0:05
Everyone, welcome back to SemiAnalysis Weekly.
Search this video
0:07
We've got our second guest ever on the program today.
0:10
That's John from Asianometry, top one YouTuber in the world.
0:16
Uh, at least my favorite.
0:16
Doug said it's top two, but we won't hold him to it.
0:20
Uh, this week we're planning to talk about some big news and everybody leaving Google, uh, leaving Gemini hanging out to dry maybe, or just going to work on some of their passion projects now that they've spent their first 35 years in one job and had enough.
0:34
Uh, the Chinese ban on transceivers, Elon pulling in his, uh, one trillion revenue forecast from 2031 to 2030, and maybe some DVDs.
0:44
Uh, guys, welcome to the show. >> Thank you. Thank you for having me.
0:50
This is, uh, it's a weird, you're so professional compared to Transistor Radio. Uh, we've grown up. It's so grown up, John.
1:02
>> All right, I have a burning question for you, John.
1:03
So, uh, one of the times when many people may have seen you talk is on the Dwarkesh podcast where a few years ago there was this like famous clip where you're sitting beside Dylan.
1:12
And you say, "Wait, wait, wait.
1:12
We're assuming GPT-5 is going to be good?
1:18
We've just assumed that?"
1:21
Do you remember this clip?
1:23
>> I do remember this clip. >> Okay.
1:24
That was like a year and a half or two years ago.
1:26
So, now we're sitting in August 2026.
1:28
Do you feel like GPT-5 was good?
1:31
Did it surpass your expectations?
1:35
>> I mean, I think GPT-5 was I think the concept of GPT-5 was, uh, was probably was probably right in the sense that like was it good?
1:46
Five itself probably wasn't that great as compared to others.
1:48
But, you know, I was it would kept going and then I think, uh, when the really ate when the injected code stuff really hit then everything hit the, you know, everything hit the pedal to the metal.
2:01
So, I was very impressed by that.
2:02
So, um you know, but what they said, you know, when I went on the when I went on the pod, when I sat down on the pod, the first thing I thought to myself, I chugged like five cups of coffee and I thought to myself, you know, it doesn't really matter if I'm accurate as long as I'm entertaining. That's what matters.
2:20
>> Hopefully we can aim for that uh today as well.
2:22
Entertaining and uh questionable accuracy. Doug, what do you think?
2:27
So, this was October 2nd, 2024. I just looked it up.
2:30
When uh Dylan and John went on Torcashes podcast and they talked about this stuff.
2:36
Like I kind of agree GPT-5 was a bit of a dud, but 5. 2 and certainly now 5.
2:39
6 is like mind-blowing 2 years later. >> A five 5. 2 was ass, too. I'll be honest. I think 5. 2 is not very good. Um I think 5. 6 is good.
2:53
The new The new base The new base model is good. 5. 6 good. And also soon soon six. >> Yeah. Yeah.
3:02
>> The six apparently is is good.
3:04
>> Doug is coming, right?
3:05
>> Yeah, Doug is coming as they say.
3:05
Doug in the is John, I don't know if you know this, but Doug is the largest spud in the world. >> Really?
3:13
>> Yeah, that's the name That's how they named it.
3:14
So, it's like the spud the spud model. I just like the spud.
3:19
It's very So, the the Doug pre-train is done.
3:23
>> There's also rumors that it's going to be really good at writing and I think that's pretty poetic that uh one of the models that finally really knows how to write some incisive commentary is going to be named Doug. >> That's blazing.
3:39
It's going to It's going to It's going to teach us a masterclass in uh in in writing.
3:45
>> His number I'm looking forward to it. >> Yeah.
3:47
Hopefully it knows it's not this, but it's that.
3:49
>> That's where Doug gets his his The references to like 1940s Taiwanese uh >> Well, you say that, but there's only the only person who's probably better at it than me is John. And John's here.
4:00
So, John's like I'm like, "Oh, yeah, yeah, 1955 semiconductor cycle.
4:07
That doesn't That doesn't exist.
4:08
Semiconductors weren't around then."
4:10
But then John's like, "Yeah, here's what a DVD player is."
4:11
I was like, "What the hell?"
4:15
So, the only person in the world who goes deeper and more random, I think, is John.
4:21
>> I've always I I like like Doug for like, you know, I've I read this really weird book on Japanese history from the 1700s on the Mitsui family, and I'm talk about it, and then Doug's like, "I read that.
4:30
First half of First half was great. Second half was okay."
4:34
Like >> I guess when I got into the modern era >> I was like, "Yeah, I don't really care."
4:38
But yeah, the first half was pretty good. But yeah.
4:39
Well, no, you got that from Jordan, though.
4:41
We all got it from the same source, though.
4:44
>> I had like I picked it up from my uh when Japan ate the rich video.
4:47
But I like I didn't I skimmed through it.
4:50
It wasn't a really good like >> It's a good It's a good It's a good narrative.
4:55
>> It's too No, it wasn't even narrative.
4:57
It's really really boring.
5:00
>> Yeah, so this is Jordan Schneider of China Talk you're talking about.
5:01
I think it's unlikely that anybody subscribed to Something Else's Weekly doesn't know um Asianometry or China Talk.
