0:00
Wait, what?
Search this video
0:00
You're adjusting the Nvidia logo to make sure it's on display? >> Yeah, man.
0:04
When you have GPUs at home, you know, you got to show that to the world. >> Actually, a GPU.
0:09
>> It's a It's a decommissioned GPU.
0:09
Uh you have like all the heat sink, all the heat sink, but there's no GPU.
0:13
If you remove all of that, you'll see there's no GPU.
0:17
>> You're going to start running some local models to stop this token burn that you've been jacking up or what?
0:22
[laughter] >> Got to cut costs, man.
0:22
But they keep telling me that it's cheaper on the cloud, you know?
0:25
So [laughter] I'm like, "Okay, my electricity bill can't handle it."
0:31
>> Jeremy Rake, we're going to talk about SpaceX doing 10 gawatt in 27. You guys ready? >> Uh, welcome back. Something else Weekly. We got Jeremy and Rick.
0:42
Like I said, we're going to talk about the article we put out that got a lot of traction.
0:44
Uh, SpaceX 10 gawatt 2027, why it's real will drive $300 billion of ARR for SpaceX and why Microsoft will be the largest offtaker.
0:54
We're going to try and talk through some of Elon's statements on the earnings call.
0:57
Talk through the economics, how we get that 100 million per megawatt per year that they're going to sell this stuff at.
1:04
Um, when you pass through the tokens, Microsoft as a potential customer, how they pay for it, and how they get enough chips and people. So, guys, yeah, welcome. Excited to dig in. >> Yeah, let's go.
1:19
>> Yeah, thanks for having us, Jordan.
1:19
All right, Jeremy, you're first author here.
1:23
So Elon's statements during the earnings call conservatively, what does what does this mean?
1:27
What's the takeaway when he has SpaceX first ever earnings call and says that he's got these gigawatt ambitions?
1:35
>> Look, I think it it all it all really starts from the economics.
1:37
Like that's really the key thing.
1:38
Um when like what we've observed over the course of 2026 is gross margins for these labs just kept going up.
1:46
Um and that has been the key driver of their ARR acceleration, right?
1:51
right? And if you look at them combined today that is anthropic and open AI they're both adding combined over20 billion of AR per month actually close to 30 now u you know so but by itself that that drives uh AI revenue adding close to 400 billion you know uh per year um and they're doing that by
2:10
increasing their gross margins and increasing their gross margins essentially that just means uh for any given amount of compute which typically is roughly affects cost like we've seen you know 12 13 14 million bucks meat here from the likes of Cororeef uh increasing their gross margin just means that revenue per watt goes up. Um and so
2:25
that revenue per watt goes up. Um and so for us and that's where you know that's what I want to flip it to you is for us we've done a lot of work on trying to understand like what is the actual revenue per megawatt and obviously we have inference x and in this article
2:37
what what we put out is that we think like um they can fairly easily today uh today on API that is open anthropic make a $100 million per megawatt per year right and just tying it back to like why is Elon trying to build so many gigawatts so fast well it's because no one else is doing it so fast no one else is really banking. No one else is really banking on the
2:59
No one else is really banking on the opportunity of like, okay, for any mewatt that they have, they can make so much money.
3:05
And so, he was sort of the first to realize, hey, man, if you want this free month from now, you want 300 megawws right now.
3:11
All right, pay pay me 50 bucks per you're going to make 50% gross margin, right?
3:15
[laughter] And so essentially, you just want to replicate that playbook and do that at much larger scale.
3:21
And if the economics are not that good, everything downstream is is much easier, right?
3:25
So the the the the first question, you know, we should ask ourselves is is 100 million dollars per megawatt per year realistic for open anthropic on API uh today right so I don't know what do you think of that Jordan?
3:43
Yeah, I mean I think it is real.
3:43
So it's definitely realistic.
3:46
Um we we we've dug into this in great detail.
3:50
Let me share this chart so that we can actually like explain the details.
3:53
But um yeah, when we talk through uh maybe the different ways in which GPUs transact, you can start at the beginning like you said 12 or 13 billion per gigawatt which is 12 or $13 million per megawatt.
4:07
Um, and then that's like the 5-year average infrastructure as a service price across Yeah.
4:16
Corewave, Oracle, Nebus, long-term close to self-build pricing for these guys.
4:20
Like maybe they're making double- digit margins on these things, but it's it's not massive.
4:24
And where things change is when you sell for a premium because either you're getting on demand, which is the green column here, the B300's, or you're doing these deals where SpaceX is selling their existing compute to somebody because they can turn on so much of it immediately.
4:41
And like you said, that's the reason you can transact at such a significant premium.
4:45
and specifically this Google deal at like the equivalent of 14 bucks an hour is a significant premium over um the average which might land around three bucks an hour for a GB300 right now. Right.
5:01
>> I would I would just like super quickly add that I think an amazing like term in the contract that they have is the 90-day cancellation policy because for a Google for Microsoft for an anthropic they have zero risk.
5:11
They can cancel this if they realize the economics are not working anymore.
5:14
Um anyway, so it just makes it way easier.
5:16
It's really like emergency uh megawatts.
5:17
You want them right now big uh easy to cancel, you know, that's the price.
5:22
So I think the price is justified because that service is unique in the world. >> Yeah.
5:27
And so what is the service?
5:27
How do they demand these margins?
5:29
It's really selling tokens at the inference API costs which as we model it and you look specifically from two angles.
5:35
Both the inference X data that we have where we actually run real workloads on the latest chips.
5:42
This is GB200, GB300 with the latest optimizations from VLM, SG Lang, uh, TRT LLM from Nvidia, um, Mory and other stuff from AMD, whatever chip you're trying to use.
5:55
The point is that uh, we get that real data from open source models and increasingly these models are approaching the size of the frontier models.
