Skip to content
a16za16z

Why AI Demand Is Outrunning Compute Supply

a16z’s David George sits down with Gavin Baker to unpack the state of the AI boom, why demand for intelligence may still be dramatically underestimated, and why the outcome doesn't necessarily have to be winner-take-all. David and Gavin explore the possibility that frontier labs, open-source models, applications, clouds, and NVIDIA can all capture significant value as AI adoption expands. They dig into the economics of the infrastructure buildout, why compute investments can have unusually fast payback periods, and what happens when today's relatively small group of heavy AI users expands to hundreds of millions of people. They also debate the risk of an AI bubble versus an AI shortage, the backlash against data centers, orbital compute, the rise of multi-model architectures, and NVIDIA's position at the center of the AI supply chain. Gavin makes the case that the AI buildout could help reindustrialize America, while David explores whether the bigger near-term risk is not overbuilding, but failing to build enough. Timestamps: 00:00 - Intro 01:06 - Finding the Bear Case: Why Gavin Can't Find One 08:06 - How's This All Gonna Go Wrong? An "And" Thing, Not an "Or" 09:09 - Will Labs Reinvest All Their Profits Into Training Forever? 14:44 - The Demand Side: 30 Million Heavy Users & the Diffusion Question 19:16 - From Reactive Coding to Fully Autonomous Agents 30:18 - What Happens If There's a Massive Supply Shortage? 34:21 - Orbital Data Centers: The SpaceX Compute Play 40:00 - Starlink's $2 Trillion Market & the Heads-You-Win Compute Bet 44:04 - The Most Futuristic SpaceX Idea: Asteroid Mining 54:01 - Who Becomes the Abstraction Layer of Intelligence? 57:57 - Harvey, Cursor & Vertical AI Winners 01:00:26 - Jensen, Nvidia & the Central Bank of AI 01:12:16 - How Chip Deal Structures Reveal True Customer Preference Resources: Follow Gavin Baker on X: https://x.com/GavinSBaker Follow David George on X: https://x.com/DavidGeorge83 Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Show on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Show on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see http://a16z.com/disclosures.

Gavin BakerguestDavid Georgehost
Aug 31, 20261h 14mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:001:06

    Intro

    1. GB

      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.

    2. DG

      What happens if there's like a massive supply shortage?

    3. GB

      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.

    4. DG

      One of the things that I think has been correct but ineffective is this idea that we need to stay ahead of China.

    5. GB

      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 reindustrializing America, and it's awesome.

    6. DG

      Assume that you're right. There's not a physics reason why this can't work.

    7. GB

      An increasing fraction of the world's compute is gonna be in orbit. This sounds crazy, but asteroid mining is gonna be a very real thing. It has more gold, silver, platinum, every precious metal in it than exists in the Earth's crust.

    8. DG

      Every LP conversation that we have starts with like, "How's this all gonna go wrong?"

  2. 1:068:06

    Finding the Bear Case: Why Gavin Can't Find One

    1. DG

      Gavin, uh, 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 like trying to find someone to make you, to give you like a, a bearish case, like to make your sentiment more negative. Um, have you found anybody?

    2. GB

      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 it's at least in July and August, I haven't been able to find a single person. Now, if we're, if we're being honest, you know, Anthropic is, um, you know, in a quiet period, so maybe they've slowed down a little bit. But I do think the rest of the world has accelerated. You know, OpenAI's clearly accelerated. Open source I think has accelerated more. And then I do think Groq, particularly after GroqBot, has had a pretty experience-- uh, 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, you know, public stocks have kind of fallen out of bed over the last, you know, two months. And I mean, you know, it's, uh, the, you know, the, you, you can, you can drown crossing a river that's on average two feet deep. And so, you know, there's not a lot of action at the index level.

    3. DG

      Right.

    4. GB

      But some of these AI names are in pretty significant drawdowns, and they bo- they bounced a little bit, um, in August, but still pretty big drawdowns, and things are broadly accelerating.

    5. DG

      Yeah.

    6. GB

      It's, um... You know, our, our friend Eric Fisher did a podcast with Patrick O'Shaughnessy and he said maybe everyone wins.

    7. DG

      Yeah.

    8. GB

      You know, Anthropic wins, OpenAI wins, SpaceX wins, Meta wins. Um, you know, Google wins by selling a lot of GPUs. Um, open source wins. Neoclouds win. Infra- you know, inference cloud, uh, the in- inference clouds win on top of the neoclouds. Um-

    9. DG

      Applications win.

    10. GB

      Yeah. Every... Yeah, and that ki- may-probably, maybe not all applications. Appli- 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, what is... M- my hypothesis would be if you're Anthropic, one, I think they probably trued up and cleaned up some accounting.

    11. DG

      Yes, definitely.

    12. GB

      You would, you'd rather do that.

    13. DG

      Yes.

    14. GB

      So you rebased-

    15. DG

      Yeah

    16. GB

      ... and now you're comparable to OpenAI.

    17. DG

      Yeah, in terms of revenue added. Like in terms of the definition and now I think kind of revenue added.

