When the history of the 21st century is written, you know, there was like the Victorian age. I think this will be like the age of Elon and Jensen because they are fundamentally altering the fabric of human society and civilization.
What happens if there's like a massive supply shortage?
Every time you've had a real profound new technology, you get a bubble because the markets get really excited and they get ahead of themselves. Things get overvalued. That overvaluation leads to an overbuild. One of the things that I think has been correct but ineffective is this idea that we need to stay ahead of China.
You're opposed to data centers. Well, you know what? It's probably the best thing that has ever happened to working-class Americans. We are re-industrializing America and it's awesome.
Assume that you're right. There's not a physics reason why this can't work.
An increasing fraction of the world's compute is going to be in orbit. This sounds crazy, but asteroid mining is going to be a very real thing. It has more gold, silver, platinum, every precious metal in it that exists in the earth's crust.
Every LP conversation that we have starts with like, how's this all going to go wrong?
Gavin, you've been out here hanging out on the West Coast over the summer and you've been talking about the fact that you're trying to find someone to give you like a bearish case, like to make your sentiment more negative. Have you found anybody?
No. And I ask everyone, my standard question is: can you tell me one quantitative data point in your business that's getting worse? Just one. That's my standard question. And at least in July and August, I haven't been able to find a single person. Now, if we're being honest, you know, Anthropic is in a quiet period, so maybe they've slowed down a little bit, but I do think the rest of the world has accelerated. OpenAI has clearly accelerated. Open source, I think, has accelerated more. And then I do think xAI, particularly after Grok 2, has had a pretty dramatic acceleration.
And so AI overall, it accelerated in July, it accelerated in August, and it can't keep accelerating forever. But it's just kind of wild that public stocks have kind of fallen out of bed over the last two months. You can drown crossing a river that's on average two feet deep, and so there's not a lot of action at the index level—
—but some of these AI names are in pretty significant drawdowns. They bounced a little bit in August, but still pretty big drawdowns, and things are broadly accelerating.
Our friend Eric Vishria did a podcast with Patrick O'Shaughnessy and he said maybe everyone wins.
You know, Anthropic wins, OpenAI wins, SpaceX wins, Meta wins. Google wins by selling a lot of TPUs. Open source wins, neoclouds win, inference clouds win on top of the neoclouds.
Yeah, and probably maybe not all applications, but applications that I think execute well and navigate this. But that feels like a very possible scenario to me, and there's so much zero-sum thinking in the world.
And by the way, on Anthropic, my hypothesis would be if you're Anthropic: one, I think they probably trued up and cleaned up some accounting—
Yes, definitely. You'd rather do—yes.
So you rebased and now you're comparable to OpenAI.
Yeah, in terms of revenue added, like in terms of the definition, now kind of revenue added.
Exactly. So you kind of rebased, and then they did their testing the waters. And then, I would hypothesize, because they've executed well, probably the next disclosure is a reacceleration. And then there's always this kind of funny game between the frontier model companies. They always have more advanced checkpoints. Anthropic is clearly waiting for OpenAI to release Strawberry.
And then it's like the next—
Here's Claude 3.5 Sonnet.
Magically, and just happened to be available several hours after Strawberry.
So I think they're being thoughtful heading into this IPO, and everyone is shooting at them.
Yes, everybody's shooting at them and they're in a quiet period, so they can't really shoot back.
There's a lot of gamesmanship, but I do think having OpenAI and Anthropic be public companies is going to be helpful for the market just because it's such a powerful force. A lot of public investors, you hear, "Oh, Sarah Friar said this at an all-hands meeting and it's on the cover of the Wall Street Journal. Okay, we're going to put that into our model."
I think it'll be better for them to be public. I am a little—Anthropic is now in their culture interviews saying, "How would you feel if the equity went to zero?"
Because we're looking for people who are mission-aligned.
Yeah, missionary not mercenary, yeah.