6:05
K3 from Kimmy is a three trillion parameter model.
6:08
three trillion parameter model. Deep Seek is using sparse attention and all these I mean Kimmy's using all of their sparse attention approaches too but Kimmy linear but the um you know the point is that these are well over a trillion parameters total that takes up all the memory space they have million context windows which is the same as the frontier models that blows up your KV cache and they have all these
6:29
optimizations to do KV offloading and so you can you can get a proxy or a lower bound let's say of what a 3 trillion or a two trillion model is going to perform at assume that The frontiers are bigger because they're higher performance in
6:43
terms of total parameters and active parameters because they have better GPUs that they can serve these things on compared to the open source models which are typically Chinese developed on the Chinese SKUs. So this is you know not
6:52
So this is you know not the latest and greatest from Nvidia or they're increasingly now being developed on you know Huawei chips and and others from China.
6:59
So, so that's a lower bound of performance is our public data on inference X and then the upper bound is really uh our simulator which takes you know op by op.
7:11
know op by op. So these are like the gems, the um collectives, the like all of the actual operations simulated on uh the current hardware where we test the actual performance of the collectives and the gems and and stuff on the real
7:28
hardware and then just, you know, trace it through what we think is is the shape of the frontier model and then forecast this forward for what we're going to see on Reuben when we see increases in the flops, the memory bandwidth, the networking, you know, bandwidth. Um, all
7:42
Um, all of this is, you know, the power consumption, all of this is going to change when people start deploying Vera Rubin next year, which is going to be all of SpaceX capacity next year.
7:54
They're not going to be de, you know, deploying a bunch of GB300's.
7:55
They're going to be deploying the latest and greatest.
7:59
And so, you're going to see a performance increase from the GPUs on a relative like per megawatt basis.
8:03
That's going to allow people to produce more tokens.
8:07
And like you said, we think they're going to be able to turn on more of these than anybody else in the fastest amount of time.
8:13
Um, so the forecast to go from SpaceX and Google signing a deal that, you know, gives them GPUs at roughly $48 billion per gawatt and then saying there's going to be a 50% gross margin to get to 100 billion.
8:33
You know, you you'd expect somebody like Google to make a financially sound decision to do that.
8:36
they've got to see returns on that, you know, capital that they're investing in those GPUs, let alone the fact that they, you know, the the reporting, the leaked financials that we're seeing from the labs right now is showing 85% gross margins and and that matches what we are seeing with our simulator and our inference X data, which would imply, you know, around 100 billion of selling costs at a cost of around $15 billion per gigawatt, right?
9:04
which which matches this chart that's on screen there in the in the blue.
9:07
>> And that 85 being a blend which includes older or less performing GPUs, not just the GB300.
9:12
So you could argue 100 is actually conservative for GB300 or for VR in fact. Um yeah.
9:20
>> Yeah, I'll scroll down and show another chart here.
9:22
Um you know there there is like really significant differences even as we go from GB200 to GB300.
9:26
the increase in memory capacity, memory bandwidth and flops like FP4 flops on GB300 was like a this is a minor change for Nvidia in the architecture.
9:37
Um and you can see it in revenue per megawatt 73. 4 million to 99.
9:42
7 million is what we forecast for those um those chips on Fable 5 like our fake uh Fable 5 assumptions for the architecture.
9:52
And uh I yeah this is telling man like the the the fact that uh they're selling tokens for such a premium on the cost of the compute just implies that they will buy any compute that they can get their hands on um not just to serve these tokens but also to train the models.
10:16
Like you know selling tokens at such a premium means that you have more money to spend on compute. It's a cycle. Yeah.
10:24
Yeah. Yeah, and I think like um uh the thing is for especially for these AI labs um they obviously have to make decisions on like what kind of risk am I ready to take and Dario has talked about this extensively like I don't want to put my company bankrupt so I have to plan for a certain amount of of compute
10:41
and and what all has happened like time and time again is he under under forecasts um and that makes sense it makes sense to not put his company at risk and so the thing is that these guys know very well that the fuel the core of their business is training because that that's what generates future revenue growth. And so what ends up happening is
10:56
And so what ends up happening is that you have this uh this core compute that is not enough to support the revenue growth.
11:01
And so they end up having to pay for whatever is available spots uh on demand like right now at a premium uh effectively, right?
11:07
And so maybe we're we're going to talk about Microsoft later.
11:11
I don't want to open it to rake on like can they actually build the 10 gigawatts, but like the the point is hey there's a hu there's a huge gap because you cannot build data centers fast.
11:22
So if you have a way to have access to a data center like right now is gigantic, that is, you know, amazing.
11:29
And so uh if if if you can make a 100 mil per megawatt per year, then there's no reason why your underlying provider can't charge you, you know, 50 million bucks per megawatt year, right? And perhaps even more.
11:40
Like maybe we're being conservative actually with that pricing. Uh we'll see.
11:42
But I think it's actually fair to say 50 million bucks a megawatt.
11:46
Um and so it all comes down to like can that company um build this data centers? Can they finance them?
11:54
Financing being extremely important because that is the the core reason why we don't have enough data centers right now is that the lead time is everyone in the industry especially the non-hypers scalers all the third parties.
12:02
Um they need the capital up front to then start making the the orders to like you know switch care supplier, cooling suppliers, build the data centers.
12:10
Um they want to build them well.
12:12
they do commissioning and whatnot.
12:13
And so the data centers are only available generally 12 months from now and much more commonly 18 months from now.
12:18
Um and so we keep being in that shortage where demand grows faster than supply.
12:22
And so you know we're always short.
12:24
So there's really a need for someone to take kind of that speculative risk to some extent accept that hey maybe you're going to get it wrong and be underutilized and whatnot but if you get it right like you get that that you know that premium.