5:10
But in case they don't, clarifying that.
5:12
Uh Interesting to get the take without Jordan on here, but John, what's your What's your take on the Trump administration administration prepping to ban a bunch of Chinese data center components, including like transceivers, and then possibly banning humanoid robots as well, like the Unitree stuff we were talking about a few weeks ago on here.
5:35
>> You know, on one hand, you know, I'm generally of a I'm I'm generally more hawkish.
5:39
I think we should be banning this stuff.
5:41
You want to drive You want to make sure that you don't have You're not drive You're not uh uh driving revenue for people for for for the Chinese.
5:50
But on the other hand, I mean, they have the expertise and it's not exactly like this stuff is um I mean I mean They're very good They're very quite good at like once they read the the letter of the law, it's pretty good to evade the letter of the law, so you can to meet the letter of the law, so you evade the spirit of the law.
6:09
And I think like >> Yeah.
6:12
>> The real followers would be numb. >> Yeah.
6:14
>> The The real followers.
6:14
Yeah, Thailand Thailand is actually a big uh big area.
6:18
The other thing, too, um is unlike past bans, I think this is an example where InnoLight um is actually better than the West.
6:26
This is like This is like us banning like it's like inverse EVs, right?
6:30
EVs, they didn't have I don't think they will have EVs for some time, but uh when it comes to 1.
6:35
6 T transceivers, um yeah, maybe there's some components in the West, but the entire supply chain is primarily in China.
6:43
So, this ironically is probably more painful for the Western companies to bear because the supply chain's already there. So. >> Yeah.
6:55
I I There's one thing that I I feel like the Chinese are really really good at, it's optics.
6:58
It's I think the I have Maybe it's the rare earth thing.
7:03
Maybe it's the fact that a whole bunch of people study it. I don't know.
7:07
>> [snorts] >> There's >> Yeah, they're also good something They also made the optics in the iPhones like uh Sunny Optical or whatever, right?
7:13
Those are the two um companies that make the lenses.
7:16
>> Yeah, I mean, they're also pretty good at Logan Logan's a Taiwanese company, isn't it? >> Ah, whoops, sorry.
7:21
It's Logan's a Chinese company.
7:21
Sorry sorry sorry sorry sorry.
7:27
>> I mean, they're also pretty good at humanoid robots, drones, solar panels, now EVs, like I think there's a a growing list of real technology where banning the best product in the world is what it takes to ban China from competing in the American market.
7:47
Whereas in the past it was like you're banning cheap knockoff competitors who are trying to bleed their competitors dry.
7:57
Now you're just like kind of >> losing the domestic manufacturing base cuz the main I guess that's that's what I would guess.
8:06
>> I feel like the idea would just be kind of like innovation, you know.
8:10
>> You ban one side, you ban the the import and then you you raise up a domestic competitor.
8:15
So, I would imagine that the Americans have a good number of domestic competitors or if you really want to go 1970 1960 Japan, you kind of force the Chinese to enter a joint venture with the Americans they want to enter that market.
8:29
A 49 45% joint venture and then IP transfer, baby.
8:36
>> You you're you're foreseeing Trump in a light industries coming soon?
8:41
>> You just have to do a 50 a 51% to the Trump family. >> Yeah, there you go.
8:50
>> It's actually really easy in a light.
8:50
If anyone is listening, you know, tip us for the idea. >> Yeah, 2% for us.
8:57
>> But demand is not slowing down.
8:57
So, uh maybe biggest news story from this week was like everybody leaving Google.
9:06
Um Jeff Dean, Sanjay, Oriol, like a lot of uh there's news about Demis stepping down to become the chairman.
9:17
>> One and all, he's stepping up in theory cuz it's up the org structure, but um I mean Demis has been really focused on Isomorphic Labs.
9:26
But some say that's that's this is that was um allegedly according to the investor bros um relatively well uh you know, kind of forecasted or like, you know, communicated beforehand.
9:39
I think there is a real chance that Demis would be could be the next CEO of Google.
9:43
And so, like, I think Demis is, you know, failing upward as fast as possible, you know, that's in the spirit of uh Google.
9:54
>> Does he actually want to be CEO of a company?
9:56
Like, doesn't seem to like the admin stuff, right?
9:59
He looks like a scientist scientist.
10:03
>> He's chief sec- He's chief scientist now. Uh, Jeff Dean's gone. >> Yeah, I know.
10:07
>> Yeah, and he's and he's currently CEO of Isomorphic Labs as if chief scientist wasn't enough.
10:11
Um, I mean, the guy reportedly works insane hours.
10:13
Like, he'll just work a normal day, go home, have dinner, and then come back and just like read papers by himself in the office until 2:00 a. m. every night.
10:23
This is like >> Isn't that how I work? It is how you work. Why is this acceptable?
10:30
Why are you saying this is unusual behavior?
10:34
>> It's Taiwanese behavior, bro.
10:36
>> It's a Taiwanese Singaporean, man.
10:36
So, he's got to he's got to he's got to he's works until he's got to work 18 hours a day, man.
10:42
>> He's got that fire in him. >> Yeah.
10:45
Even if this is just like some, you you know, story to gas him up, you do not hear the same story about Jensen or Zuckerberg or like anybody who fancies themselves a scientist who is also the CEO of a leading company.