And that's great. We want missionaries, but we also want people to make money. At the end of the day, you can't afford the compute you want for your mission if the equity goes to zero. I'm no expert, but I'm pretty sure on that. I do think they are—
They're like the accidental enterprise company.
Oh, for sure. Oh yeah, they're kind of like the accidental everything.
Enterprise is just a byproduct of the mission, the objective at the end. Whereas I think OpenAI is a little more commercial, and obviously xAI a little more commercial.
But all of these companies, let's just say they have 10 gigawatts of power and they're allocating 8 to inference, and let's just say they're monetizing that inference at $60 billion a year, so $480 billion a year in revenue—
Which on a revenue payback basis would be like a one-year payback on a revenue basis, not a gross profit basis.
Yeah, on a revenue basis. And I've tried to use conservative numbers. People seem to think Anthropic and OpenAI are both monetizing at $100 billion a gigawatt today.
Let's say they have a big research breakthrough and they decide, "Wow, it is to our long-term advantage to go from 8 gigawatts allocated to inference and 2 gigawatts allocated to training, to 8 gigawatts on training." Then your revenue just went from 480 to 120 annualized revenue. And I actually think they would make that decision.
Yeah. And this is just something that public markets are going to really have to get used to.
As you say, OpenAI may be a different animal. I do think the realities—everybody has these ideals about how they're going to manage their business, then they go public and the stock is volatile and it really impacts employee morale, recruiting, retention. So I'd be surprised if they did such a dramatic cut, but a lot of the revenue is kind of under their control based on what checkpoint they release—
—where they price along this Pareto curve, and then how much they allocate between training and inference. Meta and Google and these kind of internet companies, it was pretty smooth fundamentally, even if the stocks were volatile.
Well, there was no massive trade-off they had to make in terms of the cost or infrastructure to serve the revenue side. They were totally separate.
Yeah, it's fascinating. If you go back to Eric's point of like it's all going to work, I actually think that's a great point. I describe it differently. I've had this conversation with LPs a lot, because every LP conversation that we have—it's probably the same for you—starts with like, how's this all going to go wrong? What's going to crash? And like, are the large models screwed or the labs screwed because of open source? And I'm like, this is all wrong. This is not an "or" thing, it's an "and" thing, right? Frontier is going to work really well, N minus one models are going to work really well, open source is going to work really well, there's going to be a bunch of application companies that work really well. The clouds are probably going to be fine; they're probably going to work really well. Like the five lab companies are probably going to do really well.
Yeah, and Nvidia is at the center of all of it.
Yes, yes. They're probably going to do pretty well.
Yeah. The last 26 years have taught me not to bet against Jensen.
Yeah, he's in a pretty good position here. I want to come back to that. The point that you made about training versus inference is an interesting one. It seems to me like the labs will decide to take all incremental profits, and probably much more than their profits, and invest them in training for a long period of time. Would you think that's fair? This is very different than the internet companies and the clouds. They just end up being supply-demand driven and they generate tons of profit, and they can still grow a certain amount. But they don't have, maybe with the exception of Meta, some big long-term bet that's like a multi-year payoff.
Yeah, I think it's important to be precise. For sure, I don't think they will generate free cash flow anytime soon. I think they're going to generate a lot of operating cash flow and then they'll use that to buy a lot of GPUs, XPUs, whatever we're going to call them. Or maybe they subsidize heavily. We do know that that's happening at the labs.
Their first-party products. Token consumption of their first-party products. They're doing all this research and they're spending a lot on data, on compute—
And the first-party products are like a heavy subsidy.
Products today, right? So it's 8 gigawatts of inference and 2 gigawatts internal research, and then 2 gigawatts is actually training.
Including probably the inference that goes into post-training.
Given the belief systems that they all seem to have about scaling laws, which continue to hold, I don't think any of them are going to be that focused on generating free cash flow. And you've seen—we saw Satya blink.