    18. GB

      Exactly.

    19. DG

      Yeah.

    20. GB

      So you kind of rebased and then they, you know, they did their testing the waters. Um, [clears throat] and then, you know, I would hypothesize because they've executed well, probably the next disclosure is a re-acceleration. 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 Astra.

    21. DG

      Yes.

    22. GB

      And then it's like-

    23. DG

      Then the next Fable

    24. GB

      ... the next day, here's Fable 5.1.

    25. DG

      Yes, exactly.

    26. GB

      Magically, it just happened to be available several hours after Astra.

    27. DG

      Yeah.

    28. GB

      So I think they're being thoughtful, um, in, you know, heading, heading into this IPO and everyone is shooting at them.

    29. DG

      Yes.

    30. GB

      Everybody's shooting at them and they're in a quiet period, so they can't really shoot back. Um, and so it's, you know, there's a lot of gamesmanship, but I do think having OpenAI and Anthropic be public companies is gonna be helpful for the market just 'cause it's, you know, it's such a, uh, powerful force and a lot of public investors, you know, you hear, oh, you know, Sarah Friar said this at an all-hands meeting and it's on the cover of Wall Street Journal. Okay, we're gonna put that into our model.

  3. 8:069:09

    How's This All Gonna Go Wrong? An "And" Thing, Not an "Or"

    1. DG

      all gonna work, like, I, I actually think that's a great point. Like, I, I, I describe it differently. I've had this conversation with LPs a lot, 'cause every LP conversation that we have, it's probably the same for you, starts with, like, "How's this all gonna go wrong?"

    2. GB

      Yeah. Is, is AI a bubble?

    3. DG

      And it's like, it's like, "What's g- what's gonna crash?" And I'm like, th- this is the, this is the... And like, "Oh, are the, are the large models screwed? Are the labs screwed because of open source?" And I'm like, "This is, this is, this is all wrong." Like, this is not an or thing, it's an and thing, right? Like, this is an and thing. Um, Frontier's gonna work really well. Like, N minus one models are gonna work really well. Open source is gonna work really well. Um, there's gonna be a bunch of application companies that work really well. Like, the clouds are probably gonna be fine. They're probably gonna work really well.

    4. GB

      And-

    5. DG

      Like, the five lab companies are probably gonna do really well.

    6. GB

      Yeah, and NVIDIA is cinti- is sitting-

    7. DG

      NVIDIA's at the center of it all

    8. GB

      ... is at the center of all of it.

    9. DG

      Yes, yes.

    10. GB

      Yes. [chuckles]

    11. DG

      They're probably gonna do pretty well.

    12. GB

      Yeah. But, uh, uh, the last, um, 26 years have taught me not to bet against Jensen.

    13. DG

      Yeah, he's, he's, he's in a pretty good position here. Um, I wanna come back to that. The, the point that you made about training versus inference is an interesting one.

  4. 9:0914:44

    Will Labs Reinvest All Their Profits Into Training Forever?

    1. DG

      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? Like, it's, it's very different than, like, the clouds, you know. 'Cause like the c- the cloud, like 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. Like, but they don't have some, maybe with the exception of Meta, like, some big long-term bet that's like a multi-year payoff.

    2. GB

      Yeah, I think it's important to kind of be precise. They, I, for sure, I don't think they will generate free cash flow anytime soon. I think they're gonna generate a lot of operating cash flow, and then they'll use that to buy a lot of, you know, GPUs, um, XPUs, whatever, whatever we're gonna call them. Um-

    3. DG

      Or maybe they subsidize heavily. Like, we do know that that's happening at the labs.

    4. GB

      Subsidize what heavily?

    5. DG

      Their first-party products.

    6. GB

      Oh, yeah, yeah.

    7. DG

      So token consumption of their first-party products.

    8. GB

      Oh, yeah, yeah, yeah.

    9. DG

      So, like, they're doing all this research and they're spending a lot-

    10. GB

      Yes

    11. DG

      ... on data, on compute.

    12. GB

      Yeah, so it, it's-

    13. DG

      And then first-party products that are like a heavy subsidy products today, right?

    14. GB

      Yeah, so it's eight gigs of inference and two gigs-

    15. DG

      Yes, exactly

    16. GB

      ... is for internal research. And then, you know, two gigs is actually training.

    17. DG

      Yeah, exactly.

    18. GB

      Um, and, you know, including probably the inference that goes into post-training. Yeah, I don't, I, I think 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 gonna be that focused on generating free cash flow. And you've seen, right, we saw Satya blink.

    19. DG

      Yes.

    20. GB

      And Satya really regrets that, I think.

    21. DG

      Yeah, yeah.

    22. GB

      Um, you know, he kind of blinked, I think it was last year. You know, 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." [chuckles]

    23. DG

      Right. Of course.

    24. GB

      And, and I think they blinked a little. They slowed down. They regret that. And then Dario famously, he went on a podcast and he made f- and he said, "Listen, some people are being super irresponsible with their spending, and 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. And, like, those are both bad things, but bankruptcy is worse than losing share, so I'd rather be conservative." And he was conservative.

