:47
And you know, you don't even have to think about like am I going to run into someone because you're just everyone's just codified pattern of operation because you don't have to think about that.
1:54:55
And so we need the equivalent for spaceports where it's like, yeah, there's this corridor that allows, you know, for re-entry, you know, so regularly and just like it's not like it's a one-off rerouting of planes, but the planes just don't go there because they know that that's where things reenter. >> Yep.
1:55:07
Anything on the >> what's the update on Firefly Aerospace?
1:55:12
They went out they went public on uh August 7th at 60 down 18ish% since then.
1:55:20
But this is the company that had the the rover >> moon landing.
1:55:24
U what's the general update on the company and um yeah I would >> like how do they fit in the ecosystem?
1:55:33
>> Yeah, I've been spending a bunch of time with uh some of the investment bankers that you know sort of led that you know deal uh and then some of the like long um you know hedge funds that did it.
1:55:39
Um it was exciting to see that of the call it like top eight you know uh mega IPO allocators that are like the long only you know sort of public equity types I think something like seven out of the eight did like meaningful you know sort of positions in that uh Firefly um IPO round.
1:55:56
Um, and so I I think this kind of goes back to like the, you know, super cycle stuff that we were talking about where um, I do think that this stuff starts to become a little bit of like a self-fulfilling prophecy where like you have the president of the administration really leaning in and leaning forward.
1:56:11
Obviously, there's like Jared Isacman no longer administrator, secretary Duffy stepping in.
1:56:14
They're trying to find, you know, interim.
1:56:15
you know, there's obviously things that they're still sorting out unlike our space leadership, but clearly, you know, sort of leaning forward um in a way that is, you know, sort of marching towards, you know, wanting more launch providers, more activity, you know, both military and civil and space and then obviously towards the moon and Mars.
1:56:28
Um Firefly, you know, has had a, you know, somewhat, you know, sort of complicated history.
1:56:33
It's no longer the original founding CEO.
1:56:34
They've had a swap out CEO a few times.
1:56:36
But I do think in this, you know, sort of category of aerospace company, it's really important to focus on just like ultimate like flights and, you know, success in those.
1:56:42
And for what it's worth, Firefly has accomplished two things that I think are quite, you know, phenomenal.
1:56:46
They are the only um rocket launch company that uh basically had this contract with the DoD where the DoD basically had them on guard and then they were only given 24 hours notice and they had to launch a satellite.
1:56:55
And so it wasn't like a oh pre-planned etc.
1:56:56
It was like basically like responsive you know sort of launch is I think that you know term that the Air Force um gave it.
1:57:03
I think it was Air Force um and so very impressive to have done that.
1:57:05
And then obviously you know they've managed to actually you know sort of land on the moon and not tip over.
1:57:08
Um, intuitive machines is still working on that, you know, sort of second uh, part.
1:57:11
Um, so they've clearly accomplished some things.
1:57:15
Financial profile of the business, I admit is like, you know, it's a lot of capex, a lot of burn.
1:57:17
It's not super repeatable, you know, sort of yet.
1:57:21
yet. But the fact that like you know these public you know long you know equity folks are you know buying in at such scale again improves their balance sheet reduces their cost of capital clearly shows some signaling from the equity markets at least that they really
1:57:32
do believe that this type of like lunar infrastructure launch infrastructure is something that has such macro tailwinds behind it that you know even though they're still you know choppy in terms of operations etc um that there's you know sort of clear potential there and you're seeing that with Rocket Lab. I
1:57:43
you're seeing that with Rocket Lab. I mean look at the company it like you know was you know I think a year ago probably trading at like three and a half you know sort of four billion in market cap it's something like a $22 billion you know sort of market cap company today as like you know amongst the you know sort of higher you know sort of you know re multiple you know
1:57:56
sort of ratios um you know in all things you know and >> let's give it up for high revenue multiples >> we love them here we love them >> love them give them to the rocket companies babies >> we don't have enough of them >> yeah not saying going public anytime soon but it encouraging to see you know all the interesting activity >> well you're the one who said you were talking with investment bankers. >> Yeah. >> Yeah.
1:58:15
>> Oh, you know the job of a capital allocator, you know. >> Right. Right. Right.
1:58:18
Um, so I I posted earlier somewhat of a joke, somewhat real breaking group chat sentiment flips.
1:58:24
Bearish investor competence uh confidence dips to the lowest levels since late 2023. Source my phone.
1:58:29
Uh, how are you feeling about general sentiment right now in private markets?
1:58:37
There's been a lot of exuberance.
1:58:37
A lot of people >> should we be fearful?
1:58:41
>> Yeah, a lot of people being greedy.
1:58:42
Should we be is is there any alpha in being fearful?
1:58:46
>> Yeah, we always um you know try to you know you know zigg when others are zagging at a faf obviously.
1:58:50
So when we see this level of exuberance I think that always makes us a little more um hesitant.
1:58:55
hesitant. Um, but yeah, it's been wild to see just the, you know, sort of depth of interest, but in this like sort of K-shaped, you know, sort of dynamic, right, where, you know, the equivalent of like I, I forget who tweeted this today, some relatively prominent venture
1:59:08
capitalist, but or maybe reporter that was basically saying that like private capital markets are basically replicating the like mag seven strategy, but you know, in the private markets where it's like, you know, basically if you look at the like mag 7 versus the like S&P 493, 493 is basically flat, mag 7 or up. you kind of have the equivalent
1:59:21
you kind of have the equivalent dynamic in private capital markets both literally with like the individual logos of like SpaceX, OpenAI, Android and etc.
1:59:28
where it's like, you know, representing the biggest chunk of, you know, uh, you know, capital allocated, you know, ever in the history of private capital markets.
1:59:34
But I think it's also generally true in like the categories where it's like if you're not A and D and if you're not AI, oh my god, like there's like, you know, basically no interest and nobody, you know, wants to you should talk to you.
1:59:43
And then if you're in those categories, the rounds get done in like a, you know, sort of weak and are done at like the, you know, sort of craziest, you know, prices.
1:59:48
Um, you know, I think we always stick to our general strategy of like, you know, ignore the noise, focus on the things that we think are in the long term.
1:59:54
And so, you know, my >> Well, it's just it's it's it's worth it's worth remembering because we just had, you know, we we you know, Figma IPO was, you know, this amazing moment, but if you look back, it's like Andrew Reed did the C at 400 million post.
2:00:07
And you look at the comp the you look at the companies that are doing rounds of 400 posts right now and they were not they were not anywhere near that kind of profile >> in those categories.
2:00:20
you know, this is where, you know, again, uh, you know, obviously there's lots of benefits to doing the HOD category where you're like downstream, you know, financings typically are much less risky, but there's also the, you know, you know, you know, return, you know, compression there.
2:00:31
Uh, so my, you know, sort of piffy joke on it is like while everybody else is focused on, you know, A&D and like AI, um, I'm just doing the series A of like fish killing robot companies where, you know, nobody's really think we tasted Shink on the stream. It was a lot of fun.
2:00:45
Um, what what what does A&D stand for? >> Oh. Oh. Oh. Aerospace and defense.
2:00:51
>> Aerospace and defense. >> I'm sorry.
2:00:51
Spending too much time with capital out here. Lots of jargon. Lots of jargon. Too much jargon.
2:00:57
>> What else uh what else needs to be clarified in space?
2:00:58
Obviously, the the space port, the actual path the rockets travel along is important, but is there is there, you know, appetite for more clear regulation around LEO, Middleear orbit, geo orbit?
2:01:13
Are there even just like re-entry of the pods that Vartis sends up? Like what else?
2:01:19
What's the next thing that we need to kind of standardize around?
2:01:23
>> I like that in my head now I have Middle Earth orbit going.
2:01:25
So >> yeah, Middle Earth orbit.
2:01:26
What's it actually called? >> Medium Earth orbit. >> Medium Earth orbit. I like Middle Earth.
2:01:32
>> I like Middle Earth orbit better.
2:01:32
That's >> Let's take it back. Let's coin it. >> Exactly. Exactly.
2:01:35
Um yeah, I mean there has been, you know, I've heard some like, you know, sort of light rumblings, you know, sort of around this.
2:01:40
Um right now when you look at you know sort of uh general space policy it is kind of split up in a bunch of different areas.
2:01:46
You basically have FAA responsible for all launch and re-entry licensing.
2:01:49
So basically like when you're in the air um then once you're in orbit you have everything from if you want to take photos of other things in orbit that's all regulated by Noah.
2:01:58
Kind of crazy because it's like you know stands for like ocean whatever administration etc.
2:02:03
So it's like why are they regulating pictures?
2:02:04
But it's because they're responsible for all things OSENT you know in in you know photography land.