And Satya really regrets that, I think.
He kind of blinked, I think it was last year. He gave that great interview for Davos and they asked him about all the capex, and he said, "I know I'm good for my 80 billion." And I think they blinked a little, they slowed down, they regret that.
Then Dario famously went on a podcast and said, "Listen, some people are being super irresponsible with their spending. It's a hard decision because if you don't spend enough you could lose a lot of share, but if you spend too much you could go bankrupt. Those are both bad things, but bankruptcy is worse than losing share, so I'd rather be conservative." He was conservative, and OpenAI was aggressive, and now OpenAI is back in the game.
There are clear high ROIs on those independent of supply-demand mismatches that are happening. Clearly that seems to be the right decision short term and long term.
Yeah, absolutely. We calculate—Nebius and CoreWeave both gave some interesting disclosures, but you can kind of get to a 9- to 10-month payback for Nebius because you bring on a gigawatt, it costs $50 billion, and you can get an upfront payment for 50 to 60% of that from customers.
So now you're talking about $25 or $30 billion, and then you can monetize it if you put it into the spot market.
Yeah, spot paybacks are probably much faster than $9 or $10.
Yeah, you've got to assume a smoothed-out level like two bucks, three bucks, even with that. Now you can get like five bucks or eight bucks.
And then xAI, because they build these really big clusters and—I think a really important point—they bring them on fast—
They have an even faster payback and they can monetize at higher rates. I have tried to shift to think of pricing per megawatt rather than per GPU because it seems like where the world is. But xAI, the payback feels well inside of that.
In my career as an investor, there haven't been that many opportunities where you have companies that could deploy tens, hundreds of billions of dollars and get sub-one-year paybacks.
Yes, and it's kind of crazy. And then also, we should talk, particularly if you're buying Nvidia GPUs, to a lesser extent TPUs, you can finance these.
And there's a very low cost of capital to finance them today.
Yeah. Everybody's worked up about circularity and it's like, well, I don't know. I know a lot of smart people who work at Blackstone and KKR and Apollo, and they're the ones that are financing—
The ones who are financing it at a relatively low cost.
At a relatively low cost. I think one reason that's happening is useful lives just keep getting extended. As these models get better and better, and the ROI on token spend goes up, the monetization rate per gigawatt goes up. The true equity payback might be way inside of a year.
Exactly. Look, there's a case you could make that the prices of all this stuff go up, which could make the supply-side economics even more compelling. On the supply side, that's the dynamic today. It just is what it is. There's a ton of data points out there that paybacks are within a year.
I think it's actually interesting to think about the demand side, too, because the knock would be, "Well, in all these cycles you get some overbuild and then that destroys the economics of the supply side." The demand side today—what is the monetization of these companies which are doing call it $80 billion of revenue or something in that direction on the back of? What, like 30 million actual heavy paying users getting real value? I'm talking about developers.
I might take the under on 30 million.
Actually, what we see inside our companies is obviously there's a power law in which companies are spending a lot on tokens. Old banks are probably spending 1%; very tech-forward companies are spending high single digits. But if you actually look at the power law of what's happening with the actual engineers in those companies, the highest spending engineers are spending 10 or sometimes 100x more than the median engineer. Your 30 million is probably way overstated; it might be sub-10. So there's this question of where are we at in diffusion? There's 1.5 billion knowledge workers. It feels like we're nowhere on the demand side and we're massively supply-constrained.
I'm just curious, across the a16z portfolio, what are your best companies spending on tokens per month relative to human compensation? What rough range?
High single digits, some at 10%, like some of the very AI-native ones like 10% plus. Old economy companies, the ones that are probably doing a good job, are spending like 1%. So it feels to me like when I look at the supply-demand characteristics, people say, is that sustainable? When you pair it with the demand stuff, it feels specific. There could be things that disappoint us in terms of diffusion into the real economy, but it feels like over a 10-year stretch we're nowhere.