12:35
So that's what SpaceX does.
12:37
They feel a major gap in the market.
12:39
major gap in the market. they've done it successfully kind of luckily to some extent because that wasn't the plan and so now the the the I think it I think it shouldn't be a question of whether they can sell at this rate uh given the economics that we see on the market today the real question frankly for
12:54
SpaceX you know whatever investors or or whatever is can they actually build u half 10 gawatt by the end of the year uh 2 gawatt by the end of this year is pretty uh pretty easy so can they build 8 gawatt next year >> yeah let let's come back to the motivation at the end uh and bring rake kick in here. So like yeah, the key
13:10
So like yeah, the key question is how do they get enough chips online and how do they get enough people to actually do this?
13:16
So that means having sites.
13:18
Um can you talk through high level without revealing too much what you've what you've got there? >> Yeah. Yeah. So this was fun.
13:26
This was basically um me and Zoo Hair like two days ago doing um just melting GPUs running through every single permit across the US to see what's kind of available.
13:39
We scan through like a million sites basically within like 2 days.
13:44
Um because if the sites are available, right?
13:48
Like if you like, think of it this way, if you're Elon and like what we've seen from the past and the track record, which is the track record just says all you need is a warehouse and a gas pipeline basically or even a gas pipeline and you can green field uh which he's recently done but we'll get into.
14:05
And so if that's really your only constraint is like getting this gas pipeline access and finding the turbines, which we'll get to as well, then you have a number of sites to start from to begin with, right?
14:19
If you wanted to go warehouses, you can go warehouses, too.
14:22
We even found a few of those when we dug through the leans against uh MZX or Elon, which is >> and we found like five very good candidates. >> Yeah. Yeah.
14:33
>> Yeah. Yeah. that are like a million square feet which is at least from a space point of view obviously there's much more but it's like you know over a gigawatt per like a million square feet potentially two >> yeah exactly like given the sizings on um macroart and macro harder like comparing I think macro hard was it
14:49
macro hard like 700 megawatts and the macroart is 500 megawatts and so yeah we had 1 million square foot warehouse we had an 800,000t warehouse I think another milliont warehouse and then a few sites where pipelines are popping up But broadly speaking, even though these sites look kind of like funky, right? It's just a warehouse on a plot of land.
15:08
It's just a warehouse on a plot of land.
15:10
Doesn't look like anything, it can be turned into this um just based off what we've seen in the past.
15:14
And so, as much as people hated it internally and probably externally too, it is possible, right? Like it is possible.
15:23
I got, you know, people in my DMs on Slack going, "This is an insane take.
15:30
like what are you guys talking about? I don't know about this.
15:34
Um it is possible and like we found enough sites that we can kind of show for it.
15:39
And then if you look at like power side of things, we also wanted to well we wanted to add a chart to the article uh regarding like the turbine availability but we didn't have enough time.
15:49
We kind of pushed it out pretty quick.
15:51
Anyway, when you look at the turbine availability, we found like 7 gawatt of like undisclosed like unidentified turbines, right?
15:58
And this is like just including those.
16:00
It's not including the ones that might be bought out on a secondary market like from a fermy or from someone else or from a different state, right?
16:07
And so these turbines are kind of there as well.
16:10
And so the power is there.
16:12
Now we have the warehouse kind of all that's left is the execution front.
16:17
And we've already watched in the past these 2 gawatts being built faster than anyone else in the industry. It's all kind of there.
16:26
all kind of there. as much as people don't like it and as seen as the number is >> so so I I think like if you if you count like the bottlenecks uh I think on the power side as you said people don't realize that there's actually much more available than expected and I would just
16:42
add like hey guys think of Oracle in New Mexico right like all these turbines are on the market uh and then if for coal that the the you know the pipeline is going to be delayed the fuel cells are also some of them are going to be available nebas New Jersey the engines are available they're on the Okay. And
16:56
And there's much more of these, right?
16:59
Anyways, so there's much power I think that people realize.
17:00
And Elon already has something like 9 to 10 gawatt of turbines on order or in operations.
17:05
So there's already a lot of of it in the fleet.
17:09
Um, but then what are the other bottlenecks?
17:11
Um, labor obviously is a huge one.
17:13
We just came up with a massive article on that.
17:14
Um, so labor is a gigantic bottleneck.
17:16
Um, electrical and cooling equipment can be a big big bottleneck.
17:20
Switch gear, uh, all that good stuff.
17:23
Now, lab labor is is really fascinating.
17:26
Um, I'll start with switchg and overall just electrical and mechanical and mechanical equipment.
17:30
Um, what I think is going to happen is he's going to extensively use equipment that comes from China.
17:37
Uh, it's not like he doesn't know Chinese supply chains.
17:41
Obviously, Elon knows them extremely well, better than probably any other firm that builds data centers in the US.
17:47
Um, he knows the electrical landscape extremely well because of, you know, what he does with Tesla and SpaceX as well.
17:53
uh you can buy pre-assembled, you know, modules out of China, uh extremely large scale.
17:58
Now, do customers want it?
17:58
Um if you're in for a, you know, 20-y year offtake, uh with high SLA, probably not.
18:07
Uh if it's an ondemand cluster, I think you kind of don't care what is the electrical equipment so long as the cluster is is usable and there's some kind of SLAs and whatnot that induce a penalty if it doesn't work.
18:17
penalty if it doesn't work. But I yeah I think basically you have to assume that when you when you have a unique product on the market which is you know a gig well three months from now [laughter] right a gig well three months from now um the kind of like SLAs and the demands from customers are way different and I
18:32
think the best proof of that is Google signed with them like that's the most unthinkable thing Google is a company that hates turnkey leases for data centers uh they they hate they just like to do everything themselves and yet they still signed with SpaceX right because they were kind of you know bullish quote unquote by the to market that was unbeatable. So, electrical equipment
18:48
So, electrical equipment from China pretty extensively.