2:02:07
know in in you know photography land. Um and then if you want to communicate with your satellite you know it's the radio spectrum from the FCC and they even govern like so for example you know one of the you know recent back and forths you know more in the prior administration around Varta was uh they were trying to debate whether or not ISAM so inspace assembly and
2:02:24
manufacturing needed to live in its own specific you know sort of spectrum and everybody had to communicate in there rather than getting to be in like the general you know sort of space you know sort of spectrum um uh uh and so you have all these like sort of split you know regulators and as a space company it's very burdensome relative like okay
2:02:40
if you're like a pharmaceutical company you kind of know that like your only regulator is the you know FDA and that's all that you're you know sort of working with and so um yeah I think you know probably when I think about what needs to you know happen next beyond just the spaceports it's just like make it so that there's sort of more of a single point of contact if you're a space
2:02:54
company rather than having to establish like four different you know regulatory um you know relationships which by the way none of them talk to one another right so that and then probably like a space traffic coordinator like there isn't the equivalent of like we have air traffic control here there's regulations on like how you need to report your position how you communicate etc space traffic control. They like kind of
2:03:09
They like kind of started something actually in the first Trump administration.
2:03:12
It worked through Biden administration and then oddly enough the Trump administration actually shut down their own project.
2:03:17
And so now we're kind of back to like wild wild west again on all things space traffic.
2:03:21
So, it's mostly just like literal like the companies like individually email each other and there's like this one like nonprofit foundation that like collects everybody's space traffic data and then it's like and by the way there's no rules and like by the way if like you're going to be interacting with a Chinese satellite and then like the nonprofit introduces you and the Chinese
2:03:36
company over email it's not obvious like who's going to maneuver what there's no there's no rules of chicken >> it still just counts helps that it's mostly just pretty empty up there so there's not that many you know incursions >> we're going to need regulation around drunk space >> we're going to need uh pilots who live their life one mile at a time. >> I have I have a sort of an odd I have
2:03:56
>> I have I have a sort of an odd I have >> pretty short going 14,000 m an hour.
2:03:59
That's 14,000 miles at a time.
2:04:02
>> What do you what's what's your read on the the future of um the the sort of pilot career path?
2:04:07
It seems like obviously we like many many different types of planes have like great autonomous systems yet we obviously still want pilots uh in the cabin.
2:04:21
How do you how do you when you kind of look at the progress happening on the on the autonomous side?
2:04:28
Where do you think human pilots will fit in?
2:04:35
Man, there's clearly some things that are moving forward like you know uh Secretary Duffy um you know Secretary of Transportation you know has really pushed forward more aggressively than I would have expected on all things like EV tall um you know beyond like visual line of sight you know sort of drones um uh uh you know on like the small scale
2:04:51
you know sort of vehicle autonomy side of things when you think about like a Boeing 747 you know sort of getting actually fully automated man the commercial airline or you know commercial passenger stuff is just it feels so hard and a part of it is other than obviously a recent whole set of events, you know, DCA crash obviously included. Generally, the United States
2:05:07
Generally, the United States has such a flawless safety profile.
2:05:09
I think we've talked about this in like, you know, other >> Yeah.
2:05:15
>> to like get things changed because you have, you know, so much, you know, such a, you know, great safety record.
2:05:19
So, um, yeah, I think, you know, jobs. >> Yeah.
2:05:22
When you I mean, when you think about if you have, uh, you know, hundreds of people in a single metal tube flying through the air, is it worth having a couple more people to be in there?
2:05:32
It's also there's also an economic dynamic where uh I believe the the the the cost of labor for an Uber is about 70% borne by human labor whereas the cost of a 747 flight is like 7% pilot fees even though the pilot No, it's like 50% fuel. 70% fuel. Yeah. Yeah. Exactly.
2:05:55
Whereas the gas in the Uber is like 10% of the cost of the ride.
2:05:58
70% is is the actual human.
2:05:58
And so there's a lot there's a very different economic dynamic there.
2:06:03
But we're actually going to talk to Merlin Labs today that are uh and and Merlin is going public.
2:06:08
They work on autonomous uh flights for planes, autonomous pilots for planes.
2:06:13
So we'll dig in with Matthew George later.
2:06:16
>> My last comment for uh the session is this is my uh you know sort of favorite one where I didn't realize I was going to be on TBPN until the TBPN account tweeted and it was like Dell's on today and I was like I didn't know that.
2:06:24
Look, >> well lo and behold, look at it. I am. So >> that's amazing.
2:06:29
We should just communicate purely over x. com.
2:06:29
It is the everything app after all.
2:06:34
>> Anyway, this is always a great time.
2:06:35
Thanks so much for hopping on. >> Later, boys. We'll talk later. Bye.
2:06:39
>> And uh whether we talked to Delian about the markets, whether you're long, whether you're short, whether you think it's top or a local bottom, head over to public. com.
2:06:46
Investing for those who take it seriously.
2:06:48
They got multiasset investing, industryleading yields, and they're trusted by millions.
2:06:51
Folks, we are joined next by some folks at Anderl today.
2:06:56
Ander posted they are waiting in the reream waiting room.
2:06:59
I believe we will bring them in in just a second.
2:07:00
Uh and Industries said introducing the Anderl 250, the first ever NASCAR race on an active military base.
2:07:07
So welcome to the stream.
2:07:10
I believe we have Jeff and Jen joining.
2:07:12
So welcome to the stream from the W stream waiting room. Whoa. Okay.
2:07:18
Hey, we got >> Can we flip them?
2:07:20
Can they flip their camera?
2:07:23
seeing you horizontal or flipped somehow. >> But good to meet you. Good to see you.
2:07:29
Actually, I've met both of you online.
2:07:31
>> Um, and Jen's been on the show before.
2:07:33
Let's try and rotate the camera 90° if possible. There we go. >> Looking good. And I see some vehicle. Stunning. Both of them. >> Fantastic.
2:07:44
>> How are you doing today? >> Hi everyone.
2:07:45
Uh, my name is Jeff Miller. I'm great.
2:07:47
It's good to see you John. Good to see you, Jordy.
2:07:48
We're here >> in the hangar in Morrisville, North Carolina.
2:07:53
So, we wanted to give you guys a little bit of a window into our big announcement around the Anderl 250. >> Why North Carolina?
2:08:02
Isn't this happening in San Diego?
2:08:07
>> Well, NASCAR headquarters uh in Daytona and in Charlotte.
2:08:11
So, we've been here this week uh not only working with the NASCAR team to make sure we show up into the sport in an authentic way, uh but we've also been meeting with uh several of the teams themselves.
2:08:22
Uh because you might be seeing a Andrew stock car with a paint scheme come June 21st, 2026.
2:08:32
Getting to learn the sport as we make the announcement. >> Yeah.
2:08:34
What is the Andal colorway?
2:08:34
I think of when I think of Ander, I think of like all the mut the slate gray, but then I also think of all the anime colors and so there's a little bit of both.
2:08:43
Is is yellow and gray going to be a brand pattern going forward?
2:08:50
>> Well, black is primary color.
2:08:54
>> We are not getting a lot of audio from you.
2:08:55
I don't know if there's Is there one microphone? What's going on here? >> Can we hear you, Jen?
2:09:02
>> Yeah, >> we can hear we can hear.
2:09:02
What Joe was saying is that okay great.
2:09:08
So saying that black is in not only one of our core colors or not a color as she like to say but our accent tone there is safety yellow.
2:09:18
So anything that you see whether it's in incredible products that Jen and team are designing the the finish you're going to see that safety yellow.
2:09:27
So, in the paint scheme here, uh, we like to call it using the language that they use here at NASCAR versus what you'll hear more in F1 called a livery, what we call livery on our aircraft.
2:09:38
Uh, for our paint scheme here, we really leaned into that black and that safety yellow.
2:09:43
>> The car looks incredible.
2:09:43
You guys cooked as usual.
2:09:45
Can you can you give us a tour of the livery?
2:09:48
I got to I want to see a closeup. >> Is the Yeah.
2:09:51
Can the camera move around?
2:09:53
Can you show us the car up close >> a little bit?
2:09:55
I can hear you like loosely on his microphone, I think.
2:09:58
>> Uh, but if it's like if it's a shared pair of AirPods, probably switching back and forth.
2:10:03
>> Let's see if we can get a tour of the car.
2:10:04
>> Okay, I'm going to take you around the car. >> Okay, >> very cool. >> This is why insane.
2:10:10
That's a huge Andre logo on the front.
2:10:13
You really dominated this. Yeah. Safety. Safety yellow.
2:10:16
>> Primary color is black, right?
2:10:16
The safety yellow comes from our livery scheme on our actual products where we're calling out high viz >> critical warnings and markings. >> Sure.
2:10:28
>> Um and you know as we think about bringing these two worlds together, I really thought there was a lot of symmetry, a lot of kinship between our two identities, right?
2:10:36
I think I love also spending time here getting to know the NASCAR brand.
2:10:40
uh like bringing uh they're leaning really hard into their legacy and their heritage, but they pair that with such contemporary, you know, bold graphics and high, you know, visibility colors, which is um you know, there's a lot of synergy between our identity and their identity.
2:10:58
>> We got to get some Mustang GTDs in the Anderl parking lot after the next tender.
2:11:04
>> Have you Have you seen the Mustang GTD?
2:11:06
It's like a $500,000 Mustang. It's beautiful. Oh, wow.
2:11:08
Is that a Fury in the background, too? Yeah, >> that's right. >> It's hard to see. That's amazing.
2:11:13
>> Jeff Jeff can Jeff can walk you around the Fury.
2:11:16
>> That's a Yeah, that's awesome.
2:11:16
Uh, talk to me about the actual race.
2:11:18
Would this race have existed without Anderal stepping in?
