Gavin Baker and David George on Positional Strategy in AI

Live from Runtime by a16z

This conversation took place last week at Runtime, presented by a16z. Our own David George sat down with Gavin Baker of Atreides Management, LP, and talked about the positional strategy game taking place between the hyperscalers and AI model companies. The conversation has been lightly edited for clarity. Enjoy!

The big question: is AI a bubble?

David George (a16z): All right. So the big topic is, is AI a bubble?

Let’s take a macro view of things, and start with a couple statistics to set the stage, and then Gavin, I want your take on where we’re at.

So, we have about a trillion dollars of data centers in the US. Over the past three years, we have already built out more in data center capacity, in inflation adjusted dollars, than the entire US interstate highway system, which took 40 years. And our plan is to add $3 to $4 trillion dollars more over the next 5 years. Open AI alone, I believe, has committed to more than a trillion dollars of deals set up. All of these big numbers scare people, and they tell people, “Bubble!”.

So. Are we in an AI bubble?

Gavin Baker (Atreides Management): I do not believe we’re in an AI bubble today. I had, depending on how you look at it, the privilege and the misfortune of being a tech investor during the year 2000 bubble, which was really a telecom bubble. And I think it’s helpful to compare and contrast today to the year 2000. Cisco peaked at 150 or 180 times trailing earnings, NVIDIA is at more like 40 times. So valuations are very different. More importantly, however, is that the internet bubble or telecom bubble in 2000 was defined by something called dark fiber.

If you were around in 2000, you’ll know what that was. Dark fiber was literally fiber that was laid down in the ground, but not lit up. Fiber is useless unless you have the optics and switches and routers that you need on either side. So I vividly remember companies like Level Three or Global Crossing or WorldCom would come in and they say, “We laid 200,000 miles of dark fiber this quarter. This is so amazing. The internet’s gonna be so big. We can’t wait to light these up.”

At the peak of the bubble, 97% of the fiber that had been laid in America was dark. Contrast that with today, there are no dark GPUs. All you have to do is read any technical paper, and understand that the biggest problem in a training run is that GPUs are melting from overuse.

. . . the internet bubble or telecom bubble in 2000 was defined by something called dark fiber . . . Dark fiber was literally fiber that was laid down in the ground, but not lit up. Fiber is useless unless you have the optics and switches and routers that you need on either side.

So I vividly remember companies . . . would come in and they say, “We laid 200,000 miles of dark fiber this quarter. This is so amazing. The internet’s gonna be so big. We can’t wait to light these up . . . Contrast that with today, there are no dark GPUs.”

And there’s a simple way to kind of cut to the heart of all of this.

It is to look at the Return on Invested Capital of the biggest spenders on GPUs who are all public. And those companies, since they ramped up CapEx, have seen around a 10 point increase in their ROICs. So, thus far, the ROI on all the spending has been really positive.

There’s an interesting and open debate about whether or not it will continue to be positive, with how much spend we’re going to have on Blackwell. (I personally think it will.) But there’s no debate that thus far, the ROI on AI spend has been really positive. And valuation wise, we’re just not in a bubble.

David: The other thing that I would say is you can contrast the actual adoption and usage of the technology versus back then.

With building out the internet, you had to build a two-sided network. You had to build websites and then you had to get users. Whereas with AI tools, all you have to do is just light ‘em up, turn on your website, and everybody has access to them. They’re built on cloud computing, on internet infrastructure, and you can get instant distribution to a billion people right away.

The other thing is the counterparties. These companies happen to be some of the best companies in the history of the world, and collectively, these people writing checks for all of this Capex out of pocket, they collectively generate around $300 billion of free cash flow a year. And they have around $500 billion of cash on their balance sheets. So when people are like, “Oh my God, it’s a bubble, is it going to pop?” I think it’s kind of fine. There’s an $800 billion buffer, growing $300 billion every year.

Gavin: Yeah. Free cash flow at some of them has begun to, well, you know. This goes to your point on Return on Invested Capital. It might, we should see that next year creep down a little bit. But you know, Larry Page apparently said internally, I’m happy to go bankrupt rather than lose this race. And I think that is the mentality at Google and perhaps Meta: it’s just seen as existential, and that you have to win.