Yeah, absolutely nowhere. At Atreides, our internal token consumption has gone up 100x from the month of March—March through August, 100x our token spend. And we just got access to Grok Enterprise, and with two people using it, it looks like token spend might 10 or 20x in a month from August.
Yes. But it's actually extremely valuable. We have some heavy Grok users here, and it is very productive use. This is not wasteful tokens.
Listen, I try super hard. When I use AI, I just remember when I was trying to get my parents to shift to an iPhone and an iPad and get them used to it. They did a good job, I give them loads of credit. But I'm 50 years old. How old are you, David?
42. And you see these 23-year-old kids and just the way they use AI, they're fluent and native in it. I just feel like maybe in a way that no matter how hard I try, I will never be, and I'm trying really hard. We got Claude Code, and I built some stuff, did some cool stuff. In like three minutes of creating Grok bots, I had much better versions of everything I created.
I went on Patrick O'Shaughnessy's podcast like 5 months ago and I said, "I love having a podcast summarizer." Everybody was like, "How'd you do it?" I was like, "Well, just use AI and do it."
It takes 10 seconds in Grokbot. It's amazing and it's so good. And then a Substack summarizer, an X summarizer, an X sentiment tracker for topics and stocks. All of those would have taken me hours working with Claude Code, and they each took 7 to 12 seconds with Grokbot, and it's better. Grokbot does feel like another—at least for me—kind of ChatGPT moment. Because Claude Code, as I can see in the data, was powerful. I did some really cool stuff with it that was empowering and neat, like family calendar apps, things like that. But this is just 10 seconds and it's way better than what I was able to do.
The Claude Code thing was obviously the shift in coding. Our most sophisticated engineers were doing 20% of their code with AI to 90-plus percent. Everything you described in what you built with Claude Code or Codex is still kind of reactive in a way. It's summarizers, preparation, it's all knowledge enhancing, which is part of your job, but it's not actually doing the work for you.
Yeah, and now you have a Grok bot that says, "What are the recommended actions based on everything the other bots have learned today? What recommendations do you have for me today?" And that for sure is—
And it was so easy to build. I now have it—I'm horse-racing all these, which is I have Grokbot doing it, Codex doing it. All the action-taking for just "make me better at my job, look at everything I do, give me recommended automations you can do." I have Cognition doing it as well, which is one of our companies, very good at it. We're kind of on the bleeding edge of trying to do this stuff.
Just wait till everyone does this stuff.
And then when we actually click, "Yes, go just automate this." It feels like that's endless token consumption.
We should acknowledge the history of financial markets, dating back to the South Sea Bubble. Whenever you get this transformational new technology—I actually went on a podcast and said I thought the South Sea Bubble was connected to the invention of longitude and the ability to sail; turns out it was not, it was more of a tulip episode. But every time you've had a profound new technology—whether it's the automobile, the TV, the radio, internet, the PC, railroads, steel mills—you get a bubble because the markets get really excited and they get ahead of themselves. Things get overvalued. That overvaluation leads to an overbuild.
Particularly if you're funding it with debt—and even today a majority of this is still being funded out of operating cash flow, which I think is really helpful—debt-funded buildouts demand immediate ROI, not an ROI in three years.
You can't be off on the timing.
I talked to Patrick Watson about how wafers are these fundamental constraints, and that the buildout is so big and we're so early that it's impacting the raw productive capacity of so many industries. Everybody in copper, there's an AI thesis, and we're going to have to think about it to fill the—if 10% of what we just talked about comes true, we're in this acute shortage with several million people driving a crazy global compute shortage. What happens when that's 500 million? How many copper mines do we need to build to support this? It's kind of a wild thought. These fundamental constraints, I think, are slowing us down, and I think that's good for society.
I would now say rates and regulation: real rates are going up, and it just is what it is, which makes sense because we're investing a lot. And then regulation, man—I'm kind of shocked at what's happening in America.