2:11:22
It is like It's not just the logo on a car.
2:11:25
It's not just the logo on the entire car.
2:11:28
It's the logo and the name Andereral on the actual race. >> A whole new race. >> And a whole new race. >> Yeah.
2:11:36
What I'd like to say is that this this race first and foremost is the brain shop of NASCAR leadership and the US Navy.
2:11:45
So, it's our privilege to to be a part of it.
2:11:48
Ben Kennedy, you are going to have the opportunity to to speak with who's got such incredible vision along with Amy, the the race director.
2:11:54
And so, we came on in April.
2:11:57
Jen and I had a chance to go on the Navy base.
2:11:59
And the second you step foot on Naval Base Coronado, this is where the films Top Gun and Top Gun Maverick scenes film that famous eyebar is there and you stepped on an air US aircraft carrier and you feel empower you feel like you are you're there and with people who are doing work that really matters.
2:12:21
So yes, this race would have absolutely worked without us.
2:12:24
I think what NASCAR saw in handrol was a company that was so deeply committed to our American values and our war fighter and wanted to celebrate that story that it was a perfect fit. >> Very cool.
2:12:35
Well, we're going to hop on with uh the Ben Kennedy from NASCAR.
2:12:40
Thank you so much for taking the time to chat with us.
2:12:41
Uh when is the actual race?
2:12:43
Can people buy tickets now? Is it on sale? Uh what's the next step?
2:12:48
>> And how do we get some merch?
2:12:49
>> People are usually asking us that.
2:12:49
Now we're asking you want some.
2:12:52
We can we can talk offline, but but yeah, when when is the >> when is the race?
2:12:59
>> The race is June 21st, 2026.
2:12:59
You can go ahead and get your place your deposits now.
2:13:04
Uh we'll definitely be having merch.
2:13:06
Uh I'm expecting a live broadcast from TBPN.
2:13:10
>> We will be there 100%. >> I can't wait.
2:13:13
>> So yeah, can't wait to uh run this back with you guys in about uh 11 months. >> Fantastic. Incredible.
2:13:18
>> Uh thank you so much and congratulations on the partnership.
2:13:20
We're very excited to beautifully to keep tracking it and can't wait for the race. We'll talk to you soon. >> Cheers. >> Thank you all. >> Thanks for coming on.
2:13:28
>> Andrew also notorious for running uh snarky billboard ads on 101.
2:13:30
If you want to run an ad, go to adquick. com.
2:13:33
Out ofome advertising made easy and measurable.
2:13:36
Say goodbye to the headaches of out of home advertising.
2:13:38
Only adqu combines technology.
2:13:39
Out ofome expertise and data to enable efficient, seamless ad buying across the globe.
2:13:43
Did you see the the the the billboard that they actually dropped on the 101?
2:13:46
Yeah, we talked about this where uh uh Jeff Miller, who we just talked to, uh put up a billboard of the Fury saying uh like no more enterprise SAS.
2:13:56
>> Um because every single I think Trey Stevens got sick of seeing another another uh Enterprise SAS billboard and he said, "I need an Android billboard up there." And so they did it. Um >> they did it.
2:14:07
>> And so we will be hopping on with Ben Kennedy from NASCAR in just a few minutes.
2:14:10
But in the meantime, let's tell you about Bezel. getbasel. com.
2:14:12
Your bezel concier is available now to source you any watch on the planet. Seriously, any watch.
2:14:18
Uh Jordy, where where would you like to go in the post?
2:14:21
We got about >> I have a post here from Ian Roundtree, friend of the show.
2:14:25
He says, "A friend hadn't heard of TBPN and another described it as follows.
2:14:28
TBPN is a very popular RAMP infomercial in the style of ESPN.
2:14:33
Instead of rankings tied to sport leagues, they are tied to how early and how much you invested in ramp or how close you are to someone who did." I said fact check. True. >> We love it. We love it.
2:14:46
>> Uh Daniel Ten Tanero says happy Cororeweave lockup expiry day.
2:14:50
Basically the Super Bowl for guys like us Cororeweave uh down 10% today sell.
2:14:55
I'm not sure I'm not sure how the actual lock down 15%. >> 15%.
2:15:04
>> Still a 48 billion company and a bad run.
2:15:07
Uh, shout out Nick Carter, angel investor in the >> Oh, yeah. That's right. >> What was it? Seed round. >> Yeah. Yeah.
2:15:15
>> You can run the numbers on >> We've most recently uh beat on topline missed on bottom line.
2:15:19
They are uh burning something around 300 something million dollars a quarter.
2:15:23
Um but making a lot of revenue and so >> who hasn't burned?
2:15:29
>> They're investing >> 300 billion investing.
2:15:31
>> 300 million a quarter. It happens.
2:15:31
Uh Perog Agrial uh says he's just setting up his Twitter again.
2:15:38
>> Coming on the show today.
2:15:39
>> Heads down building parallel with some of the best people he's ever worked with creating infrastructure for AIS to search and use the web.
2:15:44
Uh he will be on the show in not too long. Very excited for that.
2:15:48
Buck says STS s a subsidized tokens as a service i. e.
2:15:53
most of the VCbacked AI app layer.
2:15:56
And on that note, a Chris Pike post has hit the timeline.
2:16:03
Oh yeah, >> a Google doc has hit the timeline.
2:16:04
I repeat, a Google doc has hit the timeline.
2:16:09
>> It's called cursor's problem.
2:16:11
>> Product market fit is the story founders love to tell.
2:16:14
Business model, product market fit is the part they often skip.
2:16:18
Product market fit is users repeatedly choose your product.
2:16:21
Business model, product fit.
2:16:22
The extraction of value is sustainably in excess and proportional to the cost of delivering value.
2:16:27
Curser has relied on a subscription model that historically allowed for quote unquote unlimited use.
2:16:33
That's a fixed revenue variable cost setup.
2:16:35
Insurance companies are the canonical example and they employ actuaries to accurately price risk and segment users.
2:16:42
Hypergrow startups rarely have that muscle.
2:16:44
When variable costs scale with intensity of usage, but revenue doesn't.
2:16:47
You're not selling software.
2:16:49
your underwriting risk without actuarial actuarial discipline pricing segmentation caps exclusions.
2:16:58
These models drift into the same ditch that killed movie pass Oyster and force class pass to retire quote unquote unlimited cohorts invert your most profitable user turn because they use the product the least and get the least value.
2:17:09
Oftentimes they can get better value from a competitor who prices them more accurately or with less breakage.
2:17:15
The remaining users are those who extract more value than they pay over time.
2:17:18
Older cohorts morph into deeply negative gross margin. Topline mass rot.
2:17:24
New larger c cohorts can briefly offset the drag, hiding the deterioration in earlier cohotes.
2:17:28
Revenue uh cohorts >> uh revenue grows margin quality quietly decay decays.
2:17:34
A large cause of this problem in fast growing companies is the conflation of subsidies and marketing.
2:17:41
They are both expensive growth strategies, but they do very different things.
2:17:45
>> Marketing buys attention.
2:17:45
It changes who hears about you, not what the product is worth in equilibrium. Subsidies buy behavior.
2:17:51
They put economic value on the product scale, distorting the read on willingness to pay.
2:17:57
>> The 15-minute delivery boom made this vivid.
2:17:59
PMF seemed strong when prices were artificially low.
2:18:01
When prices rose to true cost, demand snapped back.
2:18:04
The quote unquote fit was with the discount, not the service.
2:18:09
Venture lore points to Uber and Door Dash as counterexamples.
2:18:14
The lesson isn't negative gross margins are fine.
2:18:16
The lesson is if there's an operational path to positive margins, i. e.
2:18:21
density, batching or utilization and future pricing power uh quote modes, the temporary subsidies can be a bridge.
2:18:29
Indeed, venture capital is precisely the right instrument to facilitate these kinds of companies, but most businesses don't have such a bridge.
2:18:34
Does this bridge exist for cursor?
2:18:36
Cursor's users expect the best coding performance which is currently delivered by the Frontier models that pins Cursor's cogs to OpenAI anthropic price cards.
2:18:45
Cursor doesn't control two critical dials.
2:18:47
One model performance frontier which is what users demand and model input output pricing what Cursor pays.
2:18:55
If Cursor steps down to cheaper, weaker models, the users who care about performance will notice and churn.
2:19:00
Those who can tolerate weaker models can get them cheaper elsewhere.
2:19:04
If it stays at the frontier while keeping prices flat, the variable real cost to service their heaviest users will explode.
2:19:10
In an effort to combat this cursor has been forced to raise prices and institute usage caps leading to user outrage and churn anytime unlimited shows up in variable cost businesses, PMF becomes a permanently open question.
2:19:22
Are users here for the product or for the subsidy?
2:19:24
Would they still use it as much or at all at true marginal cost?
2:19:28
until cursor prices consumption in proportion to cost, it cannot know.
2:19:33
cannot know. uh last couple >> so interesting that we're going into we're leaving the world of zero marginal costs like this is the Ben Thompson intellectual like aggregation theory regime of the past two decades like since you know if you're trying to understand Google you had to understand the what it meant to
2:19:56
serve an incremental user at zero marginal cost and it was something that Wall Street and retail and lots of folks Even a lot of VCs like could not wrap their minds around for a long long time and there was huge alpha in that and now it's become consensus right when the the economic like uh equation is now flipping. I just wonder how long how
2:20:16
I just wonder how long how long we'll be in this in this regime if it's here to stay if we plateau and reasoning models are always expensive.