You can watch David & Gavin’s conversation from Runtime here.

“Round tripping” as competitive strategy, not financing

David: Okay. So, lots has been written about these round tripping deals, which is understandably a scary concept.

In the internet build out, that was a big problem. What do you make of it here?

Gavin: So here the round tripping, it is objectively happening, although at small scale. Money is fungible, after all. What is driving this “round tripping” isn’t the need to finance GPU or data center purchases. It’s actually competitive dynamics. So Nvidia’s biggest competitor isn’t AMD, it’s not Broadcom. It’s certainly not Marvell. It’s not Intel. It’s Google. And more specifically, it’s Google because Google owns the TPU chip. And - today at least - it’s the only alternative to Nvidia for training. And maybe the best inference alternative.

And Google’s a problematic competitor. Because they also own a company called DeepMind and they have a product called Gemini. And I think you could argue that they’re the leading AI company today. I think they’ve taken 15 or 20 points of traffic share in the last two or three months. That’s just traffic to Gemini; it does not include search overviews. I suspect on an actual traffic basis, Google is bigger than OpenAI, Anthropic, anyone today. And that business is gonna run on TPUs.

And then we have three other labs that are relevant today. There’s Anthropic; that’s an Amazon and Google captive. Anthropic is really going to run on TPUs and Trainiums. And so you’re left with xAI and OpenAI at the forefront. And if Google is going to a lab like Anthropic and saying, “I’m going to help you fundraise and give you chips, for competitive reasons, it’s very hard for Nvidia not to respond. And as Jensen said, he thinks it’s going to be a good investment.

What is driving this “round tripping” isn’t the need to finance GPU or data center purchases. It’s actually competitive dynamics. . . . So I think the round tripping concerns are pretty overblown. What Nvidia really needs is for Meta to get their act together, or another American open source player to emerge, or maybe some sort of detente with China in AI.

So I think the round tripping concerns are pretty overblown. What Nvidia really needs is for Meta to get their act together, or another American open source player to emerge, or maybe some sort of detente with China in AI.

David: Yeah. When people ask me about Nvidia and round tripping, my reaction is, “Everything they’ve done is completely rational.”

Long term, that is. Some of the things they do may not have as high of a Return on Capital as other things, but strategically, I think they’re all kind of the right moves.

Gavin: Jenssen’s one of the two best CEOs along with Elon I have ever known. And I think he’s playing a strong hand really well.

Market structure is changing

David: Alright, so you started getting into the model companies. What do you think happens with market structure? Who wins, where?

Who are you most optimistic about? Where do you have concerns?

Gavin: I think humility is an important virtue for an investor. And, if we’re gonna make an analogy and say that ChatGPT is to AI as Netscape Navigator was to the Internet: at this point in the internet boom, Google had not been founded. Mark Zuckerberg was in middle school. Travis Kalanick was in kindergarten. So it’s just very early. So I think it’s important to be humble about making high confidence predictions at the application layer. It’s one reason I think the infrastructure layer is often a safe place to be at the beginning of one of these new technology waves.

I think it’s hard to have high conviction other than to observe: the internet was a very disruptive innovation. I think there’s reasonable arguments that AI could be a sustaining innovation because the raw ingredients of data, the capital to buy compute and distribution, which is what you need: all of the big tech companies have all of those in spades.

So as long as [Big Tech] execute well, hire good people, and have a sound strategy, I think you could see it be a sustaining innovation for a lot of members of the Mag7. On the other hand, I do think it’s existential. And if you don’t execute, you know, IBM might be a good fate.

David: Yeah, that’s tough. Data, distribution, compute dollars, talent; and they have every right to win. And it seems now more than before, they’re taking it quite seriously. Yeah. Maybe Google in particular; and obviously Meta is making the dramatic moves they’re making.

What’s your forecast for the platform piece of their business, the infrastructure piece? How do you think it shakes out in terms of business model; market structure? Do you think they end up as a high margin businesses, like the clouds or like aircraft manufacturers, or do you think they end up very competitive in low margin businesses like airlines?