    25. DG

      And now it's-

    26. GB

      And OpenAI was aggressive, and now OpenAI is back in the game.

    27. DG

      And SpaceX was aggressive.

    28. GB

      And SpaceX was aggressive.

    29. DG

      And so, you know, like, there are clear high ROIs on those, independent of s- supply-demand mismatches that are happening. Like, clearly that seems to be the right d- decision, short term and long term.

    30. GB

      Uh, yeah, absolutely. I mean, we, we calculate, you know, n- Nebius, um, and CoreWeave both gave some interesting disclosures. But you can kind of get to a 9 to 10 month payback for Nebius because, you know, okay, you bring on a gig, it costs 50 billion. You get, you can get an upfront payment for 50% to 60% of that for customers.

  5. 14:4419:16

    The Demand Side: 30 Million Heavy Users & the Diffusion Question

    1. DG

      The demand side today, like what are we moneti- like the monetization of these companies which are doing, call it 180 billion of revenue or something in that direction, um, is on the back of what, like 30 million actual heavy paying users, like re- getting real value. I'm talking about like developers. It like-

    2. GB

      I might take the under on 30 million, man

    3. DG

      ... there's, so, so call it... Yeah, actually, what we see inside our companies is, you know, obviously there's a power law in which companies are spending a lot on tokens. Like old banks are probably spending 1%. Very tech forward companies are spending high single digits. But if you actually look at the sort of the, the power law of what's happening of the actual engineers in those companies, the highest spending engineers are spending 10 or sometimes 100x more than the median engineer. And so, yeah, your 30 million is probably way overstated. It might be sub 10. And so there's this question of like, where are we at in diffusion? There's one and a half billion knowledge workers. Like it feels like we're nowhere on the demand side-

    4. GB

      Yeah

    5. DG

      ... and we're massively supply constrained.

    6. GB

      And what are, I'm just curious, across the a16z portfolio, if, what are your best companies spending on tokens per month relative to human compensation? What rough range?

    7. DG

      Oh, high single digits, some at 10%, like some of the very AI native ones, like 10% plus. And so, you know, and, and then old economy companies are spending, the ones that are probably doing a good job, like 1%. So it feels to me like when I look at the supply demand characteristics, it's like supply stuff w- people say, "Is that sustainable?" Well, like when you pair it with the demand stuff, I, I, it feels, it feels pretty sim- Like there could be things that disappoint us in terms of like diffusion into the real economy. But it feels like over a 10-year stretch, like we're nowhere, right?

    8. GB

      Yeah, absolutely nowhere. And I just... My... So at, at Atrentes, our internal token consumption has gone up 100x from the month of March. March through August, 100x-

    9. DG

      Yeah

    10. GB

      ... our token spend. And we just got access to, uh, GroqBot Enterprise, and with two people using it, like it looks like it, token spend might 10 or 20x in a month-

    11. DG

      Yes

    12. GB

      ... from August.

    13. DG

      Yes.

    14. GB

      Like, [chuckles] like I think-

    15. DG

      But it, but it's actually extremely valuable. Like w- we, we have some heavy GroqBot users here, and like th- it is very productive use. Like this is not like wasteful token spend.

    16. GB

      Y-yeah, I was... And, and listen, like I, I try super hard. I, you know, I always, uh, when I use AI, I'm, I just remember when my parents, like I was trying to get them to shift to an iPhone and an iPad and like-

    17. DG

      [chuckles]

    18. GB

      ... you know, get them used to it. And like, you know, they did a good job. I give them loads of credit. And but, you know, I'm 50 years old, you know, like how old are you, David?

    19. DG

      42.

    20. GB

      42. And you see these like 23-year-old kids and just the way they use AI, they're just 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. But you know, like we got Cloud Code. I try, I, you know, I built some stuff, did some cool stuff, and in like, I don't know, three minutes of type cre- creating GroqBots, I had much better versions of everything I created. You know, so I went on this Patrick O'Shaughnessy pod- podcast like five months ago, and I said, you know, like I love having a podcast summarizer. Everybody's like, "How'd you do it?" I was like, "Well, just use AI and do it."

    21. DG

      Yes. Pretty simple, yeah.

    22. GB

      It takes 10 seconds in GroqBot.

    23. DG

      Yes.

    24. GB

      It's amazing, and it's so good.

    25. DG

      Yeah.

    26. GB

      And then, you know, a Substack summarizer, an X summarizer, um, an X sentiment tracker for topics and stocks.

    27. DG

      Yeah.

    28. GB

      And like that, all of those would've taken me, I don't know, hours working with Claude Code, and they each took seven to 12 seconds-

    29. DG

      Yeah

    30. GB

      ... with GroqBot.

  6. 19:1630:18

    From Reactive Coding to Fully Autonomous Agents

    1. DG

      Yeah. Yeah, the Claude Code thing like was obviously the shift in coding and, you know, our, our, our most sophisticated engineers, you know, were doing whatever 20% of their code, you know, with, with, with AI to like, you know, whatever, 90-plus percent.

    2. GB

      Yeah.