11:01
So, I I do think it would be a change from the current day-to-day to become CEO of Alphabet, I guess, and take over for Sundar, but maybe the Okay, the the more important thing here is like it Jeff Dean left.
11:12
Like, the legend, employee number 30, MapReduce, Spanner, Bigtable, TensorFlow, TPUs, like, what, you know, >> He's an infrastructure guy, you know?
11:26
He's just been focused so much on infrastructure.
11:28
>> That's what >> I did that Sorry, I'm not going to go to go into that.
11:31
No, um I do think it's a big deal.
11:35
Uh I definitely think there is a lot of rumors that not like I guess but thing is Jeff Dean's also at every conference, you know, he's like in every conference, man.
11:43
He's like always on the speaking circuit.
11:44
He's always talking about something.
11:47
I feel like Jeff Dean is uh as you know like the Chuck Norris of Google like that's that's his like refer his reputation.
11:53
And I think even if he was or was not working there day-to-day, I feel like you have to be a little sad for Google cuz that's like they just lost Chuck Norris, bro. Chuck Norris just died.
12:06
>> Yeah, this is right after losing John Jumper, right after losing Noam Shazeer, right after losing Dave Silver.
12:09
The prototype seems to be that these guys leave and then they raise money from a bunch of investors including Google Ventures.
12:19
>> And Google Ventures, yeah.
12:20
>> And then go buy a bunch of GPUs and do whatever they want.
12:22
Maybe because they couldn't get compute allocation to pursue the ideas that they wanted to.
12:27
I mean you may be able to say Jeff Dean like was in this senior role and quite a public figure, but I don't think you can say the same for Quoc Le or um Oriol [clears throat] or Sanjay like these were ostensibly the leads of Gemini.
12:46
Like that's I think that's pretty crazy that they all left.
12:52
Um >> [laughter] >> like Well, can I say something though?
12:57
Is this vibe is like from a vibe from a SF vibes perspective, is it surprising?
13:01
Cuz uh you know, the the joke is Gemini hasn't been a frontier model for a long time and now they definitely won't be.
13:09
Um you know, there's uh I guess I don't know.
13:11
I guess you just like lose all the legends and you think how are they going to come back?
13:15
But they do have a lot of compute, but I just um I don't know.
13:20
Everyone at GDM is not AI pilled, right?
13:22
All people who want to do the best AI don't work at Google.
13:26
Uh and I think that that's like an important mindset.
13:28
I think it's part of it too is Google's not a startup, right?
13:31
Google's maybe the largest bureaucratic organization in the world.
13:35
And OpenAI and Anthropic, who are the best in the West, are definitely kind of startups.
13:40
I'm sure they're going to speedrun being a bureaucracy really quickly.
13:44
But, you know, they have one job and one job only.
13:47
It's like A, be the best at AI.
13:49
Versus Google is like, we do ads and search and YouTube and billion services every day.
13:55
And then they're like, okay, well, how do we shove AI into all these different places? But, I don't know.
14:01
I feel I feel like it's poetically poetically it makes a lot of sense to me.
14:07
Um my hottest take, or not even hottest take, it's actually pretty mild.
14:09
Google has an L culture and has never actually made anything inside internally.
14:13
They've only ever acquired all their innovation. Number one.
14:17
And their execution is very is very poor.
14:20
And they were also terrible go-to-market until TK got a GCP.
14:24
But like just seriously, let's do some accounting.
14:26
Everything other than core search has been purchased. YouTube. Uh Uh what you call it?
14:32
AdMob, DoubleClick, AdSense.
14:36
>> I hope you make it by saying Android, yeah. >> Android, yeah. All these guys. >> That's not true. This is not true, man.
14:44
They didn't They didn't acquire Maps.
14:46
They didn't They didn't >> No, they acquired Maps. Waze. They acquired Maps. They acquired Maps.
14:51
>> No, they didn't They didn't acquire Maps.
14:52
They They bought the team that Dude, there's >> whole book about this, man.
14:58
>> They did not acquire Maps.
14:58
They They had no acquisitions after they started Maps.
15:02
>> They effectively acquired Maps.
15:02
There's a whole book about this. >> Okay.
15:06
[gasps] Didn't you just say that this Doug is the guy who has the book very >> They also did not acquire GCP, which was like the crux of your argument that they suddenly have >> They acquired TK, who then made GCP good.
15:18
>> They also, to be clear, they invented half of the concepts of TCP and then they were the last place despite being pretty close to the front you know, showing up at the beginning of the of the line and then out you know, under executing to be last place.
15:30
So, I just think like Google has like the >> TPUs?
15:34
Did they acquire Kubernetes?
15:36
>> No, okay, so I'll give them a TPU. I'll give them a TPU. How's this sound? Give them a TPU.
15:40
Um >> How generous of you. >> Yeah, you're welcome. How how nice of me.
15:45
Um no, but I think the thing is like there are these companies that are like known for their research and they never execute on the >> Bell Labs. >> Yeah, Bell Labs.
15:56
>> Bell Labs is like Google that comes up with all the stuff.
15:58
>> like Bell Labs >> They make a bunch of They make a bunch of science stuff.