Gavin: I don’t think they’ll be airlines. But look, anybody can just look at the P&L of a SaaS company circa 2021 and 2022 and you see 80, 90% gross margins. And the nature of AI, because of scaling laws; “the bitter lesson”, they’re just more compute intensive. So their gross margins are structurally going to be lower, but that doesn’t mean they can’t be great businesses. It’s going be a long time before we see a frontier AI lab with gross margins anywhere near SaaS. Or internet era margins.

Now their opex can be a lot lower, and maybe that’s how you square it. But the gross margins are fundamentally different. And until scaling loss changes, and the importance of Test Time Compute changes, which I don’t see happening, they are gonna be lower margin.

Don’t be embarrassed by shrinking gross margins!

David: Okay. So let’s talk about the application layer. Every few months there’s a fight on Twitter, that SaaS is terrible, and it’s dead, and it’s all gonna go away.

What do you think happens with SaaS and software?

Gavin: You know, I probably first said in early ‘24 that I thought all of Application SaaS might be a zero. That’s different from infrastructure SaaS. I would say I have a more nuanced view now, and I think there could be some really big application SaaS winners, especially if you serve a more fragmented SMB customer base. Google is making it really easy, if you are a customer of theirs, to use your data and essentially make any SaaS app you want, and then your data isn’t shared with anyone else.

But the critical mistake that I think a lot of retailers made in dealing with Amazon is, they looked at Amazon’s margins and they said, “We don’t want to be in that business.” And that was obviously a terrible mistake. And here we are 25 years later and, Amazon has really healthy retail margins. And I worry that Application SaaS companies are trying to preserve their existing gross margin structures, because they believe that if their gross margins go down, their stocks will go down.

It is definitionally impossible, given what we just discussed, to succeed in AI without gross margin pressure. And I do not know why they have concerns, because we have an existence proof that a software company can deal well with declining margins, in Microsoft (and Adobe ‘till the whole AI thing came along.)

You know, it used to be that companies were scared to go from on-premise to the cloud ‘cause margins were lower. Cloud margins are lower, but they’re still good. And Microsoft, they transitioned from on-premise, perpetual licenses with maintenance, to a cloud model and it was a pretty good stock for 10 years.

So if you are in an application SaaS company, don’t be scared. And look at declining gross margins as a mark of success, rather than as a badge of shame.

David: It’s actually so funny you say that, because whenever we have these discussions about companies, basically every company that comes to present to us is like, “We’re an AI company.”

And we always look at the gross margins, and it’s become a badge of honor for them to actually have low gross margins, because people are actually using their AI stuff! But if you show up and you’re like, “I’m an AI company, and I’ve got 82% gross margins” you’re like, “I don’t think anybody’s really using it.”

So yeah. If you’re one of these public companies, would you rather have 10 bucks of revenue with 90% gross margins, or 50 bucks of revenue with 60% gross margins? It’s not that complicated.

Gavin: It’s hard to do in public markets, but when you communicate it, draw parallels to the cloud transition. I mean, I’m an investor, and I would be excited about it! And I don’t think I’m alone in the world.

And the big advantage these legacy application SaaS companies have, is they do have these really profitable existing businesses. And so they can run new AI products at breakeven, and catch up to the leaders. I’m surprised more people have not done that.

Like, why are none of the public coding companies even trying to compete with Cursor? And the reality is, Cursor now they have a trillion tokens, and there will be a point where they have enough coding tokens that it’s tough to catch ‘em. But today, if you’re a public coding company and you said, “I’m gonna lean in, I’m gonna run it break even, I’m gonna attach it to everything.” Hey, you have a chance. And you know, the prize is clearly really big.

David: It’s like Dumb and Dumber: “So you’re telling me there’s a chance.” And there are examples.

Figma for example, they are an extremely high gross margin business, and they said, “Hey, we’re gonna, pretty aggressively distribute our AI tools, and our gross margins are gonna go down.” And then investors asked a few clarifying questions, and then realized, “Oh, that would be a good thing.” So I’m surprised more people in the public markets aren’t doing it. It’s a long game to play, but it’s working out well.

What about on the consumer side, the application layer?