2:20:25
And then I wonder I I do wonder how how bad the gross margins are at various companies.
2:20:32
how bad this subsidization subsidized tokens problem is.
2:20:35
And I also wonder if there are any companies like there have been plenty previously leg like like previous era SAS companies that have said have gone out and said hey we're an AI company now and I wonder if that's hurt their gross margins.
2:20:49
I've been picking at that with a few founders that have come on said like so like is your inference bill growing or is it material at all?
2:20:57
And basically what I'm getting at is like is it affecting your gross margins?
2:21:00
Yeah, we we were asking Rahul from Julius about this and and he said that it's it's it it's a thing, but it's not >> incredibly material.
2:21:11
>> There's going to need you just have to find a way with with the application layer company that you build to deliver enough value that the actual inference is under the hood very efficient.
2:21:20
And then I do wonder if cursor is in a unique unique position because it's such a raw interface to to to the frontier coding model and the quality really matters.
2:21:32
Uh but yeah, could they could they swap out Deep Seek R2, save 10 times as much money and then >> you know and also to finish off post he says Kersel will have to choose what it wants to know.
2:21:43
If it keeps subsidizing the heaviest usage, it can keep growing but it cannot claim >> uh unambiguous PMF.
2:21:48
If it starts charging in proportion to cost, some usage will fall and what remains will be the market.
2:21:54
The burning question that should nag founders, do I have demand for my product or for my subsidies?
2:21:57
So, >> it seems cursor is without a doubt beloved product. >> Yeah.
2:22:04
And it seems very reasonable to subsidize for a while and really get people installed and hooked on the product and then raise prices to to the point where you are delivering value.
2:22:14
Like when you think about what the value of an AI software engineer that could be justified in the $50,000 a year, $100,000 a year territory, that doesn't seem that crazy to me if you're actually getting that leverage out of the system.
2:22:28
But the idea of saying, "Oh yeah, my my my IDE bill is 10k a month is crazy." Um, but it could be. I don't know.
2:22:36
>> Um, anyway, >> before we get in to our next guest, uh, Ben, uh, we have some news breaking.
2:22:42
Coher just raised 500 million at 600.
2:22:45
>> Go for Aidenz >> on the transformer paper.
2:22:48
One of the greatest to ever do it.
2:22:50
>> Nick on X says, "We are so back." >> We are so back.
2:22:52
Congratulations to the folks coher.
2:22:55
>> Well, let's bring in Ben from the reream waiting room.
2:22:58
>> Ben Kennedy from NASCAR. How you doing? Ben, good to meet you. >> Good. How are you guys doing?
2:23:02
Thanks for having me on today. >> Of course. Of course.
2:23:05
Um why don't you kick me off with an introduction on yourself and how you fit in the Orton NASCAR organization? Yeah.
2:23:12
So, my name is Ben Kennedy.
2:23:12
I am a EVP here at NASCAR, but we get the opportunity to wear a bunch of different hats.
2:23:19
So, um we're responsible for the NASCAR own track properties, NASCAR regional team, emissions team, people um making improvements to our racetracks, and then importantly, what we're talking about today is schedule.
2:23:29
So, dreaming up the next year's schedule, and we're getting ready to announce our 2026 schedule here pretty soon.
2:23:35
But, you know, one of our big projects, which I know we're talking about today, is the race on the base next year.
2:23:43
>> We're bringing our NASCAR Cup Series, our Expinity series, and our Craftsman truck series to San Diego for the first time and for the first time ever on a military base.
2:23:50
So, really excited about it.
2:23:52
take me through the shape of like all of the different NASCAR events because there's like, you know, the there's the tentpole events that people kind of know and love and then there's a whole bunch of other different uh uh properties, activations, events, races.
2:24:06
Um, kind of give me the shape of what NASCAR is today. >> Yeah.
2:24:11
So, we compete across 38 race weekends throughout the year.
2:24:14
36 of those are points races and then we have two exhibition races.
2:24:17
And uh our season it starts in February with the biggest race of the year, the Daytona 500.
2:24:22
So we're a little bit different in the sense that we start with our Super Bowl and then we will go pretty much straight through the entire year.
2:24:30
We'll have one or two weeks off to the first week of November where we'll have our championship race which will be in in Phoenix this year.
2:24:37
So we uh we're competing every single season.
2:24:39
We have um three different national series.
2:24:42
So you you can almost think of it like high school or or college that kind of leads to the pros.
2:24:48
Um but we have our Cub series or Infinity Series which is our last step for a lot of drivers before they go up to the Premier Series.
2:24:54
And then our Craftsman truck series which are um a little bit different style of of vehicle.
2:24:59
It's more of a pickup truck style.
2:25:03
>> Uh and that's where a lot of drivers will start um their national series career as they they begin touring with us.
2:25:09
So, >> and then how does the race on the base fit into that structure? >> Yeah.
2:25:14
So, the race on the base will be um it'll be in the middle of our season.
2:25:18
So, Amazon Prime will cover it for the Cub series >> and we'll have the Expandy series and truck series which will be on Saturday and Friday that weekend.
2:25:25
But, it's June 19th to 21st.
2:25:27
And essentially what we're going to do at Naval Base Coronado is we're going to build a temporary threemile circuit that will go around the base.
2:25:36
So, anyone that comes out, we're going to create kind of these different neighborhoods of experiences that fans can go and see.
2:25:41
So, you have the start finish line where you'll have, you know, a lot of the hospitality, the garage area.
2:25:47
People will be able to get down and and see and hear the cars.
2:25:51
They'll go out towards the bay side of San Diego where they'll go past an aircraft carrier or two, and then eventually we'll come out onto the tarmac where they'll be weaving their way through F-18s and Ospreys and all sorts of fun stuff.
2:26:04
So, >> don't have the the course design finalized yet, but around a three-mile course, we're going to build temporary on the base. >> Yeah.
2:26:11
Talk about >> what do you think the uh expected lap time will be for for a course of that size? >> It's a good question.
2:26:19
I would say, you know, we've run a couple models on it.
2:26:23
We do a lot of our sim work virtual before we we commit to anything.
2:26:26
It's around a minute and a half, a minute 40 seconds or so.
2:26:30
It's it's a pretty high speed track.
2:26:33
120 miles an hour on average, something like that, I imagine. That's that's wild.
2:26:36
That'll be a lot of fun to watch.
2:26:38
Uh talk about partnering with Anderoll.
2:26:40
Um who they are trying to uh to reach.
2:26:44
Uh they have customers, they have employees, they have they have fans of the merch.
2:26:47
Uh uh why was it a good partnership?
2:26:49
Uh what are you excited to like deliver to them on the marketing side?
2:26:53
And what did they bring to the table other than you know just money?
2:26:56
>> Yeah, we couldn't ask for a better partner than Anderil and and we forget if we approached them or they approached us.
2:27:02
us. I think we reached out to them >> uh you know through the grapevine and we had some conversations earlier this year, you know, probably five or six months ago with them and uh they had a ton of energy around the event and you
2:27:15
know for us we want to find partners and and companies that we can do business with that are both strategic in the sense that we're going to add value to them and hopefully they'll add value to the weekend which I know they're going to do. um but partners that are organic
2:27:26
to do. um but partners that are organic as well and you couldn't think of a more natural fit than a partner like Anderil um you know you look at all the different types of technology that they're creating both on the software side and the hardware side it's incredible and you think about precision speed and innovations in technology that's what our sport is all about so couldn't ask for a better partner than Android really fired up about the
2:27:51
announcement today this is been a long one coming we had a couple of Easter eggs that snuck out um about a month ago when we had the announcement, but glad to be able to be talking publicly today about it and uh you know excited to see
2:28:05
there's a lot of elements that they're going to be bringing to the table over the next 11 months and then especially on that weekend that that fans will get to to go out and experience as a part of it. >> What's the path? A lot of our audience >> What's the path?
2:28:15
A lot of our audience uh are founders and executives at tech various technology companies.
2:28:22
what's the path into partnering with uh NASCAR itself or uh various teams and and uh events?
2:28:34
>> Yeah, I would say um if anyone's interested, reach out to us.
2:28:35
Um we can I'm sure we can find a way to provide contact info and whatnot, but you know, typically um we'll have conversations with whoever might be interested.
2:28:44
I think the good news is there's a lot of interest around this event and whether it's NASCAR um as a sanctioning body or a team or a driver, we have connections to to everyone in the industry.
2:28:55
So happy to put them in touch along the way. >> That's amazing.
2:28:59
Uh thank you so much for hopping on the stream.
2:29:01
Congrats on the partnership >> on the announcement. Looking forward to it. We'll see you there. >> Yeah.
2:29:07
Look forward to seeing you guys out there. It'll be a blast.
2:29:09
>> This is going to be great. Talk to you. >> Cheers, Ben. >> Bye.
2:29:11
If you're looking to go to San Diego, you got to book a wander. Find your happy place. Find your happy place.
2:29:19
Book a wander with inspiring views, hotel, great amenities, dreamy beds, top tier cleaning, 24/7 concier service.