16:02
>> No, but they literally make a bunch of science stuff and they literally get it they get uh massively disrupted by it in the end.
16:07
Bro, uh the first uh digital camera was made by Kodak. Kodak died. >> Okay.
16:15
>> Bell Labs invented the transistor and gave it to everyone. >> Exactly, dude.
16:18
It makes so much sense that Google is going to invent the transformer and then get like giga giga screwed as it pertains to like just lose and become a TPU.
16:29
I mean like this is >> It's to lose and be the third most valuable in a company company in the world with a market cap of 4. 3 trillion dollars. Lose to who? Lose to Nvidia?
16:39
>> There's a short term >> AT&T was the biggest company in the world in 1982.
16:43
>> It's a from a from a technology perspective, I think to me there is these moments these important branches in a To be clear, this is not saying that it's a bad stock or bad company.
16:52
They're going to make a crap ton of money, but there's these important branches in a in like an innovative technologies company where they start to become you know, financialized and they say, "Hey, let's do the let's do the way we make money and like stop doing the like pursuing technology."
17:05
You can argue this is like Intel giving up at mobile.
17:10
You can argue when IBM effectively IBM has systemic problems, but honestly IBM's vibes are very similar to Google when they were you know, they're like, "Yeah, yeah, yeah, no, no, we're not going to do all this new stuff.
17:17
We're not going to do PC.
17:18
We're going to focus on mainframe." Right?
17:19
They're going to do what they did and then they will proceed to lose.
17:23
And to be clear, IBM had like effectively 90% market share in 1950.
17:26
Like, there are these points in a company's history and if I I feel like today more than any other point, I think Google has that opportunity where the they're like, okay, we're going to all in on TPUs.
17:38
We're going to go on infrastructure company.
17:42
We're going to do good enough.
17:42
We're going to have a good enough product on the transformer.
17:47
But they're probably not going to stay for the next and the next and the next.
17:50
And then they'll quietly bow out.
17:52
And then when we look back in 5, 10, 20 years and and and look like, "Oh, what was the moment that they like where things like where things started to to change?" It's like right now.
18:00
But to be clear, I think that it's really good for the stock in the short run and medium run, honestly. So.
18:08
>> The government should basically anoint Google's monopoly in a search.
18:11
And maybe and but force them to outsource to to license all of their science that they've been doing inside to every startup, just like they did with AT&T.
18:24
That would that would supercharge the the the the startup network.
18:29
>> You think there's a new Bell Labs coming?
18:32
>> They're effectively Bell Labs.
18:34
>> They're They're effectively Bell Labs, yeah.
18:36
>> Big, profitable and then then they can't even monetize their own crap. >> Yeah, yeah, exactly.
18:41
Like it's it's like the perfect example where it's like, yeah, yeah, they have all this cool stuff.
18:45
My favorite is killed by Google. If you go www. killedbygoogle.
18:47
com, it's all these things and they're like, you can argue many of the companies that they had or they killed look like very successful valuable companies today that are work like they literally get killed by Google.
18:59
They go leave and they start it up again as a public company as like an independent company and it goes well.
19:05
Um and I just think that search uh search was a giant monopoly that gave them the ability to just do Yolo random stuff and positive R&D, but they're never going to make money out of that. It's kind of based.
19:20
>> I have another I have a question.
19:22
Like with these big guys leaving, does it really matter?
19:27
Like is it if an NBA team if it's like if some guy if some if LeBron leaves the Lakers and then people are like, "Oh, does the like are the Lakers over?" Does that matter? I mean >> To win?
19:40
>> They've They've kind of done their stuff, right?
19:43
>> Yeah, I I I think that that's kind of agreed.
19:44
Maybe maybe the answer is it's like the dream team is breaking up as they all turn 38, you know? That makes sense to me.
19:51
You're like, "Yeah, their their best days are behind them.
19:53
Now they're going to go do one last glory tour.
19:56
Maybe they're going to go play baseball or something and then go back." Who knows, you know?
20:00
I That That actually makes sense to me, but I still think you should in your heart of hearts still feel really sad about And if you're like a psychologist, you know?
20:10
>> Yeah, but that like like TSMC, right?
20:13
TSMC's R&D structure, they have incredibly federated like incredibly split up teams, especially within Taiwan and Hsinchu.
20:23
And something they always emphasize to me is that they work as a team.
20:25
There's no There's no single ego.
20:29
Which means that one guy leaves or a couple guy leaves, it's not a big deal.
20:33
So, they they tell me when big people leave to go do something else, you should always disregard that.
20:40
Not disregard, but like don't take it as for larger than what it is because in the end R&D, at least in semiconductor world, is a team effort.
20:50
Everyone has to put a crap load of effort to rubbing down the corners and no one person knows it all.
20:57
>> I agree in semiconductors specifically, but I think I disagree in the Jeff Dean case specifically cuz cuz it's such a strong counter example.
21:09
Uh as Jordan says, uh what were all the things that he he's made so much stuff.
21:13
Where it's like it's so rare you have one dude who's so cracked and is like literally Oh my It's like it's like a I mean, he's not von Neumann, but it's like Oh man, it's crazy, and he also did this, right?
21:22
He's on all these different software, hardware, uh coding infrastructure, like all these different things across the entire spa stack.