Obviously Google was the portal to the internet, and it still kind of is the portal to the internet, and the whole business model was predicated on taking some intent, and directing you to someone’s website, where they would do stuff with you. It’s not going to be that way anymore. Although, I tried the new browser today and I tried to do some basic shopping, and, there’s still some work to do. But I think it will get there.

So what do you actually think happens with the market structure of the consumer internet companies? Do they get subsumed into a component of a chatbot interface, or do you think it’s something else?

Gavin: So, humility; hard to say. I think the AI companies that have launched these AI browsers may come to regret it. Cause there’s something called Chrome that has, whatever it is, 5 billion users. And if you’re Google, you know know, you can look at what happened with Google Buzz. Google has to be very cautious; they’re currently in litigation with the government. And they could easily do this probably do it even better, but they didn’t want be first.

So now you have two AI-native companies with their own browsers. Let ‘em run for three to six months, get a little headstart, and then, “Wow, here we are, we had to do this.” And I don’t know how that’s gonna work, for the companies other than Google who don’t own Chrome. The data and distribution is pretty powerful.

I do think it’s tough to bet against the companies with large existing user bases today.

And I also think reasoning has fundamentally changed the economics of these frontier models. Pre-reasoning, I often said, “if you are a frontier model without access to unique, valuable data and internet scale distribution, you’re the fastest depreciating asset in history.” I think reasoning really changed that. Because the of way Reinforcement Learning works during post training, having a big user base now unlocks the flywheel that was at the center of every great consumer internet company.

You have a good product, you get a lot of users, the users make the algorithm better, the algorithm makes the product better, and it just spins. It’s not quite spinning yet in AI, but you can squint and see it.

Pre-reasoning, I often said, “if you are a frontier model without access to unique, valuable data and internet scale distribution, you’re the fastest depreciating asset in history.” I think reasoning really changed that . . . You have a good product, you get a lot of users, the users make the algorithm better, the algorithm makes the product better, and it just spins

. . . that fundamentally changes the economics for Anthropic, for xAI, for OpenAI . . . In a strange way, the Chinese open source model ecosystem is a godsend to any American company that’s trying to catch those four leading labs

So I think that fundamentally changes the economics for Anthropic, for xAI, for OpenAI. But Mark Zuckerberg’s trying hard. A lot of smart people are. In a strange way, the Chinese open source model ecosystem is a godsend to any American company that’s trying to catch those four leading labs.

Because the problem is, if you don’t have Gemini 2.5 Pro or a later checkpoint of it, or a later checkpoint of Grok that we don’t see, or a later GPT checkpoint… training the next model, you’re at a big disadvantage.

By the way, one thing that drives me crazy is all these people who say that GPT-5 is the end of scaling laws. GPT-5 is a smaller model. It was not designed to be better! It was designed to be more economical for OpenAI and Microsoft to run it. Its scaling laws is crazy.

Dispatch from the chip wars

David: We’ll get the pedestal up here if you want. So, do you want to talk about chips? I know you love Nvidia. Talk about your view of Nvidia, AMD, TPUs, ASICs.

How do you think market structure shakes out the competitive advantage that the various players have?

Gavin: I think it is really a fight between Nvidia and the Google TPU. And then something that I don’t think is broadly appreciated is the extent to which Broadcom and AMD are effectively going to market together.

Nvidia is no longer just a semiconductor company, as I’m sure you’ll hear from Jensen tomorrow. It was a semiconductor company, then a software company with Cuda, now a systems company with these rack level solutions. And now arguably, a data center-level company with the level of architecting they’re doing with scale up, scale across and scale out networking.

So the networking, the fabric, the software, it’s all important. And what Broadcom is saying to companies like Meta is, “Hey, we will build you a fabric that can theoretically compete with Nvidia’s fabric, which is a mixture of NVLink and either Infiniband or Ethernet. We’ll build it on Ethernet, it’s gonna be an open standard. And we’ll make you your version of TPU, which took Google three generations to get working. And you know what, if your ASIC isn’t good, you can just plug AMD right in.”

I personally believe most of those ASICs are gonna fail, particularly in the fullness of time. I think you’ll see a bunch of high profile ASIC programs canceled, especially if Google starts selling TPUs externally.