2:29:24
It's a vacation home, but better folks. Jordy, warm up.
2:29:30
>> We uh I just want to say Jessica waiting in the waiting room.
2:29:33
>> Live streaming from an aircraft carrier for the race.
2:29:37
>> We can do it for sure. For sure.
2:29:37
Well, let's bring >> let's bring in Jessica from the from the reream waiting room and from Solo. >> How you doing? Good to meet you. >> Welcome to the show. >> It's good.
2:29:48
Uh would you mind kicking us off with an introduction on yourself and the company? >> Absolutely. I'm Jess.
2:29:52
I'm co-founder and CEO of Sol.
2:29:55
Uh we're an agentic process automation platform that helps businesses automate their most operational workflows uh using AI by doing workflows the way that humans do.
2:30:06
name one process you've automated with an agentic workflow. >> Yeah, absolutely.
2:30:10
Um, as an example, we we work with companies across a whole bunch of different verticals, but >> um, for example, for some of the logistics companies we work with, you can imagine they operate on internal portals, external systems, spreadsheets, everything in between.
2:30:25
And to actually get a shipment out end to end, it requires a lot of manual work.
2:30:28
And so, uh, Solo will type data in, it will take down data, it will, you know, do all the coordination that gets the shipment end to end, um, and work in that fashion.
2:30:40
>> Uh, give us the news today. Anything to share? >> Very exciting day.
2:30:43
Uh, we just raised a 17 million series A led by A16Z. >> Congratulations.
2:30:52
>> Don't leave out conviction.
2:30:55
>> I was just about to get there.
2:30:55
uh with continued support from our seed lead conviction as well as Y cominator. >> Oh, fantastic. >> Amazing.
2:31:01
That's a great uh great lineup.
2:31:04
Talk to me about what the state-of-the-art is in actually building an agentic workflow on top of kind of an internal dashboard.
2:31:11
We've been hearing a lot about um like the big labs setting up reinforcement learning environments with verifiable rewards.
2:31:18
They're cloning Door Dash.
2:31:21
They're creating copies of Amazon. com.
2:31:23
So, the future version of Shiao GPT will let me order headphones or gongs or whatever I need for the show um just within that chat interface.
2:31:30
I imagine that that if you're talking about some custom piece of logistics software or some internal portal um there's probably some things that you can get out of the box just by using Frontier Reasoning models and and uh and web browsers and agentic browsers.
2:31:44
But what what extra steps do you have to take to actually deliver a ton of value? >> Absolutely.
2:31:51
Um, let me give a little context.
2:31:53
Uh, 100%, you know, everyone's talking about enterprise AI agents and computer use and VLMs, but there's a there's a gap between where those are at today and how that gets into productionized workflows, right, for enterprise companies at scale.
2:32:07
>> Um, you can imagine that, you know, for the first time we have these models that can reason and they can act like humans.
2:32:14
they can understand how these platforms are working and we don't need to you know build an integration or like an API uh connection into platforms in order to be able to do work on them.
2:32:22
Um but uh instead we can just you know mimic the way humans do.
2:32:27
Now that said, uh if you've tried operator or you know computer use agents or any of these models, they're they're pretty unreliable for individual tasks and then you can imagine if you're scaling this up to, you know, hundreds of thousands of hours of real world work, it's going to be really tough uh to do so.
2:32:42
And so what sol does is we take an approach where we use a combination of deterministic as well as more deterministic uh non-E non-deterministic automation um and sort of change the abstractions on which things are built so that you know uh you are able to for example send a shipment across five different platforms reliably and at scale.
2:33:03
And so we use these models but also a whole lot of different architectures on top of that to make things uh very accurate and be able to plan and you know build guard rails around the processes that people have.
2:33:15
>> Did you start with a focus on logistics or is that somewhere you ended up after you know talking with a bunch of different companies in various >> after I demanded a single example to dig into? >> Yeah, absolutely.
2:33:27
I mean uh today we we work we have workflows across a lot of different verticals.
2:33:33
So Sol does finance workflows and KYC.
2:33:36
Sol does uh healthcare workflows and enters data into EMRs.
2:33:39
Logistics is a big part of that.
2:33:42
>> Um I would say that the two biggest verticals that Sol has a ton of done a ton of work in today is logistics and healthcare.
2:33:49
>> It's by no means where we started actually.
2:33:50
Our early customers were in uh legal and in healthcare I would say mostly and some insurance as well.
2:33:55
Um, but I think logistics just turned out to be an industry that we discovered a few months ago.
2:34:02
Uh, turns out there's a million perfect automations for solar there and today we work with a whole bunch of really large companies uh, in the industry who are using it for all kinds of different applications.
2:34:14
>> What's been the employee reaction at the companies that you guys work with?
2:34:17
I imagine a lot of these employees are stressed out, overworked, kind of, you know, exhausted from managing all these different surface areas.
2:34:24
So my my sense would be that they they welcome a tool like this, but what's your read? >> Yeah.
2:34:31
>> Uh for sure I I think that it really meaningfully changes the the way people work.
2:34:37
Like for example, some of the companies that we operate at, you know, we're we're saving people from doing the very boring copy pasting, very manual, repetitive work so that they can focus on sales and customer relationships and everything else that goes into their business.
2:34:49
Um one of the customers that we work with, they do a few hundred million in revenue.
2:34:54
We started working with um uh about half a year ago and their their kind of famous data point is that they haven't had to hire a single person at their company since bringing on Solo because they've just been able to do more with the same amount of people that they have on hand and they can kind of scale the revenue of their business without having to scale headcount.
2:35:11
Yeah, I've heard that from uh a founder building uh in the AI rollup space and he's buying companies and he's telling the management team, we're not like traditional private equity that's just going to like >> gut gut the company reduce headcount >> on growth.
2:35:26
>> We're going to focus on growth, but we don't need to add a bunch of incremental headcount.
2:35:29
We're just going to make you more effective.
2:35:30
So, it's a great value prop.
2:35:32
>> There's a big uh kind of discussion online today about gross margins at startups.
2:35:37
Can you walk me through the the the economic thinking of the of your buyer, your customer?
2:35:42
Uh how are they thinking about um the value that you deliver and then your costs?
2:35:47
Because obviously I imagine that these workflows use reasoning models.
2:35:54
The token bills are probably significant, but you're delivering hopefully a lot more value than that.
2:36:00
So, how are people responding to uh the idea that that okay, there's there's a company that has real variable costs with the amount that I demand or that the amount that I'm using this tool.
2:36:13
Are they receptive to a consumption model?
2:36:14
Are they are they receptive to price increases?
2:36:16
What's the what's the vibe been like from the from the AI buyer world? >> Yeah, absolutely.
2:36:23
I mean, I think the most direct way to justify value is just how many hours are you saving, right?
2:36:29
That's the the most straightforward way and the traditional way automation is justified.
2:36:32
Um on top of that for a lot of businesses you actually see topline revenue lift like if you start powering their core operations and they can just do more business.
2:36:40
So it becomes a very different equation and I think that's a lot of the ways customers see sols.
2:36:49
It really doesn't make sense to you know paying a computer use agent at every single step that you're doing.
2:36:53
You can imagine that will get super costly. It will be slow. it will be unreliable.
2:36:57
will be slow. it will be unreliable. And so we have a lot of infrastructure built on top of that so that um the things that you know you don't necessarily need models for that are very straightforward we can just you know apply more traditional automation and then areas where you need reasoning you need
2:37:10
flexibility you need workflows to adapt then obviously we use more models there uh and so costs aren't crazy um and I imagine those will keep going down over time >> fantastic well congrats on the round and congrats on all progress >> yes hiring hiring lots of engineers we're based in New York um if you're interested please reach out. We are
2:37:27
We are growing very quickly. >> Fantastic. >> Amazing.
2:37:31
Congratulations on the milestone and we'll see you back here soon, I'm sure. >> Yeah.
2:37:35
Have a great rest of your day. Cheers.
2:37:36
>> We'll talk to you soon.
2:37:36
And we have Prag Agarwal, the former CEO of Twitter, >> man.
2:37:41
The myth >> in the reream waiting room.
2:37:42
He is launching a new company today and we are pleased to have him join the show. Welcome to the stream. How are you doing? >> Hey. Hey John. Great to be here.
2:37:55
>> Great to have >> so much to talk about.
2:37:55
Um, let's start with the new company.
2:37:58
I I mean, I'd love your thoughts on on live streaming on on Twitter and X and all the things that we're doing, but let's let let's stay focused on your company.
2:38:06
What did you announce today?
2:38:08
Uh, should we ring this gong for you? >> Let's do it.
2:38:13
>> Uh, give us a give us a breakdown.
2:38:13
What what how are you describing the new company?
2:38:19
>> Um, we're building parallel to build infrastructure for AIS using the web. >> Mhm.
2:38:24
It's born out of this observation that the web was built for humans. We built Twitter.
2:38:28
You're thinking about people in browsers, in apps, and you're designing stuff for humans. >> Mhm.
2:38:35
>> And two years ago, as I was thinking around like AIS are going to use the web, and that's going to be the primary user of the web.
2:38:41
And it's going to be at massive scale like thousandx, a millionx of what we've ever done on the web. >> Mh.