    3. DG

      And so now I think everything you described in what you built with Claude Code or Codex is still kind of reactive-

    4. GB

      Yeah

    5. DG

      ... in a way, right? Like it's, it's still, you know, it's like summarizers-

    6. GB

      Yeah

    7. DG

      ... prep- preparation. It's all like knowledge enhancing, which is part of your job, but it's not actually doing the work for you.

    8. GB

      Yeah, and now you, uh-

    9. DG

      Now you can actually do-

    10. GB

      You have a GroqBot that says, "What are the recommended actions?"

    11. DG

      Yes, exactly. I'll go do that.

    12. GB

      Based on everything the other bots have learned today.

    13. DG

      Yeah.

    14. GB

      What recommendations do you have for me today? And that for sure is like... And it, it was so easy to build. Um-

    15. DG

      I now have it... It's o- I'm, I'm like horse racing all these, which is like I have, uh, uh, GroqBot doing it, Codex doing it, all the like action-taking for it.

    16. GB

      Yeah, yeah.

    17. DG

      'Cause I just, I wanna know, uh, make me better at my job.

    18. GB

      Yeah.

    19. DG

      Look at everything I do. Give me, give me recommended automations you can do. I have Town doing it as well, which is one of our companies-

    20. GB

      Yeah

    21. DG

      ... early stage, but very good at it. Um, but and we're like kind of on the bleeding edge of trying to do this stuff.

    22. GB

      Yeah.

    23. DG

      Just wait till everyone does this stuff.

    24. GB

      Yeah.

    25. DG

      And then, and then when we actually click like, "Yes, go just automate this"-

    26. GB

      Yeah

    27. DG

      ... it feels like that's sort of endless token consumption.

    28. GB

      Yeah, and but I do... We should acknowledge like the h- the, the history of financial markets, you know, dating kind of back to like the South Sea Bubble, is whenever you get this transformational new technology, um... I actually went on a podcast, I said I thought the S- South Sea Bubble was connected to like the invention of longitude and the ability to sail. Turns out it was not. [laughs] It was just, it was kind of like a, a more of a tulip episode. But like every time you've had a real, you know, profound new technology, you know, whether it's the automobile, the TV, the radio, internet, the PC, um, railroads-

    29. DG

      You always get a bubble, yeah

    30. GB

      ... steel mills, y- you get a bubble because the markets get really excited and they get ahead of themselves. Things get overvalued. That overvalue, uh, overvaluation leads to an overbuild, and then particularly if you're funding it with debt, um... And, and even today, a majority of this is still being funded out of operating cash flow, which I think is really helpful. Um, you know, debt-funded built- build outs, they demand immediate ROI.

  7. 30:1834:21

    What Happens If There's a Massive Supply Shortage?

    1. GB

      Yeah.

    2. DG

      Which is crazy, but like we do live in a supply-demand world. Like it's conceivable-

    3. GB

      Yeah

    4. DG

      ... if the, if the demand goes massively. And by the way, it, the, 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?

    5. GB

      Yeah.

    6. DG

      So like why do people select the frontier tokens when they could use the cheaper tokens to do most tasks? There's many reasons why, but like the biggest one is because there's a tremendous amount of surplus even if you're using the frontier tokens. Right?

    7. GB

      Yeah, absolutely.

    8. DG

      And so yeah, w- what happens if there's like a massive supply shortage?

    9. GB

      Well, I think that would be the, you know, kind of funny, the consequence of like the, these like data center degrowthers, um, may be like real compute inequality, where big companies and wealthy people can afford compute. And then, you know, two years from now they'll be on about that, and it's like, "Well, that happened because of you."

    10. DG

      Yeah. Yeah, yeah, yeah.

    11. GB

      You know, that happened because you wouldn't let us build data centers.

    12. DG

      Yeah, and by the way, we've, we've seen this, right? Like 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.

    13. GB

      Yeah.

    14. DG

      Um, as we've seen with all the, you know-

    15. GB

      Yeah

    16. DG

      ... the consumer internet businesses that we've invested in over the years. Um, and so there may be a disconnect in the period where you can't actually offer that.

    17. GB

      Yeah.

    18. DG

      And that would be a terrible outcome.

    19. GB

      That'd be a terrible outcome for the world. Nobody wants that, so we need to build a lot of data centers.

    20. DG

      Yeah, exactly. Exactly.

    21. GB

      Yeah.

    22. DG

      Yeah.

    23. GB

      Like a compute in- inequality like future, that's, that's not a good, that's not a good future for anyone, which is another reason open source is so important. And just one of the things, um, you know, I u- you know, I had, uh, Grok ma- ma- make me like a meme of that like three-headed dragon, and one of the heads is like kind of confused about like-

    24. DG

      Yeah

    25. GB

      ... all of the really like stupid bearish AI narratives. But people have this idea that open source tokens are free.

    26. DG

      They're not.