18:50
Um, anything that they can get as fast as possible if it's unconventional, they're going to do that.
18:56
Now, labor, I think labor, it's fair to say, is probably the single bottleneck bottleneck these days on data centers.
19:01
Um, and that's where like it's really interesting to look at Elon's history.
19:04
Um, I think when you look at what he's done with Tesla or or SpaceX systematically, he always does things with much less labor than others.
19:16
Uh there's a precedent in the data center world as well. Uh Colossus 2.
19:17
Um the you know the numbers we can see out there uh point to about three uh thousand people workers per day peak um at that site um on a you know per gigawatt basis.
19:30
This is about 3x lower than what you see even from the very best data center developers and the ones that build very fast with like highly modular data center designs like call it you know Cruso for example.
19:39
They're amazing at what they do and yet somehow Elon needs 3x less people than them, right?
19:47
Um and again like it goes down to like you know extensive prefabrication from China and whatnot like he he he does stuff differently.
19:53
Um and and another one uh so on the supply chain uh I believe it's possible and by the way folks like we've we've we've told our institutional clients I think uh several weeks uh from now that on the trip side we can maybe go back to that when I finish on data source.
20:08
On the trip side, we saw orders on the supply chain for 5 to 10 gawatt just for next year.
20:11
So that's been uh in preparation for the last few months already.
20:16
Our Taiwan uh supply chain team tracked that in a brilliant way.
20:19
Uh and they're hiring by the way if you want to join our memory team.
20:22
Um but Rick, a question for you.
20:27
What about the permit bottleneck?
20:27
Um how how did they pull it off in in Memphis and how did they pull it off in Mississippi and how can they pull it off again at 4x scale? Is it even possible?
20:40
Well, Jeremy, I'm glad you asked. Um, [laughter] uh, yeah.
20:44
So, Mississippi was a special case, right?
20:45
Um, that's when we had the Colossus 2 article a bit ago where we talked about, okay, so they couldn't figure out how to get the kind of perpeting done for the power plant or the on-site generation within like Tennessee.
20:58
And so, the idea was, okay, well, the data center is right next to the border.
21:02
Let's just build it over the border.
21:03
And so, they went ahead and built it over the border.
21:05
They got some permits for like some some x amount of turbines. I think it was like a 1.
21:09
2 gawatt permanent power plant was the idea.
21:13
And then they started rolling in like mobile turbines and they're like, "Okay, you know, maybe down the line they become permanent."
21:18
And then they roll in more mobile turbines and it just completely goes past like the permitting allowance and they're like, "Okay, well this is this is not great." And then it keeps going.
21:29
They end up with a total of, you know, of course, 69 turbines.
21:33
Um, and so then eventually like there's some complaints, but then the DOJ intervenes and says it's okay.
21:39
Um, and so there's [sighs] it's kind of an unprecedented precedent in the sense where it's like I guess he can just kind of do these things on that front and so I don't put it past that we'll see these on the other fronts.
21:53
And then the nice thing as well is when you choose warehouses like in this instance the million square foot warehouse in the middle of Mississippi.
22:02
Um they're already permitted and zoned for the most part on these things.
22:06
U so you can actually just skip this. Right.
22:08
This is the benefit of not going with kind of a powered land solution instead.
22:12
You might take more time to get the actual construction permit across.
22:15
Or I can take a warehouse and just rip everything out and then plug it all in for myself.
22:21
All I need is an air permit then.
22:21
And so you can kind of go ahead Jeremy. [snorts] >> Yeah.
22:26
One thing I want to add to that is on the on the power side like like it's good to remember like he does things in a very unusual way.
22:33
Um and and so the warehouse can be like used for like large scale cooling, large scale whatever but maybe that parcel or even the parcels nearby can't be permitted for um you know air pollution.
22:44
Um and that's exactly what he did in in Colossus 2, right?
22:48
Like literally the power plant is something like 2 miles away.
22:52
uh from the from the warehouse.
22:52
Um and so he just like built a pri private transmission wire uh from the power plant to the to the to the warehouse.
23:03
Those are as far as I know running on medium voltage which is highly inefficient by any means.
23:07
But again like you know you want to do it fast like you're not going to so it's trade-offs.
23:13
It's not going to be it's going to be not going to be efficient but there's ways to do it.
23:16
And hey guess what they have the most brilliant electrical engineers on earth.
23:19
Um, so what they're going to do is they're going to find these warehouses and they're going to scout like everything nearby to see where can I build a power plant.
23:25
Even it's a if it's three miles away from it.
23:29
I'll just figure out the electrical way as well. Right.
23:31
So I think yeah unconventional it's going to be highly unconventional I think is the name of the game.
23:39
>> And I think like one one other thing I want to add to this too is like with these timelines right the question comes about like Jeremy brought up SLAs's earlier.
23:46
I think it's a really interesting topic as well.
23:48
Uh, I'll be brief though, but I think you're not going to these data centers for the highest SLA, right? Like that's obvious.
23:56
You're going for like a large scale contiguous cluster within the within the next 5 months or so.
23:59
And so this is fine in Elon's case, but it's also becoming more fine like broadly across the industry, right?
24:07
Like we have, you know, the anthropic self build that's like being talked about by multiple people now where you've got 99. 7% uptime, right? This is unheard of.
24:19
There's no multiple nines.
24:19
There's no tiers of this from like the uptime institute.
24:22
It's just, hey, let's remove the redundancy we need, you know, like if it's going to be internal electrical or if it's going to be the generators on backup. Remove all of it. I don't care.
24:32
I don't need these lead times.
24:32
I don't need this like capex.
24:33
And so people are also pretty fine with accepting lower SLAs's if it means more speed to market.