We're in a really bad place.
I had this exchange with Dario from Anthropic on X last weekend. Dario said, "Hey, I don't think I've been negative. I've written two essays; one was positive, one was negative." Being 50% negative, particularly when it's a terrifying negative—
An existential negative: everybody might be out of a job, like that Eliezer Yudkowsky guy says, "If we build it, everyone will die." How about if we build it, we're going to cure cancer, we're all going to live forever? I thought one of the best things Dario said was what we need to do is stop talking about curing cancer and actually cure cancer.
Actually cure cancer and actually make breakthroughs.
My favorite line in the Bible is, "The truth shall set you free." The only group that can tell the AI industry's truth is the AI industry. They need to just start telling the truth: "Okay, you're opposed to data centers. Well, you know what? It's probably the best thing that has ever happened to working-class Americans."
Going to college might be significantly NPV-negative now because you can go learn how to be an electrician, a plumber, an HVAC tech, and make ungodly amounts of money. So this has been amazing for working-class Americans.
We now have a lot of data that particularly with behind-the-meter power generation, when a data center goes in, it transforms a town. Tax revenue doesn't double; it 10xes, and it is revitalizing all of these dying small towns all over America. We're getting much better at addressing the environmental stuff. Generally, they use natural gas, which is a pretty clean fuel.
The water consumption thing is totally debunked.
It's nothing. These are really good and they're having a really positive impact on the world. That's without even considering things like curing cancer, but somebody needs to tell that story.
I think the problem with it now is the burden of proof is on not curing cancer, but actually delivering some real tangible everyday American benefits beyond using ChatGPT or Grok to answer your questions or substitute for a search engine, right?
It does feel like we're pretty close to that. And by the way, one of the things that I think has been correct but ineffective is this idea that we need to stay ahead of China.
It is true, I'm very much a patriot and I believe that, but it's way too abstract for the average American.
Nobody's worried about China invading America.
Exactly. The ocean is really big. It's about affordability and how is this going to change my life for the better or worse, right? There's a pretty immediate impact you could feel. My favorite is Loudoun County, Virginia, which is the highest per capita income county in the US and it has the highest density of data centers.
And they make a tremendous amount of tax revenue from data centers. We should do this everywhere.
It was actually very funny: someone very opposed to data centers said, "Oh, you're for data centers. I'd like to see them put in the highest income zip code and the highest income county." And they're like, "Actually, the highest income zip code in America and the highest income county has the highest per capita concentration of data centers." So we've done that—
And it worked out really well.
Yeah, but don't bother me with the details, I'm on to my next talking point.
That's good, that's good.
All those talking points, it's tragic. There is an organized CCP-funded campaign against data centers here in America. I think a lot of it gets laundered through TikTok, and it's just tragic because the other thing that's happening is this is re-industrializing America.
The combination of having natural gas here at two or three bucks—it's now 25 bucks in Europe and Asia, or 20 bucks or whatever it is. Natural gas is an important input to the cost of electricity, which is an important input to almost all manufacturing processes. So we have a huge cost advantage for that basic input now. You have that happening, and you have this kind of data center boom happening. We are re-industrializing America, and it's awesome. This is what everyone in both parties has wanted for a long time: bring industry back. Small towns that were left behind by the steel mills closing—well, data centers are bringing them back.
Yeah, but somebody has to tell that truth. I mean, I try to do it on every podcast, but I'm just a dude.
Your audience is the tech audience that already believes you; you're preaching to the choir, if you will. Meta is probably doing the best job of telling that story, I would think.
One reason is it's really wired into Meta's DNA. One of the first things they started doing as a public company—I don't remember if it was on their first earnings call—Sheryl Sandberg would run through 10 or 15 very specific small businesses that had started using Meta's advertising products and the impact it had on that business. "This cake bakery in Des Moines worked with Meta. It was two women who were single mothers working by themselves, and now they have 15 locations, they employ 50 people. This has been amazing for Des Moines and it's been transformative for them." They would just run through that every time.