    27. GB

      And it's like it takes the exact same amount of compute-

    28. DG

      Yeah

    29. GB

      ... all else equal, to make an open source token as a, you know, frontier token for a comparably sized model. Now, 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-

    30. DG

      Yeah, of course, that are being captured

  8. 34:2140:00

    Orbital Data Centers: The SpaceX Compute Play

    1. DG

      Say we are in this supply crunch. Um, it's so funny when, whenever I talk about SpaceX, and it's, it's obviously near and dear to both our hearts, um, you know, I, I say like, first of all, the orbital data center stuff, it's not like big buildings in space. Like it's helpful to actually think of it. It's like a, the size of an airplane.

    2. GB

      Uh, yeah, people are pic-

    3. DG

      Yeah, like it's like a big rack

    4. GB

      ... people are picturing like the Death Star.

    5. DG

      Yeah, exactly. Like that.

    6. GB

      And like, or the Pentagon-

    7. DG

      Yeah, yeah

    8. GB

      ... floating around in space. That's not what it is at all.

    9. DG

      Yeah. It's a, it's, you know, whatever, the size of an airplane, right? Rack of 72-

    10. GB

      Yeah, but even-

    11. DG

      ... whatever chips, whatever

    12. GB

      ... yeah, it's, it's like five of us standing together is kind of roughly the rack.

    13. DG

      Yeah, and the airplane is like the wings.

    14. GB

      You have the, you have the solar-

    15. DG

      The solar arrays

    16. GB

      ... solar wings.

    17. DG

      Yeah.

    18. GB

      And then you keep it in a sun synchronous orbit, so you have the radiator-

    19. DG

      Yeah, the back end

    20. GB

      ... that's always in the shadow of the rack. That's how you cool it. And it's, I, like I can't... It's very hard for me to engage. You know, there's all these people on X and they're like, "I am a physics PhD, and I, this is impossible." Um, and actually there's, there's, there's a friend who's another investor who actually is a physics PhD, who I had many, um, arguments with him, and he's like, "I am a PhD and this is impossible." And then he goes to the SpaceX Day and, you know, he talks to the SpaceX engineers and he's like, "Well, I was wrong." And so like if, let's say you're an astrophysics PhD, you're 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? Have you thought about it for five hours? 'Cause 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, and the sum of that, working with like very sophisticated, you know, engineering tools, is it's a solved problem, and in their minds it's dramatically simpler and easier-

    21. DG

      Yeah

    22. GB

      ... [chuckles] than a Starlink satellite 'cause a Starlink has to have the phased arrays and move around.

    23. DG

      Yeah.

    24. GB

      Um-

    25. DG

      I, I think it's, uh, like so, okay, so assume that you're right. I, I say it's like physics... There's not a physics reason why this can't work. Cost-wise, it seems really imposing, but kind of the history of the Elon companies is the cost curve gets dramatically better. Like when we first invested in SpaceX, you know, Starlink, uh, like was not commercially available, and Like, we, 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. Like, I, I, I just have to think that that will get solved paired with the fact that we're gonna have massive undersupply, self-inflicted, on Earth.

    26. GB

      Yeah.

    27. DG

      Uh, it feels clear to me at a minimum it will be swing capacity.

    28. GB

      Yeah.

    29. DG

      And, you know, in the fullness of time, maybe it will be larger.

    30. GB

      Well, no, it's really simple. Like, if we use 50-- Uh, and it, and it is the people... The question people should be asking about orbital compute, which is the one SpaceX is focused on, is Starship reusability.

  9. 40:0044:04

    Starlink's $2 Trillion Market & the Heads-You-Win Compute Bet

    1. DG

      mobile plus your broadband, whatever, it's, call it, like close to $2 trillion of a market.

    2. GB

      And then you have a really rapidly growing AI ARR base.

    3. DG

      Yeah, AI ARR. You've got the cloud, you know, the sort of neo cloud business.

    4. GB

      Yeah, the cloud. Yeah. Um, so I don't think... Great, you're an orbital compute skeptic? No problem. It doesn't matter.

    5. DG

      Yeah, exactly.

    6. GB

      We don't even need to... We can just look at things that are happening today with-

    7. DG

      Yeah

    8. GB

      ... terrestrial compute, with Cursor, with Grok, with GrokBot. By the way, I think X ads are... You know, we have telemetry.

    9. DG

      Yeah, they're going well. Yeah.

    10. GB

      They're also growing. You know, I would expect at some point you'll have like a Starlink GrokBot, um, [clears throat] X advertising bundle. You know, kind of one of the ways Google built their-

    11. DG

      Yeah. Yeah

    12. GB

      ... cloud business is they bundled it with ads and like, "Hey, we're..." You know, maybe you're bundling the ads with AI, but-

    13. DG

      Yeah. Yeah

    14. GB

      ... why not do that?

    15. DG

      Yeah. Yeah, I actually like the AI position that they're in because it's like heads you win, tails you win, in the sense that their first party business is growing very fast, and they, they caught up to the frontier, like very quickly.

    16. GB

      Yeah.

    17. DG

      Um, and so they've made the very aggressive compute investments to enable that first party work.

    18. GB

      Yeah.