24:40
And so I think I don't think it's really a bad thing if he's just ripping open a warehouse and plugging in GPUs and, you know, maybe melting them really quick, but then everything works afterward.
24:49
It kind of doesn't really matter too much either.
24:55
Yeah, I mean, uh, fair enough.
24:55
I I think it it it matters [clears throat] to an extent.
25:01
Uh but they're they're well known for uh being pragmatic as opposed to overly conservative when it comes to redundancy and like keeping things up and online in terms of all of their different services which is you know if you got to go fast you have to make some concessions there.
25:14
So >> and they bring a lot of batteries from Tesla as well which helps on the redundancy. >> Yeah.
25:21
>> There is like some kind of Yeah. Like uh thing involved. Yeah. Yeah. Go ahead.
25:26
[snorts] >> Can we go back to talking about the potential customer here because obviously they've signed deals with Anthropic directly.
25:30
they've signed deals with Google.
25:31
You guys put in this article that Microsoft is the potential or the most clear customer.
25:35
In other words, there may be in my view uh Elon's not going to sell to SAM and OpenAI directly, but there's a way to get exposed to serving OpenAI's models which means selling directly to Microsoft.
25:48
So, can we talk through that? >> Yeah.
25:53
Um like again essentially you you get to a point where um the the demand for this kind of service is only from folks that can make these economics of like a hund00 million megawatt a year.
26:04
Um and there's you know free companies in the world today that have access to frontier AI models uh ripping the full benefits of them.
26:10
No cost no revenue share whatsoever just the cost of infrastructure open AI Microsoft and Anthropic right Microsoft having the OpenAI IP.
26:18
Um so Microsoft is in a pretty amazing position because uh they can monetize at this rate.
26:23
However, what what they they have two issues.
26:26
One is they had a they did this massive data center pause in uh second half of 24 first half of 25.
26:32
Um they were you know ready to build like more than more than anyone else and then they paused.
26:37
They didn't want to spend too much.
26:38
Um and so now they find themselves a situation where uh it takes a while to build data centers.
26:44
Um and so they're not going to be able to like have as much capacity as they would like to have.
26:48
Uh and the second thing is they signed a massive offtake contract with open AAI.
26:51
offtake contract with open AAI. Um so we estimate that at about 7 gawatt and so Microsoft a lot of the magazars they're building today are serving open AI but through infrastructure as a service which you know in our chart that we showed earlier of the economics that
27:04
would be closer to the 12 million bucks magwell a year uh as opposed you know the 100 right um and so the question is like okay if they they could potentially like accelerate like like crazy and that that's why year to date they've sort of woken up pretty dramatically u in any way they can examples. Data center pre-leasing is
27:23
Data center pre-leasing is always a good one.
27:25
So you just call up third parties and bu sign contract with them to you know build data centers.
27:29
Uh they've signed you know 7 gawatt of data center pre-leasing year to date.
27:32
They've been the most active company alongside meta for fairly similar reasons you could to some extent uh on the comput side.
27:41
So extremely active on the leasing front on the self they reacelerate a lot of a lot of their big sites like you know fairwater in Wisconsin and a lot of other sites here and there.
27:48
um Neocloud offakes they've they have been still very aggressive with folks like Nscale they kept signing more deals with them um and behind the meter agreement that's completely new uh they signed this 2.
27:59
7 gawatt off take agreement with uh you know Chevron in Picos County and Microsoft is the company that would would probably never do that right like they they they've been very committed for a long time to like 59s uh and to the grid and then they did this massive pivot where they're going to go like behind the meter uh in the middle of West Texas and So I think all of that tells you there's a big strategy change.
28:22
They're preparing for a massive acceleration of their infrastructure buildout.
28:25
However, it takes time to build stuff.
28:27
Uh you know they're a targeted date and we agree with that's why we forecast you know models 2028 for the Trevon deal for example.
28:32
A lot of the leases they've signed are for late 27 to 28.
28:36
There's a gap especially in late 26 and in the first half of 2027 right?
28:43
Um, and so Microsoft is left with this option of like or or this I guess debate of how do I get megawatts to to bank on the opportunity they have of selling GPT tokens at 100 mil megawatt a year. >> Right now. >> Right now. >> Yeah. Not into 28.
29:03
Like right now, you know, or as soon as possible next year. Yeah.
29:07
Um, and can you talk through the 90-day cancellation policy like in a little bit more detail?
29:13
more detail? I know you've brought this up a couple of times, but conceptually, if you're doing the self-build and you're doing these long-term offtake agreements, that's actually quite different when it comes to serving the
29:25
OpenAI tokens because like that's almost your base load in power terms, but now you've got this flexible sort of on demand capacity, which you know, it's it's you don't want to cancel everything you have and then try to strike new deals with these guys later. But
29:38
But conceptually either side can pull the chute on these uh SpaceX deals with 90 days, right? >> Yeah.
29:47
And I think you know conceptually what this means is um as you do these deals and you can also keep you can also plan ahead of time and you can do these deals you can think hey I'm going to do this for like you know 6 month to a year whatever and I'm also going to sign for a guy that's going to deliver for me a year from now so I can swap it.
30:01
But anyways, the point is like the that the deal structured is uh anthropic and and Google and and reflection same terms is uh whenever you want to cancel that that contract uh we can do so in 90 days.
30:16
Uh and that contract includes a monthly payment uh which annually equates to you know about 50 million bucks maybe a year.
30:22
Um, and so yeah, you just like the only burden on on your on your books is these three months that where you're going to still pay at at that rate and then it's done.
30:33
So from a balance sheet point of view, it's extremely easy to sign off on.
30:35
Uh the the period at risk is really low and that's extremely different from the compute agreements or data center agreements they've signed here today.
30:45
Those 10 gigawatts that equates to, you know, over well over actually $300 billion in total contractual value binding contracts.