I don’t think is broadly appreciated is the extent to which Broadcom and AMD are effectively going to market together . . . what Broadcom is saying to companies like Meta is, “Hey, we will build you a fabric that can theoretically compete with Nvidia’s fabric, which is a mixture of NVLink and either Infiniband or Ethernet. We’ll build it on Ethernet, it’s gonna be an open standard. And we’ll make you your version of TPU . . . And you know what, if your ASIC isn’t good, you can just plug AMD right in.”

And who knows exactly how that would work? Cause if you’re Anthropic, who rumor has it wants to buy tens of billions of TPUs, maybe you don’t want Google seeing your secret sauce, but there’s ways around that. So I think this is really a battle between Google and its TPU - enabled by Broadcom for now - and Google can take the TPU away from Broadcom whenever they want.

Now, they can’t do the ethernet networking that Broadcom is doing. But they control the TPU. So, it’s really Google and the TPU v. Nvidia, with Amazon as well. The Annapurna team, that’s a very talented team. I think the Trainium 3 will probably be a much better chip than the Trainium 2. It took Google three generations to get the TPU right. And then AMD will always be kind of the second source, and you need a second source.

Business model degradation and disruption

David: Exciting. So I wanna go back to business models. One of the big things that is widely discussed is “source of disruption.”

And most of the CEOs in this room are CEOs of startups who are trying to go beat some incumbent, or find some new market opportunity. And the most ripe opportunities tend to come when you have a big platform shift that is also accompanied with a business model shift.

So there are a couple of areas where I can see it in an obvious way. We’re investors in Decagon customer support; you can pretty easily see a business model that is priced on the resolution of a task, because it’s so measurable. And similarly in coding, a lot of the business model has now shifted to consumption. For developer-facing tools that’s comfortable and pretty well-known.

What about the rest of the industry? Because I feel like there’s sort of this handwave going on, which is, “We’re gonna go get all of services.” But okay, so how do you actually go do that? It’s gonna be pretty hard. Do you have any prediction on how that plays out?

Gavin: Customer service is kind of an easy first example: we have a lot of textual data, and LLMs are good at text. And you can run some RL to make sure that they get a good verified reward, on “happy customer”, first call resolution, or whatever it is. And, I mean, humans are paid based on outcomes. And a lot of AI will be augmenting humans, but probably also replacing some humans. And that will involve being paid for outcomes.

Going back to the consumer business model, everybody’s talking about affiliate fees and for sure, the next time I want to go on vacation, I’ll have an AI that will know the hotels that I like to go to and it’ll say, “Hey, three hotels, I have Gavin coming, who’s got the best price and the best room? It’ll really upgrade the gifts that you get, and so forth.

This will be a little bit of a business model degradation.

Why did Google never start a marketplace? Because people overvalue, systematically, their ability to keep an organic customer once they’ve acquired it through Google. So they systematically overpay and they continue doing that. That’s why Google never went to outcomes, or a marketplace, because advertising leads to advertisers systematically overpaying. So that inefficiency will be squeezed out when we go to outcomes.

This will be a little bit of a business model degradation. Why did Google never start a marketplace? Because people overvalue, systematically, their ability to keep an organic customer once they’ve acquired it through Google. So they systematically overpay and they continue doing that.

. . . Google never went to outcomes, or a marketplace, because advertising leads to advertisers systematically overpaying. So that inefficiency will be squeezed out when we go to [monetization based on] outcomes.

You know, I think Elon tweeted today that “work would become optional.” Like how, instead of buying your vegetables at a supermarket, you can grow your own garden if you want. That kind of optional.

Now, who knows how long it takes us to get there. But that doesn’t sound wildly implausible to me for how powerful this technology is. And I was just struck about how Karpathy, just a couple days ago, is being painted as like a skeptic for saying AGI is 10 years away. Are you kidding? That’s insane.

David: “Shorter timelines, please!” Yeah, I mean, that’s awesome.

The future is humanoids (or not)

David: While we’re on the topic of exciting futuristic things, robotics: do you have a view?