2:38:49
>> I think that needs new infrastructure, new business models.
2:38:51
And that's what this company's about.
2:38:54
Today the product we shipped is our deep research product.
2:38:56
Now I'll tell you more about it. >> Yeah.
2:38:58
Um >> how long when when did you actually start the company?
2:39:01
It sounds like you've been in stealth for a while.
2:39:04
>> Yeah, a year and a half ago.
2:39:04
Been building a lot of infrastructure in that time. A lot of infrastructure.
2:39:09
>> So talk about talk about the the first product. Break it down.
2:39:12
>> Um we shipped a deep research API today.
2:39:15
So you're you're use you're users of Chat GPT and se I'm assuming several deep research tools.
2:39:20
Yeah, >> we have an API product that any enterprise developer can integrate into their applications, into their workflows. Mhm.
2:39:29
>> Um, since we have reimagined the search stack as well as the technology for the web from the ground up, we can outperform OpenAI's deep research.
2:39:37
We can outperform every leading models deep research quality uh in terms of their benchmarks, others benchmarks uh and with real customers and I think that's the product we've announced today.
2:39:53
>> So break down some initial use cases for the product.
2:39:56
Let's say I have a a SAS comp like a vertical SAS company is this and and I'm building or or or or something you know I could imagine somebody's building SAS for sales reps or you know a CRM provider is this something they would be running you know deep research style queries within the product.
2:40:16
>> Yeah people do all kinds of things.
2:40:16
So in the sales CRM context, right?
2:40:18
There's a lot of people who have this massive current customer list, prospect list, um they want to go, I wish I knew this about this customer.
2:40:27
I wish I knew this about this customer.
2:40:30
So they just like add a bunch of data to have the CRM now have a lot of information that wasn't in there, but was on the web and now it's in their systems, in their CRM.
2:40:39
They're ranking based on that.
2:40:41
They're prioritizing based on that.
2:40:42
prioritizing based on that. um they're able to go do meeting as if like before you meet someone pull everything from my internal information that I have about the customer and our interactions with them but also pull everything that's happening with their business what
2:40:57
they're talking about what they're doing what they're shipping >> how do you think about the paro frontier I mean Google's I mean I believe they have Gemini APIs and they have and and they focused on kind of this this idea of the trade-off between cost and quality of result and and in intelligence is a function of of cost ultimately. Where do you see pockets of
2:41:16
Where do you see pockets of value?
2:41:18
Do you just want to go way further?
2:41:20
Doesn't matter how much it costs companies are willing to pay or are there is there a different kind of angle on the paro frontier that you think is unexplored?
2:41:30
>> Glad you mentioned the PTO frontier.
2:41:30
I think we obsess about the parto frontier, but I think both are important.
2:41:35
One, at every price point, you got to be the best. >> Mhm.
2:41:39
And then for someone who who has no price sensitivity, you got to be the absolute best, >> right?
2:41:48
And I don't think you get to do the one without the other. >> Yeah.
2:41:52
>> The way we push is we p first push quality and accuracy for use cases as high as possible.
2:41:58
And once you can do that, it actually you can do a lot of work to make it cheaper by trading off a little bit of quality out. Mhm.
2:42:10
>> And that's the motion you take on.
2:42:10
But you got to be on the PTO frontier for every use case to be operating at the scale that we want to operate at, which is like every application, every workflow, anytime, everything should use AI. >> Yeah. >> So, so who are you?
2:42:27
>> So, do you do you uh how do you think about competition in the category?
2:42:29
Are you competing with OpenAI's API?
2:42:31
Uh is that the wrong way to think about it? >> Yeah.
2:42:36
And I'm interested in in where where do you see pockets of unexplored territory?
2:42:41
Like one of Google's advantages is that they have the TPU, but there are lots of uh new chip companies that offer different tradeoffs for inference.
2:42:48
And what what are you excited about in terms of like differentiation?
2:42:54
>> So one, we use models from OpenAI Google and we use several open source models in different places.
2:43:00
We also train in our own models.
2:43:02
We build our own index, crawl, ranker, reasoner.
2:43:04
Uh, and so what we are is a little bit of a complement to the be models if you're building AI applications.
2:43:15
>> Now, of course, OpenAI does bundle a search tool with their API. >> Mhm.
2:43:21
>> And that we compete with and beat on quality.
2:43:25
So we beat not just that tool.
2:43:27
In fact, like the benchmarks for the products we ship today, we beat humans doing hours of work through a deep research product.
2:43:35
Not just because we're cheaper or can do it faster, but we're more accurate for a bunch of workflows than hiring a bunch of people to do it, right?
2:43:45
And that I think is really exciting.
2:43:47
Uh as we work with more and more customers, we're able to replicate that and that's an amazing moment that we've found this year.
2:43:54
Do you think there's room for differentiation by uh finding a beach head market that is a subcategory?
2:44:02
I just imagine like we've seen this with Harvey and legal and they're obviously uh you know HIPPA compliance and medical and uh we just talked to uh Jessica and she was saying that like they were YC company.
2:44:14
They tried a bunch of different things, did some healthcare, some legal, wound up in logistics and had a bunch of luck there.
2:44:20
Are there any glimmers of beach head markets on the horizon for you? um almost too many.
2:44:25
So we we're seeing like a bunch as you mentioned like CRM, sales, sales intelligence, marketing.
2:44:32
>> There's a bunch happening there.
2:44:32
There's a lot of people innovating and there are customers.
2:44:36
>> So if you think about our customers, our customers are often >> vertical people who are in various markets.
2:44:41
We have people doing uh science research on like papers published or on sort of clinical trials that are happening or the data coming out of those we people we've like investors like hedge funds to PE funds to venture funds >> using us because like information on the web is what gives people alpha and people can be so creative and now can create like scaled programs for figuring out where to spend their time.
2:45:12
Instead of doing one query at a time at chat GPT, you write a script, you do a thousand queries on our deep research system, you get better quality and you figure out where you want to spend your time.
2:45:23
>> And we even have coding agents that use our search tools >> when they get stuck. >> Wow.
2:45:28
>> So if you an agent like it's just like so in beach head markets, I think it all converges in the way I see right.
2:45:34
>> It's all agents out there.
2:45:34
You can have an agent for sales.
2:45:36
You can have an agent for investing.
2:45:37
You can have an agent for choose your favorite industry.
2:45:43
And every agent needs >> horseback riding.
2:45:47
>> How has it been uh building you know quietly out of the public eye?
2:45:49
Have you enjoyed it >> a lot?
2:45:56
>> Also respect for founders like I I don't know this is I ran a large company, right?
2:46:01
I was at Twitter 11 years. >> Yeah.
2:46:03
Being a founder, taking something from like nothing to having like a bunch of >> This is your first startup.
2:46:12
>> It's my first startup. >> Wow. Wow. >> Figuring it out.
2:46:14
Like I'm massively unqualified for the jobs, but those are the only jobs you really want, right? Yeah, of course.
2:46:20
>> You want a job that you're qualified for?
2:46:23
>> Just got to figure it out.
2:46:24
>> I want to I want to dig into a little bit more granularity.
2:46:26
I obviously it's launch week, but chat is asking us, "What did you get done this week? break it down for us.
2:46:36
>> This week I got back on Twitter. >> Let's go. >> Ring the gong. Ring the gong. We're hitting the gong. We're hitting the gong.
2:46:43
Great to see you back on the timeline.
2:46:45
>> See, congratulations on the launch. I think that's huge.
2:46:50
>> And I don't know if you caught this.
2:46:50
I think I almost broke Twitter. >> Oh, really?
2:46:56
as I was tweeting like this morning at like 8:00 a. m.
2:46:57
I'm like trying to get my tweets out for the first time like in years, right?
2:47:01
And Twitter is glitching.
2:47:04
>> Then I go on down detector. >> Okay.
2:47:08
>> You might have brought it down.
2:47:09
>> And I was like, what timing, man? Like >> Yeah. Yeah. Timing launch.
2:47:12
Timing launch on the timeline is is like one of the hardest dances you have to do.
2:47:18
Uh you can get steamrolled by a viral thing.
2:47:22
There's a good thing going, but it was a good day. Yeah.
2:47:24
And we're glad that you were able to make the show.
2:47:26
could have been like a surprise deepseek launch. >> Yeah, exactly.
2:47:29
But I think you broke through and uh we're very excited for you.
2:47:33
So, congratulations on the launch.
2:47:34
Thank you so much for stopping by and have a great rest of your week. We'll talk to you soon.
2:47:38
>> Yeah, great to catch up. >> Have a good one. Bye.
2:47:42
>> Up next, we have Eric from Bolt.
2:47:42
new in the reream waiting room. Let's bring in Eric.
2:47:47
I'm very excited to talk to him because Bolt is launching a new business model.
2:47:51
the margins on prompting will go to zero and the companies will make money from hosting websites and apps instead. I want to talk about it. How you doing, Eric?
2:47:58
Good to have you on the stream. >> What's happening?
2:48:01
>> And good to be here, guys.
2:48:01
Yeah, long long time watcher.
2:48:02
Good to be uh good to be on the show. >> Fantastic. It's good to have you.
2:48:05
Uh uh introduce yourself.