30:52
So that's going to be spent no matter what.
30:53
Um so that's why it's so hard to plan infrastructure ahead of time is because a lot of capex is very balance sheet heavy whether you build or you lease is the same thing whereas this is on demand.
31:04
So the CFO can say hey you know uh I think we can make a whole lot of money if we take this there's no risk on my books.
31:10
I'm just going to do this for whatever again like 6 months a year.
31:14
I mean, Google has been pretty open publicly that they did this because they have this need.
31:18
Um, and then they're going to cancel the the contract like, you know, free 6 months from now.
31:23
Um, and again, like that's why this is an option at all for these companies.
31:27
Uh, and if you know, given what we've said on Google, maybe you can expand on that if it's the topic today, I don't know, but we think Google is not going to be able to like uh compete at the frontier, but Microsoft is at the frontier thanks to OpenAI. >> Yeah.
31:40
I mean I I think uh we don't know where things are going to play out in everything that's not coding today.
31:45
Uh if you believe that coding is the path to AGI and and coding, you know, turns into all of these other models, then Google is certainly behind and they they need to catch up and all of the signals we're seeing from them is that they are not doing this.
32:00
So not only is Microsoft pouring in all this money, Google is also doing this in the tune of $300 billion or something like that in capex.
32:09
So maybe the point that I didn't make on the last podcast that I like to make now is just that it's shocking to see people like Jeff Dean leave and raise even Dave Silver, John Jumper or Nam Shazir who left before Jeff Dean and Oral and all the guys now is that um they're leaving
32:25
to raise like one or two billion dollars which is an incredible seed round and you know the craziest thing ever except for the fact that you have to compare it to these guys pouring in $300 billion of capex and be like you couldn't give Jeff Dean 1% of this to do what he wants to keep him to stay, right? Instead, he's
32:41
Instead, he's got to go raise money from external parties just to pursue the research that he wants, leave all of the infrastructure, all the contacts, all the people and start something new.
32:52
Anyway, that's fascinating.
32:54
Um, but let's uh yeah, let let's get back to maybe the the question that's on everybody's mind right now, which is okay, so SpaceX has a demand.
33:05
We think there's a path for them to actually build all of this stuff in the timeline that people are talking about. How do they pay for it?
33:15
Um, you know, saying on earnings were exclusive to Nvidia.
33:17
It's the world's best hardware.
33:19
You know, Vera Rubin is so amazing.
33:21
Uh, coming from Elon Musk.
33:21
I I am assuming that's not free.
33:27
[clears throat] [laughter] I would put it this way.
33:28
No, but look, the first thing is um again like the the the revenue um you you just pay it with operating cash flow um you know we we so okay let's put it this way right right now like the the contracts that they've signed already which is just for a portion of their 2 gawatt uh it's for about a gig or one one to one and a half gives them uh 50 billion bucks of you know annualized revenue.
33:51
So already the compute they have they're they're you know they're they're making uh that amount of money which is you know pretty tremendous.
33:59
know pretty tremendous. So 4 billion a month of uh that's essentially pure cash extremely high margins like over 90% margins ebit margins of course um but even you know even a bit margins are very high and so the the thing is like if they keep in that direction and indeed what we said can happen which is
34:16
they can build 10 gawatt and monetize a 50 mil uh gawatt a year uh and we think they're still going to build for internal training and and we'll see how much and so on and so forth but let's say they they build five uh so five time 50 you know 250 billion bucks That's all operating cash flow. Um, that's going to
34:30
Um, that's going to help financing pretty dramatically.
34:32
And so, Nvidia, we think, is likely to step in as a financing partner. I I I don't know.
34:41
I I don't think anyone at all knows exactly what it's going to look like, but um the the like the point is if they can make 50 million bucks mega year, that is uh well over the cost of the GPU.
34:55
So the GPU is going to be, you know, paid back in under a year.
35:01
And so they don't even, you know, if they can get vendor financing, then it basically makes it like nearly cash neutral for them.
35:06
And they have a bunch of cash on the balance sheet as well and so on and so forth.
35:09
But and they're on the public markets as well, so they could always raise equity. There's always options.
35:14
Um, and we'll see what happens.
35:16
But, you know, at least I'm not too worried given the amount of operating cash flow they're making on these deals.
35:21
uh is >> well maybe the number one takeaway from this for me was just how big of a commitment this is to Nvidia.
35:27
So SpaceX and XAI has historically been building the bulk of everything on Nvidia but has been trying out other things.
35:36
They've been you know dipping their toes in the water of TPU and AMD GPUs and things like that.
35:41
And Elon declared on the first earnings call that they are Nvidia exclusive and he tweeted about it.
35:45
I think he specifically said we choose to go with Nvidia GPUs because they are the best.
35:53
So this is great for Nvidia too.
35:56
[laughter] It's not just uh you know them doing this for themselves uh that we take away from this.
36:02
I think this is also pretty positive on on Nvidia. >> Yeah.
36:07
And I mean I think it comes at like you know a pretty a pretty convenient time too with um at the moment we have the Nvidia backs stops kind of hitting the market overseas and then in the US there's direct data center leases um you know like with I mean the HUD 8 stuff.
36:21
Can I say that Jeremy by the way? Yeah. Okay. Uh >> public info man. >> It is.
36:29
>> Yeah it's public info. >> Okay.
36:30
Well, yeah, but the the HUD 8 leases like 7004 megawws and then you know all the back stuff's over overseas, right?
36:38
Um firmness, I guess, for example.
36:39
But it's pretty clear that Nvidia's like game for financing these situations, right?
36:46
like they understand that this is kind of the name of the game with Gemini or with Google becoming more and more of a TPU seller um and Enthropic continuing to build their workloads upon TPUs making it harder to like kind of switch back into Nvidia shapes.