21:31
He has his fingerprint on all of them.
21:33
It's really rare to lose um a talent like he's revered for a reason. Uh but yeah, we'll see.
21:39
Maybe Maybe it doesn't start on von Neumann, especially with like a lot of his stuff collabs.
21:44
He does a lot of collabs.
21:48
He does a lot of Hey, maybe that's Jeff Dean. He's just a collab guy. >> That that is. >> He's an idea guy.
21:51
He put He does He collabs with the computer thing.
21:55
>> Uh like I the von Neumann model drives me crazy.
21:58
Like if you read into this the the the science of it >> I'm I'm I'm going to be honest with you. It's very vibey.
22:03
He just showed up and he's like, "Yeah, you should make it like this."
22:06
and he peaced the [ __ ] out.
22:07
No, to be clear, I I mean, you got to respect von.
22:09
You got to put some heat on von Neumann's name. Come on, man. >> All right, I do. I do.
22:13
>> But let's not Let's not glaze him. Let's not glaze him.
22:16
Let's not Let's not glaze him the same way.
22:17
>> I think there's something to be said for somebody who can keep >> somebody who can keep Yeah.
22:25
Whether it's von Neumann or Jeff Dean, I think there's something to be said for the talents required and really the experience required to be able to fit all of the different components of the system in your head and reason about them, even if the work that you do is
22:38
pretty quick and the insight you give is pretty small in the grand scheme of a big project, nobody else can really provide the exact same value that you can if they don't have all of those little pieces of experience that you stock up, which I think is what people view there. On the model side, though, I think if
22:55
On the model side, though, I think if you view individual people leaving as like not representative of the next version of Gemini being good or bad or not not like a signal that that kind of goes against the entire ethos of all of these labs that are being founded right now and kind of implies that like every
23:16
single lab that's founded by like the guy that just left um is not going to be successful cuz it implies that like whoever's got the biggest engine with the most compute and the most data is just going to produce the best models and there's no more discoveries to be had which uh >> Maybe it's true. Could be true. I mean, didn't a bunch of Could be true.
23:35
I mean, didn't a bunch of people leave OpenAI like after '22 '23?
23:42
OpenAI kind of they staggered a bit but they got themselves back up.
23:45
>> No, but that's different.
23:45
No, that's different, dude. That's different.
23:49
That's like that's like the Protestant church, okay?
23:50
When when we go back in the history of AI, that's going to be like, "Dude, the Catholic church has lost the sauce.
23:57
They're not even They don't even really believe in the same god we do and they're going to [ __ ] They're going to nail the thing that like Dario is the guy who nailed his freaking >> He's Martin Luther.
24:08
>> Exactly, he's Martin Luther, dude.
24:08
You tar That That is a take That is a take, dude. >> We're recording now. >> a new religion.
24:15
>> [laughter] >> Exactly. Yes, he did.
24:18
And he's like, "No, no, no, no, no."
24:20
And then that's and that's why they're that's why they have uh Yeah, that's that's that's Anthropic and OpenAI.
24:26
>> So soon you're going to You're going to be like somebody's going to look over your shoulder and see if you're using Codex or Claude and it's going to be like somebody in Northern Ireland being like, "Are you a Catholic or Protestant?"
24:37
>> [laughter] >> Which Uh I'm an atheist.
24:42
Well, are you a Catholic atheist or a Protestant atheist?
24:48
>> [laughter] >> That's one of my >> [gasps] >> favorite Halloween jokes.
24:52
>> They'll have uh They'll they'll be, you know, bombing certain data centers with others, you know?
24:57
It'll be religious war, you know?
25:00
A 30-year religious war be >> Sounds good.
25:04
>> Well, I mean the the models are compete they're like currently competing each other with each other to to show off how many >> Felonies.
25:14
>> examples of their models breaking out of containment they can >> [laughter] >> Felonies. The felony chart.
25:19
Dude, I'm I will admit some of them, if they're real, do does scare me.
25:28
Do my my complete complete rando ass take that I feel like may or may not become true is I feel like the internet will I think the internet will be viewed as a toxic toxic as in like it's just going to essentially affect your computer and things will become like you can argue if we have enough local LLMs, like not local, but like why do you need to talk to some dude across the world?
25:52
Maybe talk to some dude across the world, but like referencing information, why do you need to go to a website?
25:56
You your GB200 in your neighborhood has all the thing you need has all the information you need anyways because if you actually go on the internet, all the malicious AIs will just destroy your will break your computer instantly and steal all your money.
26:11
>> Why Why do these companies not keep track on their AIs?
26:13
Like isn't it using a computer?
26:16
Like don't they own the computer?
26:17
Like why why are they why can't they keep track of It's like so like their dog is pooping on my yard.
26:22
Why do why do you not keep track of your dog?
26:25
>> Because they can't This is it's not a dog.
26:28
It's a the yard is so big and there's no fences.
26:33
And in fact, it's actually 100 acres and there's secret tunnels everywhere because we've just been building it for God knows how long.
26:41
And this dog is super intelligent and knows everywhere every little I mean, the thing that's really crazy I think is like in the same way we talk about how you if you say like too big for anyone person to understand how to make a chip, modern software is too big for anyone person to read and understand all the code.