2:48:07
Give us a little background on the company and then I want to go into the business model questions because we've this has been top of mind for us all show. >> Yeah, for sure. Yeah.
2:48:13
So, uh company's called Bolt.
2:48:15
New uh formerly known as Stack Blitz.
2:48:16
We're in the process of swapping the logo.
2:48:18
got to get new merch printed.
2:48:22
>> Well, we've been around for eight years now and we actually just hit the eight year mark two weeks ago.
2:48:25
We're actually about to go out of business.
2:48:28
>> Success like a year ago >> and um so what ended up happening was we pivoted and launched Bolt and uh in the first two months we went 0 to 20 million of ARR and wow was really one of the first text app uh tools that that out are you beating the allegations 0 to 20 million very impressive.
2:48:46
What were the gross margins?
2:48:47
Can you give us any insight here because everyone on the timeline's saying anyone with a big revenue ramp it's all suspicious but uh give us give us a white pill.
2:48:57
>> Yeah so our margins have actually been uh really good.
2:48:59
So like uh typically it's been around 40%.
2:49:00
And a big part of that is that uh you know for for the text to app stuff part of it is the inference where you're having the codegen uh you know AI model spit out code but you actually have to run it somewhere.
2:49:12
And so that's like the technology we've been making for the past 8 years.
2:49:14
Basically allows you to run full development environments in a browser using the end user CPU.
2:49:19
So it doesn't cost us anything. Right. >> Okay. Got it.
2:49:23
>> That's a big part of why our cogs are are a lot better.
2:49:25
But um but for us is like you I think it what seems to be inevitable is that margins are just going to go to zero on this stuff.
2:49:32
And so we're we're kind of skating to the where the puck is going because unpack that because a lot of people are saying margins are going to go to 100%.
2:49:43
Margins are going to go very high because inference is going to get very cheap.
2:49:45
You're saying margins are going to go to zero. Break that down. Why would it go to zero?
2:49:48
Is that just competitive dynamics from the competitors and the legacy companies and just all sorts of different competitive pressure.
2:49:55
It >> I think it's really Yeah, good question.
2:49:58
So, I mean, it's really like like we could go and use the previous generation of models, but and charge users, you know, what we're charging for Frontier stuff, but the problem is like >> you're not going to be able to build uh you know, quality apps, right?
2:50:08
Because the the the Frontier models and agents are incredible and they just keep getting better.
2:50:14
getting better. And so the kind that you know if if you're if you're expecting to as a as a company reselling inference unless you're a model provider >> uh you know your choice is you can either take a margin and give worse models than what are the frontier
2:50:29
>> or you can or you're or you're saying hey we want to always have the best most capable stuff um and and and not take a margin and instead and then but what that means for your business though is that you have to be going and making money in other ways than just reselling inference, right? And so if you kind of
2:50:44
And so if you kind of look at the previous generation of site builders and hosters as an analogy, >> you know, if you go to like Wix and Squarespace, they don't charge you to like build a website. It's free.
2:50:53
Of course, you don't have to pay for AI or whatever, but like you drag and drop.
2:50:58
What what they bill you for is to host the thing, right?
2:51:01
>> And and that's how they have a very uh healthy LTV and average customer, you know, contract length, etc.
2:51:05
, is that you have your website host and that's the primary ongoing value that's being >> Yeah.
2:51:10
And and I people uh people people that have been kind of trying to u track this market of of prompt to site or prompt to app have have that you know the critics critics have talked about churn but my read on it has always been that it seems like there's been a fairly healthy market of just website builders for a long time.
2:51:30
It hasn't been monopolistic.
2:51:32
You've got the Wixes, the Squarespaces, the Web Flows, the Framers, and you know, all those companies have been able to carve out solid uh solid businesses.
2:51:41
And so, yeah, this launch makes a lot of sense, which is, hey, come on the platform, generate whatever you want, but then if you're generating a site, a marketing site for your business, you know, we we'll host it, you know, for you indefinitely. >> Yeah. Exactly.
2:51:57
And so it kind of you move like for us as a business it it moves the focus for us to having the best services and like workflows uh for the customers we're serving.
2:52:06
So like that's for people building hosting website that's hosting services.
2:52:10
So we partnered with like Netifi and Superbase for powering the stuff that we've got today and they've scaled this for you know some of the the you know heaviest products like Uber and Twilio in the past.
2:52:20
So you can you can build like you know to millions of people now with Bold.
2:52:24
Um, and on the B2B side, we're selling to folks that are I mean, basically replacing uh Figma with Bolt for doing prototyping and rapid product development.
2:52:32
And that's just very sticky because you're you're working on a team.
2:52:35
And it's for the same reason that Figma is a collaborative product is sticky.
2:52:39
And so that's where really where we're focusing because that's like, you know, the the retention is incredible um on those use cases from our numbers, right?
2:52:47
>> Are you feeling acceleration deceleration from the frontier labs on uh coding products?
2:52:52
Uh lots of people were kind of uh on the timeline at least disappointed in GPT5.
2:52:57
It feels like it's more of a consumer product now, but uh in general, you know, it used to be, you know, uh every six months we got something that was completely breakthrough and now uh you know, the the releases feel a little bit more incremental, but what are you feeling uh just from the foundation model lab side?
2:53:16
Because we are seeing a lot of progress there.
2:53:17
The benchmarks are improving, the context windows are getting bigger.
2:53:19
Are we accelerating, decelerating, kind of in a in a rebuild mode, kind of rethink, need to go back to the drawing board and do some AI research?
2:53:28
>> Yeah, I think I think in, you know, for all the the frontier labs, I mean, they they should all they're they need they need to keep uh hitting, you know, more and more breakthroughs, right?
2:53:35
But from a from a on the ground perspective of a company like us looking at these models and running evals and and measuring them.
2:53:42
Um they're you know the the models themselves may be seeing like maybe more incremental improvements >> but what's what's really I think the big unlock is that not only are they getting better at just like coding but they're getting better at being designed to be used as part of agent systems.
2:53:56
used as part of agent systems. And that's like a that's a huge thing because I mean if you think about hum how humans think like if I asked you to like uh you know rattle off like you know the latest financial numbers or something it's like you might be able to
2:54:09
like pull that up but realistically you need to like think on it you get some things wrong on on zero shot and you need to like kind of go do some research bounce around and so like that's what really agents are kind of replicating in my view is like how we as humans actually think and work. Um, and so how
2:54:21
Um, and so how do you really tune the models to to be good at going and and going down different thought paths, killing ones that don't actually look promising, etc. , etc.
2:54:30
And you look at claude code is is probably the best example of this in the code generation world.
2:54:34
Um, and so there's remarkable uh, you know, progress being made in that realm of things, right? Yeah.
2:54:41
So there's a lot seems to me there's a lot of room to run just on you know incremental improvements on the zeroot aspect of models but then also how these things are going in reasoning and working >> system.
2:54:53
Yeah the yeah yeah the unhobblings I think about like uh I'm I'm sure that you don't vibe code a database every time someone builds a website you pull a database off the shelf and I've seen the stats of all the databases that you guys are p like pumping uh customers into.
2:55:07
Um and so yeah, as more of those tools uh roll out and and you get more of that functionality off the shelf and you can pull this from over there, pull that open source project, pull this data, like that feels like that unlocks a lot of a lot of growth.
2:55:21
Uh very >> last question from my side for now.
2:55:22
Uh Gabe in the chat asks, "What are those champagne bottles in the background? What's going on there?"
2:55:30
>> First time someone's asked about this.
2:55:32
Yeah, actually those so that you know this is these are the first enterprise deals when we closed the first cuz like we had never done enterprise.
2:55:37
sure >> sales motions before period like I've never been a sales person.
2:55:41
So you know my chief of staff and I my CTO when we closed our first kind of three major customers this is like probably four or five years ago.
2:55:48
We started with the $15 bottle of cooks for the for the 20k deal we did somewhere. >> Yep. That's good.
2:55:55
>> And then worked our way up.
2:55:55
Um but yeah those are like Yeah.
2:55:57
Those are the uh uh the things I keep on the shelf just >> What's the top tier bottle in the back?
2:56:03
>> I I think the top I think it's a bottle of Dom. I can't >> There you go. That's probably 250 350. >> Yeah. Yeah.
2:56:08
I think someone got it for us as a gift or something.
2:56:09
I can't remember what that was about, but it's that's that's like the uh uh I like never drink alcohol, but it's like that's you know, >> let's hear it for some toing enterprise deals. >> We love it. We love it. Good >> experience.
2:56:21
Well, thanks so much for coming on the show. We'll talk to you soon. Have a good rest. >> Likewise. Thank you. >> See you. >> See you.
2:56:27
>> Up next, we have Matthew from Merlin Labs waiting in the reream waiting room. Let's bring in Matthew.
2:56:31
He has some really big news.
2:56:34
We're going to need a bigger gong.
2:56:36
Jordy Hayes, how you doing?
2:56:39
>> Hey guys, how are you? >> We're great. How are you? Give us the news. Give us the update.
2:56:43
Give us the introduction on you, the company, and then the news today. >> Hey guys, I I'm Matt. I run Merlin.
2:56:47
Uh we are Yeah, we're going to need a bigger gong.
2:56:53
>> We're going to need a bigger gong.