    19. DG

      Um, and that's the kind of heads you win and like tails you win. Say they overbuilt their capacity for what they need for inference or training, they have a-

    20. GB

      Sub six-month payback

    21. DG

      ... very compelling sub six-month payback on the compute side, um, you know, with like massive scarcity of, of supply.

    22. GB

      Yeah.

    23. DG

      And so I, I think that's a really good setup.

    24. GB

      And there was a bear case that, hey, okay, well, in, in a, in the OpenAI anthro-anthropic maximalist view, where they're the only two companies and they're designing their own chips, then like where, what's the room for anyone else? Well, like I don't think they're gonna have a reusable starship and multiple spaceports anytime soon. And if the economics of com- of compute are such that orbital is where it makes sense increasingly going forward, 'cause starships should be deflationary, you know, terrestrial cooling, you know, power should be inflationary. Well, like even in, in a world where they fumble the ball with their first party AI applications, like they do still have like a-

    25. DG

      Yeah, then they're a massive infrastructure business.

    26. GB

      Yeah.

    27. DG

      Yeah. Yeah, I, I'm, I'm so fired up about the, uh, the star base Louisiana. Um-

    28. GB

      Oh, yeah. [chuckles]

    29. DG

      It's [chuckles] like-

    30. GB

      I can't wait to visit, man.

  10. 44:0454:01

    The Most Futuristic SpaceX Idea: Asteroid Mining

    1. GB

      but I mean-

    2. DG

      What's... Yeah. What's the... Okay, so SpaceX, like, again, you and I have talked a ton about SpaceX. What's, like, the most futuristic thing that you think about with SpaceX? Like the 10-year k- like, okay, so you and I were at this conference together, and there was this whole debate about, um, among a small group of public investors of, like, what's gonna be the, the first $10 trillion company, and, uh, I think what you said was, like, "I have no idea, but I know which one's gonna be the first $20 trillion company." Uh, [chuckles] so, like, what's the most futuristic, like, product or market or technology thing about SpaceX that, that you can think of?

    3. GB

      Look, I mean, this sounds crazy, but asteroid mining is gonna be a very real thing. We're gonna capture, you know, there's asteroid Psyche. It has more gold, silver, platinum, you know, every precious metal in it that exists in the Earth's crust. At some point, particularly with Starship, it, you will be, you know, and w- we may need that, um, lunar base to make this happen. You'll be able to cap- capture these asteroids. You'll bring them into a stable kind of geosynchronous orbit over some, you know, American-owned atoll in the middle of the Pacific. Um, you know, no humans within, whatever, 50 miles. You'll, you know, you can imagine, like, Optimus robots, you know, um, you know-

    4. DG

      Yeah, doing the, doing the work, yeah

    5. GB

      ... yeah, doing the work. Um, and then, you know, delivery to Earth is free, and for sure some of it's gonna burn up. But I think th- that's gonna happen.

    6. DG

      Yeah.

    7. GB

      And I always think, um, J- Jeff Bezos said something very interesting. He said, "I think in the future Earth is going to be zoned residential." And, you know, somebody asked him-

    8. DG

      Mm-hmm

    9. GB

      ... this was, like, 15 years ago, "What do you mean by that?" He's like, "All heavy industry will take place in outer space." And then this addresses the pollution concerns.

    10. DG

      Yeah.

    11. GB

      It addresses everything. You know, people always get, like, really worried about, oh, you know, will we still be able to see the stars? And it's just like, I think it's hard for, like, the human mind to understand how big space is.

    12. DG

      Yeah. Yeah. [laughs]

    13. GB

      How big outer space is.

    14. DG

      Yeah, yeah, yeah. Yeah.

    15. GB

      You know? It's like-

    16. DG

      We don't have to worry so much about emissions up there. Yeah. [chuckles]

    17. GB

      Yeah. Yeah. So I think that is, um...

    18. DG

      That's probably the most futuristic thing you-

    19. GB

      But in terms of an economic application. But it does, um, [clears throat] I mean, I, I do think in the next few years you're gonna have a fleet of Starships land on Mars. Next few years, I mean, I don't know. Let's just say at the outside this is eight years away.

    20. DG

      Yeah.

    21. GB

      They're gonna land on Mars. They're gonna have, like, you know, a little ramp's gonna come out of the Pez dispenser, and it's gonna be a modified Starship, the Mars Colonial Transporter, and it's gonna be wild. You're gonna s- have Optimus robots holding American flags-

    22. DG

      [laughs]

    23. GB

      ... like, walk down, and then, you know, they're gonna pull out a bunch of solar panels and batteries and racks of compute, and they're gonna set all of that up. They'll be dropping Starlinks, you know, and, and maybe the orbital mechanics don't allow this, but I think, you know, they'll, they'll, they will figure out a way to have, you know, capacity. So just think how crazy it is to watch, like, the views from Pathfinder-

    24. DG

      Uh-huh

    25. GB

      ... you know, or, you know, whatever these different, you know, Mars, um-

    26. DG

      Yeah, rovers and stuff

    27. GB

      ... rovers are. And, like, you know, 4K video through Optimus robots all over Mars, and then after that, there will be humans.