26:58
And then also where you understand all the interdependencies, right?
27:00
Versus it's like, "Hey, all goes in one context window and it can read every single word all at the same time."
27:06
That it might as well be the effective like, "Oh, how do I How do I understand all the context of a very specific, extremely abstract language and then be able to break and like listen to it make uses or like, "Oh, I could just do this instead of this." Right?
27:20
Like, I think it's like the perfect use case and I think they're I don't know.
27:25
I just think it will everything's not going to be safe.
27:29
Like, one day And it starts with like Python and whatever and like stuff that's like super high-level languages, but there's also super low-level exploits, too.
27:36
Like, memory slipping out of the CPU.
27:38
Like, why Like, these are really hardcore exploits that are superhuman and I think we'll just continue to see superhuman level breakage of the system forever.
27:47
So, it's going to be it's going to be an interesting decade to to say the least.
27:54
You know what's the only thing to fix it? A great firewall.
27:55
I think we should I think America should have erected a massive firewall.
28:01
And um they've provided the technology originally.
28:04
They might as well just use it. >> Yeah.
28:09
Yeah, maybe one on the internet and maybe one physical one as well just in case the robots are gun going.
28:12
Like, a big a great wall.
28:16
It's okay, the humanoid robots can't swim more than 10 minutes, I think.
28:22
>> So, you're telling me it's just going to be a great moat?
28:26
It's going to be one great one-mile trench filled with water. >> Sure, man. Yeah.
28:33
>> It's back to the, you know, we're going we're going back to the medieval times. A big moat. Big moat. Big moat. >> Okay.
28:41
Okay, what else you got in your bag of questions?
28:43
>> [laughter] >> Well, okay, the other one is uh Let's uh Let's let's get a nice segue.
28:49
So, with all this demand from all this new AI, it's going to mean more chips, including Elon breaking ground on terrafab, and then telling investors in the first combined SpaceX uh earnings call that they're going to move their one gigawatt forecast to 10 gigawatts by the end of 2027, and 20 gigawatts after that, which means that they are now, he believes, bringing in their $1 trillion forecast from 2031 to 2030.
29:21
Um That's the That's the abundance case, man.
29:26
Um Okay, let's go back to the very beginning of this.
29:29
John, I talked about you going on Dwarkesh with Dylan in October 2024.
29:35
Uh that was before like the hyperscalers decided to pour in a trillion dollars of capex, and we've already got some pretty incredible models in two years since then. So, >> Mhm.
29:47
>> what happens when 1 trillion becomes like more than a trillion in the next 5 years?
29:54
Like 10 trillion in the next 5 years? >> W- cuz of Elon?
29:58
You're saying Elon hits his numbers, and then suddenly we're all awash in compute?
30:03
Is that what you're trying to say?
30:06
>> It's going to be Elon, but it's also going to be all the other competitors, is you know, Microsoft, Amazon, Google, Meta.
30:11
They're like all taking their free cash flow to zero right now to buy chips, which is uh a fascinating thing to live through.
30:23
>> I mean, I, at that time, in '24, could not believe it.
30:28
I I I I don't even I I still don't believe it now.
30:32
I don't live in the states, so I I don't see it here.
30:33
We Taiwan doesn't really build data centers.
30:37
They need all the power for fabs.
30:38
So, >> Do Do people Do people even use AI in Taiwan?
30:44
>> They use like ChatGPT to like translate letters and stuff.
30:48
I think they watch a lot of slop.
30:50
But they don't I don't think they use ChatGPT that much there. They're very behind.
30:57
>> Do you use coding agents?
30:59
>> Yeah, I code my own video editor. >> Really?
31:02
Can you tell us about this?
31:04
>> Yeah, when did you do that?
31:06
>> I think a couple months ago I would needed a I didn't like like my work my workflow is very specific like a lot of keyboard shortcuts.
31:15
And I figured you know, I used to work in banking.
31:17
So they used to do they used to remove they used to like take away your mouse and like you have to like do the whole thing.
31:23
So I didn't like my video editor not having keyboard shortcuts and I was like I should write my own. Can I do that? And turns out I did.
31:32
My my little video editor is around 40 megabytes and it's like highly optimized to what I do.
31:38
Like a lot of stuff like cuz if I grab a picture you I'll right click it and it immediately jumps into the into the image into the video editor.
31:47
Um just a lot of small stuff that I would I never thought was possible with Adobe Premiere or something like that or even iMovie.
31:58
And it's it's it's wonderful.
31:59
I highly recommend that people roll their own software.
32:02
Like it's like it's like the hipster version of people making their own furniture.
32:08
If you're a if you're a true tech hipster wearing the with a beard and the plaid shirt, you need to you need to roll your own software.
32:16
I'm going to make my own text editor soon. I feel like I should. Yeah.
32:21
>> Like the minute I wrote >> Honestly words are so bad, why not? What's to lose?
32:25
There's literally nothing to lose. There's nothing to lose.
32:28
>> Can you explain how you built it? Like when did you start? What tool did you use? >> I tried four times.
32:33
First I was thought to myself like I tried four attempts like and I think it was a big deal where I tried to like I asked Claude to quiz me on what this video editor work did and what it how it worked cuz the workflow's a bit specific.