I do think the entire AI industry—I'd love to see xAI, Anthropic, OpenAI, Google, Meta say, "Here are real businesses and real Americans," and either name the business or get permission to name the American, or anonymize it: "This is a really positive thing it had on their life."
Already very tangible, yeah.
Nvidia, AMD, Broadcom, all of them. Just run through specifics, because the truth will set you free, but only if you tell it.
Exactly. It seems more likely than not, given that fact pattern, if you go back to the macro situation that we're in, that we underbuild on the supply side.
Oh yeah, through 2028. There's no capacity available with all the forecast builds that will happen through '28, which are probably now going to be delayed given the political dynamics. Everybody's worried about oversupply; I'm more worried about massive undersupply.
Massively undersupplied, yeah, exactly. If that's the scenario, you could see a scenario where you see big price increases to access the intelligence, which is the opposite direction of where everybody thinks this is going to go.
Dwarkesh had a wild point. He was positing the cost of a token could go up 10x or something like that, which is crazy, but we do live in a supply-demand world.
It's conceivable if the demand grows massively. By the way, the whole premise of this that's happening so far is that there's a massive amount of consumer or user surplus being generated, right? Why do people select the frontier tokens when they could use the cheaper tokens to do most tasks? There's many reasons why, but the biggest one is because there's a tremendous amount of surplus even if you're using the frontier tokens, right?
What happens if there's like a massive supply shortage?
I think the consequence of these data center degrowthers may be real compute inequality, where big companies and wealthy people can afford compute, and then two years from now they'll be on about that. It's like, well, that happened because of you. That happened because you wouldn't let us build data centers.
And by the way, we've seen this, right? The path to a low-cost product delivered to consumers in a mass market is advertising. It takes a long time to build an advertising business, as we've seen with all the consumer internet businesses that we've invested in over the years. And so there may be a disconnect in the period where you can't actually offer that.
That would be a terrible outcome for the world. Nobody wants that, so we need to build a lot of data centers.
A compute inequality future is not a good future for anyone, which is another reason open source is so important.
I had Grok make me a meme of that three-headed dragon, and one of the heads is kind of confused about all of the really stupid bearish AI narratives. People have this idea that open-source tokens are free. It takes the exact same amount of compute, all else equal, to make an open-source token as a frontier token for a comparably sized model. There's a lot of nuances there, but that's broadly true.
It's just a question of what are the margins that are charged on top of that. And even then, the Kimi license—something that I don't think a lot of people appreciate—the Kimi license stipulates a 30% share of any revenue.
Kimi is taking a 30% cut of all the revenue generated on it, and this is because it's open weights, not open source.
Exactly. But it's also extremely token-hungry, too. On a task basis, it's far more inefficient, so it's very costly.
Jensen is a great patriot, great American. We're so lucky to have him and Elon. When the history of the 21st century is written—there was the Victorian age—I think this will be the age of Elon and Jensen, because they are fundamentally altering the fabric of human society and civilization with AI, SpaceX making humanity multiplanetary, Starlink bringing low-cost internet access to the poorest communities in the world, which is an amazing surplus.
There was never going to be an economic case to build internet access in those places because of the cost and the willingness to pay. Any incremental internet capacity is not going to be built in a traditional sense on Earth; it's going to come from space, and so that is a huge unlock.
We should all be grateful for them because they're making the future as exciting and inspiring as possible.
Whenever I talk about SpaceX—and it's obviously near and dear to both our hearts—first of all, the orbital data center stuff, it's not like big buildings in space. It's helpful to actually think of it as the size of an airplane.
People are picturing the Death Star or the Pentagon floating around in space. That's not what it is at all.
It's the size of an airplane, right? A rack of 72 chips.
It's like five of us standing together, roughly, with solar wings.