    28. DG

      Who can inhabit it. Yeah. Yeah, yeah.

    29. GB

      Yeah.

    30. DG

      Yeah, that is crazy to think about.

  11. 54:0157:57

    Who Becomes the Abstraction Layer of Intelligence?

    1. GB

      I was-

    2. DG

      It sounds easy to describe. Like, the way I describe it to people is like, who gets to be the abstraction layer to the organization and the users with intel- like, uh, of, of intelligence.

    3. GB

      Yeah.

    4. DG

      It's like the most v- v- whatever, vied after space-

    5. GB

      Yeah

    6. DG

      ... or position that you could imagine in business, like in the history of business.

    7. GB

      Yeah, for sure.

    8. DG

      Right? I think it's like the answer is and again. Like-

    9. GB

      Yeah. Yes, and for sure. It's, yeah, who's the arbiter of intelligence-

    10. DG

      Yes

    11. GB

      ... for global enterprises-

    12. DG

      Yeah

    13. GB

      ... and probably consumers. I was a retail analyst, and, um, you know, everybody kinda thinks running one of these big chains is easy and there's a lot into it, and it's like, well, it's really easy. Start an American retailer in any category, 'cause America's so big, that's worth over $50 billion. Almost any category.

    14. DG

      Yeah.

    15. GB

      All you have to be able to do is have a fleet of 1,000 stores-

    16. DG

      [chuckles]

    17. GB

      ... in 50 different states that have very different climates, consumer preferences. You need to have them stocked with the right products at the right time for that region, at the right prices. They need to be staffed by friendly and knowledgeable employees who don't steal from you.

    18. DG

      Who turn over at 100% a year.

    19. GB

      Who turn over at least 100% a year. The stores need to be clean and well-lit. And if you can do that Presto, $50 billion.

    20. DG

      Yeah.

    21. GB

      And like in the history of American business, like you can... I mean, it's more than one hand, but you don't have to-

    22. DG

      That's a ton

    23. GB

      ... go through many.

    24. DG

      Yeah.

    25. GB

      It's really hard to do.

    26. DG

      Yeah.

    27. GB

      And having that abstraction layer, having it work, having it seamless, is I think way harder to do than people think. And I do think, well, something I think is very interesting about Cursor, I'd love your opinion on this, is like everybody else in the lab space, you know, had this like we're creating a digital deity, you know, and AGI and ASI.

    28. DG

      Yeah, yeah, yeah.

    29. GB

      Like we're... And the Cursor guys were just like, "We want to make great product."

    30. DG

      Yes, exactly.

  12. 57:571:00:26

    Harvey, Cursor & Vertical AI Winners

    1. GB

      Yeah.

    2. DG

      Right? So like Harvey has done an incredible job of this.

    3. GB

      Yeah.

    4. DG

      And, you know, like legal has sort of been takeoff and-

    5. GB

      Yeah

    6. DG

      ... um, and, and I think they can see the future of how to be that abstraction layer, um, and do the work. Um, but like le- legal is also unique because it's very documented.

    7. GB

      Yeah, and tax.

    8. DG

      And it's somewhat verifiable.

    9. GB

      We'll see tax.

    10. DG

      Tax. We'll see that.

    11. GB

      Similar.

    12. DG

      We'll see, see things like that.

    13. GB

      Yeah.

    14. DG

      But like the, the one and a half billion, the really appealing broad pie is gonna be very messy to go get.

    15. GB

      Yeah, although I do always think, and um, you know, I think probably in their heart of hearts, Harvey and Lagora think, "Oh, if we solve this, we could be that abstraction layer for everyone."

    16. DG

      Yeah.

    17. GB

      I think probably in their heart of hearts, Cognition thinks something like that too. Um-

    18. DG

      I think everybody thinks it. And by the way-

    19. GB

      Everybody thinks they're going after it

    20. DG

      ... this is like massive validation of the category.

    21. GB

      Yes.

    22. DG

      Because Kirkland & Ellis said, "We're gonna spend 500 million bucks to build this ourselves." Like, first of all, you know, like good luck. That's gonna be very hard.

    23. GB

      Yes.

    24. DG

      Um, but that actually tells you that the pie is really big, right?

    25. GB

      Oh, huge.

    26. DG

      Yeah. It's massive.

    27. GB

      And that's, it's... And, you know, just um... And I'm sure they, they have a very smart head of AI, head of AI, but it's not like a $500 million one-time build. That model has to be continuously updated-

    28. DG

      Constant build

    29. GB

      ... switching out the base model, and all of that has to happen transparently. But I think you're gonna have this huge collision between, you know, products like Fireworks, Nexus, these legal agents, coding agents, big companies like Microsoft.

    30. DG

      Databricks.

  13. 1:00:261:12:16

    Jensen, Nvidia & the Central Bank of AI

    1. DG

      [chuckles] Um, so okay, you, you mentioned Jensen. You know, I, I share your sentiment, like he's like carrying this industry forward. Like tell me your thoughts on Nvidia.