32:49
You start with an audio file.
32:51
So, I thought to myself, well, can I do it this way, do it this way?
32:55
And I tried the first three times and I really didn't like it.
32:59
And then finally, I stripped it down to enough I had to get another video out.
33:04
And I thought to myself, let's give myself a deadline.
33:07
And I finally stripped it down to a point where I said, just do this one thing really well.
33:13
And ended up having Claude have like this 50 page like it was a massive spec.
33:20
And I handed it to Kodak Kodak's I think it was 5.
33:22
3 or something at that time.
33:24
And I said, just make this very, very, very simple structure.
33:29
And um And it barely was good enough that one thing that I could like keep working on it a little bit. That's how I did.
33:41
I think it was >> Has it gotten better with subsequent AI because I do feel like there are things that were outside of my control or outside of the ability of what I was able to do with 4.
33:49
5 that is doable with Fable that is maybe doable with 5. 6.
33:53
And I think it's like weird cuz what will happen is every once in a while be like, wait, I should try to redo this project.
34:01
Like I I redid a lot of projects with Fable and then when I redid I was like, oh, actually it's actually good now. >> Yeah.
34:08
>> I mean, there's a lot of times where I would be like triggering Opus and something like that.
34:12
And I was just saying, improve this. Can you improve this?
34:14
And then I would have to go on to my computer to troubleshoot it and check it and it's gone a lot better.
34:23
Like I have now situations there's a there's a library I use called Easy Cheese and like it really it grills me.
34:30
They use a lot of cheese puns. That grills me.
34:34
For like 50 50 words or like 50 questions.
34:37
Like The questions are longer than the actual coding.
34:40
And it creates this titanic spec.
34:44
And um but it one hits every feature. I love it. Um No idea how it works.
34:48
Um so but it's ridiculously good.
34:51
And the fact that I can roll entire There's one feature I added.
34:57
Like recently I added a Can you when you add an image, add it to the end of the last image, but fill the like text block cuz it has a subtitle.
35:09
Fit it fill the text block cuz the text block could be different sizes.
35:12
And Opus and Cla- Opus couldn't do it for the longest time.
35:17
But I ended up asking um 5. 6.
35:19
And then I used Easy Cheese and I just the questions I kept answering I asked answered like 25 questions. I'm not even joking.
35:27
Like I went to bed and I was still asking me questions.
35:32
But after finally did it it one hit the whole one shot of the whole thing. I was just mind-blown. It's crazy. It's crazy.
35:42
>> How much uh How much do you spend?
35:44
Do you have Do you pay for token or do you have the monthly plan?
35:48
>> I have a Codex and then a GPT Pro and a $20 cloud.
35:54
>> So you've done all this with just under a thousand bucks or something? Yeah.
35:58
Like a couple hundred bucks. Yeah. >> Yeah.
36:01
And I have my own video editor.
36:03
And I've integrated Wikimedia Commons search into it.
36:05
I've integrated a video search into it.
36:08
Like it drops it in with the correct attributions.
36:12
It's ridiculously It's I don't even know how. Yeah.
36:17
It's It hasn't made me faster.
36:19
It just makes the videos longer, unfortunately.
36:21
And now I'm >> I mean I like them.
36:26
Um so, how many how many You did this you said it's like starting in April, so you've done like a bunch of videos with this new editor. And no >> Yeah.
36:38
And I don't think people would notice. No one's noticed.
36:41
And then recently I've like had GPT >> I want to start I'm going to start commenting on the videos.
36:45
I can tell this is AI sloth now. >> You're hilarious. I still do it. I still do the work.
36:51
It's just my own tool, man. No, I'm just kidding. I'm just kidding, dude. No, it's okay.
36:57
I have a ton of tools like that, too.
36:58
Like you have a tool that you use to >> Yeah. Really?
37:01
Yeah, pretty much you have all the hockey.
37:03
No, there's so many things like that where it's like this is the exact way I think about this set of problems and I want to make sure it's done exactly like this and then I'll work on it for forever and then it's like, "Okay, now this skill, press enter, go." Mhm. Mhm.
37:18
And so like um I'm it's it's fun. It's fun.
37:21
I don't know how we got into this, but I also have a bunch of other tools.
37:25
I I do audio post-processing with a little app um that I made myself that I coded as well.
37:31
I have no idea how it works, but it works and it even gives me a nice waveform. >> Who needs to know?
37:37
>> Do I need to know how it works?
37:37
Like the audio comes out better. >> Who needs to know? I don't need to know. You don't need to know.
37:48
>> I mean, my dog that that that video that little post app might be hacking the Chinese you know, Chinese government uh website right now, but like who knows?
37:56
I mean, like it's >> Code? That's for the birds.
38:05
>> So, okay, so what's left on the what's left on the question list question, man?
38:10
>> I don't I don't >> [laughter] >> I don't have any more.
38:11
I was going to ask John about GPT take.
38:15
Yeah, yeah, what's your GPT take?
38:17
>> I think he's going to start with memory. Yeah, 100%.
38:21
>> Yeah, I think he said do that actually, yeah.
38:23
>> You should choose the highest margin.
38:25
>> Why do you think that?