2:56:55
>> No, like I'm seeing you ready with a mallet.
2:56:56
Like I'm I'm here for it.
2:56:56
Like especially at the end of the show. Yeah.
2:57:00
>> Um so I run a I don't know biased, but like a really cool company called Merlin.
2:57:04
uh where we're developing essentially a a pilot just not a human one.
2:57:08
Uh we are deploying that predominantly with US Air Force and a bunch of other partners uh and customers around the around the world.
2:57:14
And we announced this morning that we're uh taking the company public, which is going to be the first time one of these big defense uh startups go public. Let's go. >> Congratulations. >> There we go. >> I love it.
2:57:28
I >> was waiting for that.
2:57:30
Um uh let's uh let's stay on the anatomy of the deal.
2:57:34
What's actually happening um from a mechanics perspective?
2:57:36
What does it mean to take this company public?
2:57:39
Where will you be listed? Do you have a ticker?
2:57:41
Like break down the like what's going on with the deal, what it means?
2:57:44
And then uh and then I I have ton of questions about uh autonomous flight. >> Yeah, 100%.
2:57:49
So we uh we we were looking at a couple different options and particularly for us going public uh sort of efficiently allows us to do a bunch of things.
2:57:58
allows us to go build uh a lot more capital to go uh sort of execute on a couple big contracts that we're working on, but also allows us some currency to do M&A uh which is important for us, especially as we as we uh you know sort of build build up.
2:58:12
>> So, we talked to a lot of folks and uh we're combining with Inflection Point.
2:58:16
uh Inflection Point has done uh two other like really high-profile deals, USA Rare Earth and then uh Intuitive Machines, both of which are trading at multi-billion dollar valuations with hundreds of millions of dollars of cash in the balance sheet. And >> let's go. I love it.
2:58:29
And then another big part of the announcement today is we announced like 120 plus million dollars of committed capital before we even announced the deal. >> Yeah.
2:58:38
So it's effectively like an investment in the company, although it'll happen after the merger, >> 100%. Yeah.
2:58:44
>> So, most companies when they're going public, they raise their pipe between when they announce and when they go public.
2:58:48
Y >> um based off the strength of the deal, we were able to pretty quickly raise about $120 million before the announcement.
2:58:55
TBD, how much more money will take uh uh sort of pre-announced, but super exciting and I think a really important moment for defense tech. >> Got it.
2:59:03
Uh talk to me about um the value of autonomy in flight.
2:59:07
Uh we were talking to Delian earlier about this.
2:59:11
There's a like like the the the model for self-driving cars is very logical.
2:59:15
Something like 70% of the cost of an Uber is the human driving the car.
2:59:18
In the plane context, it's usually something like 7%.
2:59:24
It's much lower because the fuel is very expensive.
2:59:26
You only have a couple pilots.
2:59:28
You have a lot of people in the in the vehicle.
2:59:29
Um so how are you thinking about focusing on like fully autonomous versus uh you know pilot assistance technology?
2:59:36
kind of what's the equivalent of a level five self-driving, level two self-driving?
2:59:42
How do you think about the progression to self-flying planes?
2:59:46
>> Yeah, I mean, well, Delian Delian is captaining his Cessna extreme extremely well, >> but he might be joined by a by a co-pilot >> soon.
2:59:53
No, I I >> Yeah, he's gotten himself in some situations where he probably would have been better to let the co-pilot take over, but he's always made it through, so situation.
3:00:02
All right, so everybody thinks of an airplane with like two pilots, right?
3:00:06
But for if you're like FedEx or like big airplanes, I don't know what deli are, but but like big airplanes. Yeah.
3:00:12
Uh there's anywhere between 13 and 20 FTE pilots per airplane making all in anywhere between half a million and a million dollars a year.
3:00:20
Y >> so if you're able to in the future augment that and go from two crew down to one, that's uh that's you know billions of dollars of savings for you know the airline industry which is which is super important.
3:00:31
But for our military customers, getting that out there enables uh those pilots to go act a little bit more like mission managers uh which is super important particularly given that the Air Force uh is short pilots which enabled our big contract uh that we announced last year sort of a big nine figure contract to go spearhead uh fixed wing autonomy for for certain parts of the Air Force.
3:00:53
What about um what about the role of uh of uh like not necessarily autopilot but like uh just enhancing the safety of of uh airplanes.
3:01:04
I feel like that's uh just >> something that everyone should be excited about.
3:01:08
Get about everyone's seen the the the different flight disasters.
3:01:13
Everyone wants more airplane safety.
3:01:15
They're definitely willing to pay for it.
3:01:16
Um, it seems like there's an like a an opportunity to plug in and I'd love to know about the the different FAA pathways for what it means to actually pilot the plane via a non-deterministic machine learning based AI based system versus having a you know an iPad next to you that's more uh just offering advice and diagnostics and acting as kind of a just bring like a centaur model uh bringing super intelligence along alongside a human in the loop. >> Yeah.
3:01:49
So for for our customers, particularly our DoD customers, like bottom line, most people living today have lived in a world where America has been able to go project air power and like prevent World War II from happening for the vast majority of their adult lives.
3:02:04
>> We're in a world for the first time ever >> where America does not in the West does not have the unilateral ability to go project air power and prevent World War II.
3:02:15
II. So like what we're doing and what we're building here of like getting autonomy out actually in an operational service going from two crew down to one and then we're doing some uncrrewed stuff as well uh with with with a few customers like gets us to that point of like getting out of this like kind of
3:02:32
toy iPad pilot assist kind of like I don't know kind of like a BS solution and actually going out and and really operationalizing autonomy and like getting like millions of hours of operating history so that we can start to go claw back that lead against some of our peer adversaries. >> Yep, makes sense. >> Yep, makes sense.
3:02:52
>> Super important stuff.
3:02:54
>> Come back on as this cycle progresses.
3:02:57
>> Yeah, we're very excited for you.
3:02:58
Congratulations on the news. >> Cool.
3:03:00
We're going to need a bigger gun, guys. >> We'll get it ready. We're working on it.
3:03:04
It's actually >> It's actually in the works.
3:03:06
>> Hopefully, wherever it's manufactured, we'll fly.
3:03:09
>> It's a race against against you guys uh getting out, you know.
3:03:12
Okay, >> we'll do our absolute best.
3:03:16
>> I I'll be back and we'll expect a bigger gone. >> Fantastic.
3:03:18
We'll talk to you soon, Matt. Great to meet you. Cheers. Have a good one. Bye. >> Breaking news.
3:03:22
The Trump administration is said to discuss US taking a stake in Intel.
3:03:27
>> Someone predicted this on the show. Maybe it was Aaron.
3:03:28
Um, >> the Trump administration is in talks with Intel to have the US government potentially take a stake in the bleaguer chipmaker, helping support the company's effort to expand domestic manufacturing.
3:03:41
the deal would help shore up Intel's planned factory in Ohio.
3:03:43
Who uh said the people who asked not to be identified because the deliberations are private.
3:03:49
Of course, the company had once promised to turn the that site into the world's largest chipm facility, though it's been report repeatedly delayed.
3:03:58
>> The plan stems from a meeting this week between President Trump and Intel CEO Lip Bhutan.
3:04:02
The people said the idea is for the US government to pay for the stake and the details are being sorted out.
3:04:07
One of the people said another caution that the planes remain fluid.
3:04:11
So, we'll be interesting to see here if this is taxpayer dollars or this like sovereign slush fund that we're getting from Japan and >> make Nvidia pay for it.
3:04:19
15% of H20s put that money to work by Intel.
3:04:25
>> I don't know if I like that.
3:04:25
I don't know if I like buying Intel.
3:04:26
It feels like we also I don't know.
3:04:28
I mean, the US government does own some companies like Fanny May and Freddy Mack were recently.
3:04:35
They're the the lenders for student loans or um uh all sorts of all sorts of lending that goes on.
3:04:41
Those are uh nationalized and so Trump was talking about IPOing those or spinning those up.
3:04:48
>> Is this uh do you think this is bullish for Leopold? Situation awareness big. Weren't they big Intel? >> Oh yeah, that's huge.
3:04:55
>> Intel's up 20% in the last 5 days. >> Exercise those calls. Let's go.
3:05:00
>> Buco Capital has a I got a post here.
3:05:03
We'll use this to close out the show. Pull this up. >> Team >> please.
3:05:08
>> You just drop it in the chat. >> It's in the chat.
3:05:12
>> Leopold ashen Brener today after being clowned on >> Intel is a horrible.
3:05:23
>> We're going to get demonetized for this.
3:05:25
>> Turn this off before we get demonetized.
3:05:27
You know, we just got to talk over as it's playing so that the AI gets confused and thinks, "Oh, that's not a real song."
3:05:33
>> Anyways, fun hanging out with everyone in the chat.
3:05:35
Thanks for watching, Gabe.
3:05:37
Techno Chief, John Xley, Deepo, Regard, >> so many good usernames. >> Thanks for watching.
3:05:41
We will see you tomorrow. >> I can't wait.
3:05:46
>> Friday episode, the mansion section incoming. >> Really? It's already Friday. Wow. Idea.
3:05:52
Anyway, thank you so much for watching.
3:05:54
Leave us five stars on Apple Podcast, Spotify