And then you keep it in a sun-synchronous orbit, so you have the radiator always in the shadow of the rack. That's how you cool it.
There's all these people on X and they're like, "I am a physics PhD and this is impossible." There's a friend who's another investor who actually is a physics PhD who had many arguments with them and said, "I am a PhD and this is impossible." Then he goes to the SpaceX day and talks to the SpaceX engineers, and he's like, "Well, I was wrong."
If you're an astrophysics PhD, you are brilliant; you're hanging 100 IQ points on me. Have you thought about this for an hour? Have you thought about it for 10 hours? Because you have 10,000 of the world's smartest engineers at SpaceX who've thought about this each for hundreds, if not thousands, of hours. The sum of that working with very sophisticated engineering tools is it's a solved problem. In their minds, it's dramatically simpler and easier than a Starlink satellite, because a Starlink has to have the phased arrays and move around.
Assume that you're right; there's not a physics reason why this can't work. Cost-wise, it seems really imposing, but the history of the Elon companies is the cost curve gets dramatically better. When we first invested in SpaceX, Starlink was not commercially available and we had all these questions about how the economics would proceed over time. The same on the launch side, the same with the Model 3. I have to think that that will get solved, paired with the fact that we're going to have massive self-inflicted undersupply on Earth. It feels clear to me at a minimum it will be swing capacity, and in the fullness of time maybe it will be larger.
It's really simple. The question people should be asking about orbital compute, which is the one SpaceX is focused on, is Starship reusability. Because the math is: let's just say it's $50 billion a gigawatt, and let's say 35 of that is IT. So that's the same, and maybe it grows a little because it's going into space. The rest is power, cooling, labor, all sorts of things that you don't need in space because you have the solar panel and the big radiator. Call that $15 billion, and that's probably inflationary here on Earth.
Yeah, because labor fundamentally feeds into that. We just talked about what's happening to electrician compensation and materials.
Yeah, all of it. Yeah, we're going to have—yeah, we're going to run out of—you know, the copper bulls are focused on copper shortages. All of it.
Yeah. So that 15 billion is inflationary. And so what you have to compare it to is the cost of launch. And with Starship reusability, that goes to under a billion. So the economics just instantly flip. Now, you're always going to train on Earth. There will always be advantages to having GPUs right next to each other. Speed of light limitations are a real thing. Latency matters. So, data centers on Earth, they're not going anywhere. I think they're going to continue to be very, very valuable. But an increasing fraction of the world's compute is going to be in orbit. And you know, Elon said that he and Jensen have co-designed a Rubin rack
and they're—it's going to launch in the fourth quarter of '27.
And let's just say—let's just say he's off by two quarters.
Yeah. That's still okay. That's pretty soon.
That, as Brad Gerstner says, nobody's really paying attention to this and it's happening in plain sight. And it kind of to me solves for something, mid-single-digit billions today,
which by the way, is just keeping share constant,
of what's happening with—
not presumably taking any share on Grokbot.
Yeah, from 3 billion. And by the way, man, I'd probably take the over with Grokbot.
changing by the day just based on my own usage and the number of people who are hitting their usage limits. And then you are starting to get from Grokbot like, "Hey, our servers are overloaded," every once in a while, and they have a lot of compute. So it's just like, okay, you don't want to debate orbital data centers,
no problem. Well, like Starlink Mobile, they have a pretty clear, credible plan
for how that's going to work, and wireless is another $800–$900 billion of revenue that they address. So your mobile plus your broadband, whatever, call it close to $2 trillion of a market,
and then you have a really rapidly growing AI AR base.
Yeah, AI AR, you've got the cloud business.
Yeah. So great, you're an orbital compute skeptic, no problem. It doesn't matter.
We don't even need to. We could just look at things that are happening today with terrestrial compute, with Cursor, with Grok, with Grokbot. By the way, I think X ads are—we have telemetry.