38:27
Well, I'm just reading his latest that that one he's just he's just on the SpaceX earnings call he's just talking about memory all the time.
38:35
And [clears throat] like and if you in the in Taiwan one thing I definitely do know is that memory is crimp the lack of memory is crimping every product out here in Taiwan that they're trying to make and they can't get enough of it.
38:47
And like Apple like Apple has a bunch of N2 chips that they're just sitting in the in in a locker somewhere because they can't get memory to put into their phones.
38:58
My friend Tim Copan reported that.
38:58
So like um it's going going crazy.
39:04
Um fastest way to to revenue is to just start making memory and expand it there.
39:09
All the all the stacks are open.
39:11
>> Apple Apple tried to negotiate with CXMT and got denied. I think that was Huawei. Was that Huawei?
39:18
>> No no it no it was CXMT.
39:20
>> It was Apple with CXMT.
39:22
They were trying to be like oh give us we'll give you a giant order and you'll give us a discount.
39:25
And CXMT Chad mogged them and was like why would we give you a discount?
39:32
It's just lower than the going rate of memory and then Apple's like well shoot.
39:37
They're like don't you want to work with us cuz we're Apple and CXMT's like who are you?
39:41
Everyone wants my memory so.
39:48
>> So I guess Elon's going to you need to you need to hit that one trillion revenue within a couple of three years so start making memory.
39:55
They'll contribute to it. >> Be easy. I'm excited.
40:01
I'm excited to see him smoke a cigar in the Terafactory.
40:06
And make a hundred billion dollars of revenue next year because that's the ramp that needs to happen in order for Terafactory to happen. Um yeah, I'm excited. It's going to be cool.
40:15
>> Anything under a hundred billion at the end of Q4 27 is a miss. Hey, doc. >> I would say so, yeah.
40:25
I'm I'm being conservative, you know, there's that's just the exit rate, too, you know, just to get to the next >> Does that Does that include or not include Tesla after they merge into Tesla?
40:36
>> Uh >> No, no, that's just Terafab alone, Rez.
40:41
>> I mean, I don't think I think Tesla's a drop in the bucket if they need to go to 100 billion next year, too.
40:50
>> Yeah, just who's even counting, honestly? Who even cares? I'm not even counting. Who's counting?
40:56
SpaceX investors aren't counting. It doesn't matter.
40:59
1 trillion, 2030, 2040, who's who's I think Okay, man, it it's I hate this cuz I really do Um everyone says this and I really do believe it is Elon makes the late No, the impossible late.
41:11
So, that means that means that he's going to get a trillion dollars of chip revenue in 2040.
41:19
>> [laughter] >> Yeah, okay. >> It'd be pretty good.
41:21
Some would say that's pretty good. >> Sure, yeah. Hey, power to him, man.
41:26
I'm excited for more memory.
41:29
>> Actually Actually, John, I was going to say I feel like I feel like that would be the best call that could possibly be asked on the TSMC earnings call.
41:36
Cuz they they hate the analyst asking dumb questions.
41:40
You can be like, "Well, have you heard of an American company that's guiding to a trillion dollars of revenue?
41:48
What do you think about the competitive prospects of a new entrant in the foundry market?"
41:53
And they'll be like, "Please, shut the [ __ ] up."
41:56
They would They would really That would probably be the most triggering question impossible, but I would love to hear it be asked.
42:02
>> Yeah, and the guy will never come back again. >> Yeah.
42:05
He'll be kicked He'll They'll send him to to East Taiwan.
42:10
>> Yeah, he also He also had nice tweet where he said SpaceX chooses to install Nvidia GPUs because they are the best. Pretty matter-of-fact.
42:23
Um pretty matter-of-fact.
42:23
Yeah, and then and then a bunch of other people saying Jeff Dean left Google because he didn't like TPUs anymore and he wanted to use GPUs so that which is the trend for everybody else leaving.
42:35
It's like Dave Silver leaves, raises a billion-dollar seed round and then spends the entire thing on Vera Rubin chips which they have a nice like promo post with Google Cloud about.
42:50
Um >> Why do you think that is?
42:52
I thought I thought TPUs were good for pre-trains.
42:58
>> Yeah, I think TPUs are great, but um I think that the support you can get outside of Google on the software side >> is zero.
43:04
Well, see that's what Fables for.
43:06
That's what Fables for, bro.
43:09
You going to roll your own software.
43:11
You're going to roll your own GCP sales rep.
43:13
You can roll your own console.
43:15
>> Fables figure out Palace and TP Wax and everything. >> Exactly, yeah.
43:19
>> You can make it uh downgraded to Opus 5 which is intentionally uh >> screwing you over.
43:29
>> Just not trained on the TPU documentation because that would make you a competitor to Anthropic who uses all the TPUs.
43:36
>> [laughter] >> So, they're not going to let everybody else figure out Palace kernels because they got to keep that for themselves. >> What's I believe it.
43:45
That that sounds like something the Protestants would do.
43:49
>> [laughter] >> Protestants. Love it. All right, everybody.
43:53
Thanks so much for tuning in to this latest episode of Semi Analysis Weekly with uh John from Semi Analysis Weekly.
43:59
We covered the gamut and we appreciate everybody joining and listening to us uh talk [ __ ]