Gavin: Yeah. Very real. And it’s gonna be Tesla versus the Chinese in the same way that it’s Tesla versus the Chinese in electric cars. Just cars, even, not just electric.

You can all watch the Optimus videos. Every roboticist I know is extremely impressed. There’s a giant debate: “Is it gonna be humanoids or not humanoids?” I think that debate is over because humanoids can learn from watching YouTube videos.

And it’s easier for a human being to, for example, put on a suit and show the robot how to do it. It’s crazy to watch the video of the 50 Optimus robots doing 50 different tasks, and it’s pretty simple: “did you put the glass in the dishwasher correctly, or not?”

There’s a giant debate: “Is it gonna be humanoids or not humanoids?” I think that debate is over because humanoids can learn from watching YouTube videos . . . It’s crazy to watch the video of the 50 Optimus robots doing 50 different tasks, and it’s pretty simple: “did you put the glass in the dishwasher correctly, or not?”

For some inside baseball: I have been told on good authority, - although not from Elon himself - that for a long time he was work working on the hand, which was the toughest engineering problem. And he then shifted his focus to making sure that Optimus could not outrun a human being. From, you know, a mechanical perspective. Because he didn’t want to be the person that brought Terminators that could run more than 10 miles per hour into the world.

David: So he’s programming robotically that old joke, “You don’t have to be faster than the bear, you just have to be the not the slowest?”

Anyway, with that optimistic note, let’s wrap up. This is so fun, Gavin. I always love chatting with you. Thank you.

---

Views expressed in “posts” (including podcasts, videos, and social media) are those of the individual a16z personnel quoted therein and are not the views of a16z Capital Management, L.L.C. (“a16z”) or its respective affiliates. a16z Capital Management is an investment adviser registered with the Securities and Exchange Commission. Registration as an investment adviser does not imply any special skill or training. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell — or a solicitation of an offer to buy — any securities, and may not be used or relied upon in evaluating the merits of any investment.

The contents in here — and available on any associated distribution platforms and any public a16z online social media accounts, platforms, and sites (collectively, “content distribution outlets”) — should not be construed as or relied upon in any manner as investment, legal, tax, or other advice. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Any charts provided here or on a16z content distribution outlets are for informational purposes only, and should not be relied upon when making any investment decision. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, posts may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein. All content speaks only as of the date indicated.

Under no circumstances should any posts or other information provided on this website — or on associated content distribution outlets — be construed as an offer soliciting the purchase or sale of any security or interest in any pooled investment vehicle sponsored, discussed, or mentioned by a16z personnel. Nor should it be construed as an offer to provide investment advisory services; an offer to invest in an a16z-managed pooled investment vehicle will be made separately and only by means of the confidential offering documents of the specific pooled investment vehicles — which should be read in their entirety, and only to those who, among other requirements, meet certain qualifications under federal securities laws. Such investors, defined as accredited investors and qualified purchasers, are generally deemed capable of evaluating the merits and risks of prospective investments and financial matters.

There can be no assurances that a16z’s investment objectives will be achieved or investment strategies will be successful. Any investment in a vehicle managed by a16z involves a high degree of risk including the risk that the entire amount invested is lost. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by a16z is available here: https://a16z.com/investments/. Past results of a16z’s investments, pooled investment vehicles, or investment strategies are not necessarily indicative of future results. Excluded from this list are investments (and certain publicly traded cryptocurrencies/ digital assets) for which the issuer has not provided permission for a16z to disclose publicly. As for its investments in any cryptocurrency or token project, a16z is acting in its own financial interest, not necessarily in the interests of other token holders. a16z has no special role in any of these projects or power over their management. a16z does not undertake to continue to have any involvement in these projects other than as an investor and token holder, and other token holders should not expect that it will or rely on it to have any particular involvement.

With respect to funds managed by a16z that are registered in Japan, a16z will provide to any member of the Japanese public a copy of such documents as are required to be made publicly available pursuant to Article 63 of the Financial Instruments and Exchange Act of Japan. Please contact compliance@a16z.com to request such documents.

For other site terms of use, please go here. Additional important information about a16z, including our Form ADV Part 2A Brochure, is available at the SEC’s website: http://www.adviserinfo.sec.gov