    2. GB

      So, um, I, I think he's in a very, very good position in his strategy of being vertically integrated but horizont- horizontally open. And it's like, okay, like let's just say, um, you know, let, let, let's say there's some accelerator that emerges that is really, really, really, really good. Almost certainly it will be better if it can plug into, and this is why like I know you have an accelerator investment. My number one thing is if you're a semiconductor CEO, the only thing you should ever say is, "Thank you, Jensen. Thank you for creating this opportunity. Thank you. How can we work with you? We wanna enable you. Sure, we're gonna compete with you on the edges."

    3. DG

      Yeah.

    4. GB

      But you know, my rule of thumb for accelerators, every 1% share today is probably worth 100 billion.

    5. DG

      Yes.

    6. GB

      So there's no need to go head on with Nvidia.

    7. DG

      Yeah.

    8. GB

      Um, just pick a niche, get your 1%, make sure that, you know-

    9. DG

      That, that pie is very big

    10. GB

      ... he has, he has nine chips.

    11. DG

      Yeah.

    12. GB

      Um, you know, he's got, [chuckles] he's got multiple flavors of accelerators. He's got CPUs. He's got, you know, Ethernet switches. He has two kinds of DPUs. You know, he's got... You know, we've gone from just, um, scale-out networking being a thing. We have scale up, scale out, scale across, now scale in.

    13. DG

      Yeah.

    14. GB

      So just try to find a way to plug into his ecosystem.

    15. DG

      By the way, this is not foreign. Like his biggest customers all have competing products-

    16. GB

      Absolutely

    17. DG

      ... with various of those nine chips.

    18. GB

      Yeah. And just try to find a way to plug in, but just be nice to him. Be nice. Be nice.

    19. DG

      [laughs]

    20. GB

      It's all personal.

    21. DG

      Yeah.

    22. GB

      You know? And it's just like sometimes, like, you know, you hear some of these and it's like, have you ever seen game tape of the Chicago Bulls when Jordan was... It was, you know, it's game 50 of the season.

    23. DG

      Yeah.

    24. GB

      And he's a little bored.

    25. DG

      Yeah.

    26. GB

      And the Bulls are down 'cause, you know, they're up eight games. You know, they're up eight games-

    27. DG

      Yeah. Yeah, yeah

    28. GB

      ... over the number two person in their conference, and he's a little bored. And then somebody-

    29. DG

      Somebody talks shit

    30. GB

      ... somebody who's you, who's, who's kind of young, decides I'm gonna talk shit to him 'cause we're beating him, and then he just looks.

  14. 1:12:161:14:11

    How Chip Deal Structures Reveal True Customer Preference

    1. DG

      Right.

    2. GB

      Because, like, you come out with a-

    3. DG

      Yeah, they'll take anything. Yeah, that's-

    4. GB

      You-

    5. DG

      This is how you know that, like-

    6. GB

      Yeah

    7. DG

      ... the very old gen, whatever, the price is held up of H100 is very high.

    8. GB

      Yeah, yeah. And if you have a TSM allocation, you're gonna be sold out.

    9. DG

      Yes.

    10. GB

      Particularly if you can get the DRAM to pair with it.

    11. DG

      Yeah.

    12. GB

      You're gonna be sold out. So it's actually kind of hard to infer true customer preferences, and I actually think one of the best ways you can, like, see true customer preferences is the kind of deals they cut with chip companies. So broadly speaking, you know, the first deal is where the chip company invests-

    13. DG

      Yep

    14. GB

      ... in a customer, and you saw TPU and Trainium, Amazon and Google do that with Anthropic.

    15. DG

      Yep.

    16. GB

      And that was to their admi- immense advantage 'cause it really helped their businesses, I think helped those chips really level up. 'Cause you kinda need to use a chip.

    17. DG

      Yeah. Yeah, yeah.

    18. GB

      There's a cold start problem. [clears throat] And, um, and in that scenario, as long as the dollars you invest are less than the gross profit, you can't lose money. And then there's the scenario where you do the RVG, Blackstone finances it, or whoever, Blackstone, Apollo-

    19. DG

      Yeah, yeah

    20. GB

      ... KKR, Goldman Sachs finances it. Um, and as long as that RVG is actually less than your gross profit, you can't lose money.

    21. DG

      You're fine. Yeah, you're fine.

    22. GB

      And you have upside probably through a revenue share on top of it. Then there are deals where you give warrants away, but they're tied to, um, like a fixed price per million tokens.

    23. DG

      Yep.

    24. GB

      And as long as the performance of your chip kind of outruns the performance of your stock, you're gonna do good in that situation.

    25. DG

      Yeah, that's, that's valuable. Yeah.

    26. GB

      If you just give warrants away, it could be negative NPV-

    27. DG

      Yeah, yeah. For sure

    28. GB

      ... 'cause the better the stock does-

    29. DG

      Yeah, the more value that's captured by the person. Yeah

    30. GB

      ... the worse the deal is. Yeah. And so you can kinda look at that hierarchy of deals and, like, infer something about true customer preferences.

Episode duration: 1:14:25

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode FGC4ofTcg2k

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.