37:01
Uh you end up with Nvidia pushed a little bit and having to really finance or really help out a lot of these projects to make sure that more folks get on the Nvidia like kind of side of things.
37:13
side of things. I mean uh whenever we go to the conferences too like Jeremy and I have had conversations with folks who are saying like yeah you know sometimes we'll be having a talk with Nvidia and like we're talking about taking this
37:29
site and then maybe having Entropic as an offtaker or something but then we've got you know the uh Google backs stop you know we kind of uh wave that around maybe we can you know uh we can curry some favor with Nvidia there And I mean it it tends to like happen, right? Nvidia kind of needs to like get more of
37:47
Nvidia kind of needs to like get more of these uh labs or these offtakers locked into the Nvidia ecosystem.
37:52
Um and so I think more than reasonable at the moment. >> Yeah.
37:59
And and and one thing I want to add as well on the on your on your question uh draw down the financing.
38:02
Um, what one comment that Elon made on these earnings I thought was fascinating and is he said like, you know, I'm going to try and be allowed to 20 gawatt, whatever.
38:13
But he said specifically gigawatts of power and cooling, right?
38:18
And so what what that tells me is that what he's going to do first and foremost is building data centers.
38:21
And when you have the data center built and it's like extremely close to delivery for the end customer, like everything becomes much easier.
38:30
easier. And so buying the GPUs at that point like you start to have many more financing options because you know you can like bully your customer into hey maybe it's a six-month deal maybe it's going to be a 40 mil whatever but
38:42
there's so many more options when you have the data center built and it's so close to delivery right and so u I think that's going to be his main focus currently he's obviously as we said talking to the chip supply chain we
38:53
think he's talking to everyone on the supply chain to ensure he's going to have enough chips uh and it could fall short and he said that explicitly he thinks it could fall short as well uh but he's going to try and make sure that he has enough data centers and that's going to be the number one priority is build them and then uh you know financing and and and buying the GPUs can sort of come easier once you have them built. So, uh, building 10
39:11
So, uh, building 10 gigawatts of data centers, you know, it's still a lot of money.
39:14
Uh, still a lot of money, but the SpaceX has cash and again they have like decent operating cash flow, you know. >> Yeah.
39:21
Well, it does it does feel like the ability to generate all this cash flow kind of depends on them having the chips on their balance sheet and kind of having the optionality on who to sell it for as sell it to as opposed to, you know, having a empty data center that somebody else can bring their own chips to that's on their own balance sheet.
39:39
probably not as big on the leverage side there in terms of sales, but um anybody who is asking the question about like the very natural question of like how are they going to be able to pay for this much capex I think you know what I'm hearing from you guys which makes sense is that you need to be considering
39:58
this as Nvidia's balance sheet probably the strongest in the entire world except for maybe Apple at this point and then Elon's ability to raise money on the back of an incredibly profitable incred incredibly cash flow generating business, which you know, one of, if not the best guys in the entire world at raising money. So, um, they're going to
40:15
So, um, they're going to go for it, man.
40:18
And it's, uh, it's inspiring to see.
40:21
It's an incredible, uh, it's an incredible thing to be a part of and to witness as we see them, uh, work through this.
40:29
Uh, where do we go from here, guys?
40:31
What do you think's the next step that people are going to take?
40:34
what are people going to be looking for on the next SpaceX earnings call or from Microsoft or from Nvidia or anybody else that we've mentioned this so far?
40:42
>> Yeah, I I think it really comes down to like how much do people believe in these ultimate economics and uh I I think like it's starting to happen.
40:51
Um, and I think more and more we're going to see towards the end of the year that yes, indeed selling tokens, especially Frontier AI tokens is insanely profitable and so profitable it's just going to drive the industry to do all kinds of creative things to like try try to bank on this.
41:09
So I think that's the only thing is do you believe in that profitability?
41:11
Uh, I think we'll see it from OpenAI Antropic as they're going to keep accelerating revenue because any mega they bring online gives them so much.
41:18
Um, and I think it's unrealistic to expect that if they both accelerate at this pace, Microsoft is not going to want to do the same thing.
41:29
Uh, if anything, the question should be asked to Microsoft like you can make that amount of money, why aren't you doing it?
41:35
Like, what are you doing about it?
41:38
I think I also think we somebody commented or like somebody on Twitter replied to one of the comments saying like you guys have you know 12 million per megawatt for the core weave deals but you're assuming you know 40 to 50 megawatt 40 to 50 million per megawatt
41:51
for the SpaceX deals and like yeah obviously this is these were the deals but I think what's kind of interesting to me is like in my opinion in my humble data center opinion and not the tokconomics guy um I think I think this is kind kind of overdue, right? I
42:07
I remember we at one point we did some we ran some analysis on I think it was 4. 6 or 4.
42:14
8 on like a GB300 and somebody found it was like 90 to 95% margins and this was I don't know if we maybe got that number wrong.
42:22
I mean, 85% is still insanely high, but in any case, reading these numbers was always to me like, well, I mean, somebody has to, you know, capture more value somewhere else, right?
42:32
Like, why are we selling these GPU hours for so cheap if they're making 90% margins?
42:36
Uh, to me, this kind of always made a bit of sense and they weren't ever going to like I mean I mean, maybe some world, but it never felt like they were going to keep these margins forever.
42:47
And I mean, it makes sense that somebody like Elon is coming in and charging these prices because why not?
42:53
They're still going to make so much money.
42:56
I don't really think it's going to kill them or hurt them too much.
42:57
And so I I think it makes a lot of sense, honestly. What's going on?
43:05
>> What's your uh conclusion, Jordan? Close it for us. >> Yeah.
43:09
>> Yeah. Well, I I think um look, the one thing in the back of my mind that we haven't had a chance to talk about on either of these podcasts this week is just the OpenAI uh release about the security incident with Hugging Face and how they had this uh talk at Black Hat Summit um which is from their Astra
43:28
model which is in testing right now and I'd say seems to be comparable or uh you know is certainly comparable to Mythos in terms of like the project glasswing sort of cyber security concerns that are being This is this is the one where you you kind of commented in Slack saying it like actually scared you, right? >> I mean, I've been scared for a while,
43:46
>> I mean, I've been scared for a while, but this is like two separate organizations identified a security incident where multiple agents were coordinating to attack infrastructure using really sophisticated methods.
44:02
I would say not super sophisticated in some of them, but the way the agents coordinated to do this and kind of went arai during evaluations was crazy.
44:11
Everybody should go watch that video.
44:14
I I won't even attempt to try to explain the details right now, but the point is that um if we are making the uh steelman case for the people who think Elon can't do this, the the the biggest concern that I have is not actually the execution and all the stuff that we've laid out here.
44:31
It's that demand has a problem in the future.
44:33
And not because I think the models aren't going to be good enough.
44:37
I actually think the models are going to be too good and then people will be so scared they're going to shut down access.
44:45
They're going to stop people being able to do this stuff.
44:46
Politicians are going to be involved.
44:47
And if there's any, you know, case where this doesn't work, it's more political than technical or operations related.
44:56
And uh we're yet to see that.
45:00
On the other side, the the creative part of my mind is going, well, there's lots of other use cases outside of coding and security that we can pursue in order to give people a lot of uh tokens and keep models going.
45:08
I mean, everybody's focused on coding right now.
45:12
And so, what about everything else when you point the GPUs and the researchers towards drug discovery and material science and weather prediction and video generation and robotics rake?
45:20
generation and robotics rake? like you know this is like if if you point the GPUs and the research effort towards something other than cyber security I think we will um realize a lot of benefits there but uh man this cyber security stuff is really concerning right now it's really really freaking scary man [laughter] yeah >> good yeah a nice positive note to end it on there [laughter] but yeah >> for what it's worth going to be coming
45:48
out with an article in a couple of weeks about our security experience testing a bunch of the neoclouds recently and how we found I mean these guys talk about zero days right which is like a publicly disclosed CVE meaning like a a security issue of some piece of software that's
46:02
running in somebody's infrastructure where the ability to exploit it requires downloading something from the internet and testing it works on the system and making in some cases just minor modifications to make sure it works. This is something an agent can do. It it
46:14
This is something an agent can do.
46:14
It it seems kind of obvious an agent can do this from everybody who's had experience doing research with these things.
46:19
And providers out there that are not SpaceX, but are on the lower tier, but there's plenty of them are running stuff not with like 6 week old zero days, but like three-year-old zero days that we found in a second when using this.
46:33
Just check the version of software they're running.
46:37
And the ability to develop PC exploits to show them that this is the potential concern took us like afternoons, not like weeks.
46:44
and months of effort and not needing to be a security expert or a Linux kernel expert or a Nvidia GPU driver expert or a Kubernetes expert.
46:53
You just are like, "Hey, model check for this version."
46:56
It checks for it and then it can build an exploit in a couple of hours.
47:00
So anyway, um that's when the the you know people are actually trying to work on it directly.
47:06
The really scary part about this story is that it was autonomous agents.
47:09
people weren't monitoring going and doing this by themselves because they were pursuing a goal of trying to find a data set to pass their eval.
47:17
So they went out and hacked hugging face which hosts these data sets by finding issues in like the HDF5.
47:25
You found a zero day in the HDF5 data format on hugging face. It's unbelievable.
47:30
So anyway um maybe okay because I tried to explain it I'll I'll give people another nugget so they can go watch the video.
47:39
The agents were coordinating with a message board using file names on a JROG artifactory service that ran remotely.
47:46
So an agent would go leave a note in the file name and then another agent would come back later and pick it up and be like, "Oh, you did this.
47:54
I'll keep working on that."
47:55
So they were like a swarm all pursuing the research to attack Hugging Face using file names on a like a file server. Not even.
48:06
They couldn't even have access to write the contents of the files, just the names. >> That's absurd. That is absurd.
48:19
>> Yeah, you should go you should go watch this video. It's 30 minutes.
48:21
Well worth it, guys, if you want to understand where we're at right now in AI progress at the frontier.
48:26
And it's a glimpse into what people like Elon who see Grock training and talk to anthropic and open AI are seeing with these models that they hold internally and don't release publicly because even the stuff that's public is so unbelievably profitable they don't even need to release these you know research projects that are being evaluated.
48:49
Um, so we're we're yeah, we're uh we're seeing rapid progress right now of how much these things can improve. Um, okay.
49:00
Any final thoughts after I went on that rant about cyber security when we were supposed to be talking about SpaceX?
49:08
>> Looking to the article. >> Yeah, really.
49:10
I was wondering how to change the topic back.
49:11
[laughter] It is a really out there call.
49:18
I think people will look back on it and go, "Wow, those guys are crazy."
49:20
But they got it hopefully. I mean, God, hopefully. Yeah.
49:25
>> I mean, we got a lot of hate when we had our call on Amazon and we were like, "Guys, just look at these data centers.
49:30
You know, they're going to accelerate revenue pretty obviously."
49:31
They're like, "No, you're idiots.
49:33
Never going to accelerate. They're losers of AI."
49:34
So, right now, SpaceX is the loser of AI. >> I guess, man. Yeah.
49:40
We'll we'll we'll see you at Christmas for the year end review 2027 podcast where we'll check in and see uh see if this one was right or wrong. Okay, guys. >> Looking forward.
49:51
>> I think we I think we got it. All right. Thank you, Jordan. >> All right. Good job, guys. Take care. Byebye.