In this interview, Altimeter investor Apoorv Agrawal discusses the forces shaping the current AI boom. He argues that data and compute are the two essential ingredients for AI's future, positioning Nvidia as the dominant infrastructure provider that will benefit from a massive increase in inference demand. Agrawal frames AI as the latest "supercycle"—larger than the internet, mobile, and social cycles—projecting Nvidia could reach a $4 trillion market cap. He evaluates OpenAI, highlighting its rapid user growth (over 1 billion in 2.5 years), strong brand moat, and relentless feature velocity as the key defensibility factors, while acknowledging competition from emerging models (DeepSeek, Claude, Perplexity, Copilot) has had limited impact so far. The conversation turns to valuation, with Agrawal noting current prices appear inflated but may make sense when extrapolating exponential growth out to 2030, and he stresses that power‑law dynamics will concentrate returns in a few winners, notably Nvidia and OpenAI. Talent, data, and compute are presented as the leading indicators of value, and Agrawal explains the AI value chain in three layers—semiconductors, infrastructure, and applications—drawing parallels to past supercycles. He also comments on Meta's hiring strategy, viewing it as a small‑cap bet for a slice of the AI supercycle. Throughout, Agrawal balances optimism about AI's long‑term trajectory with caution about short‑term market dislocations and valuation bubbles.
Chapters
Look, I would say the two essential ingredients for the future of AI are data and compute. NVIDIA produces one of them and is the best game in town. And so if you believe that inference is going a billion X from here, tokens are going a billion X, the need for compute will be very long term. And so that explains all that's happening and much more to come on NVIDIA.
But if I may, just on that point, Arjun was just highlighting how all of the conversations he had in Paris were very much suggesting that everyone just wants to work with Nvidia. But at a time when we're talking about perhaps too much dominance in certain parts of this market, I'm wondering, where do we go from here in terms of everyone just perhaps wanting Nvidia, Nvidia, Nvidia? At one point, is this going to reach a peak?
Look, taking a step back, if you look at the supercycles that we've had over the last 2, 3 decades, they've all created large outcomes. Right. Let's start with the internet supercycle, early 2000s. We got a great business out of there called Google, $2 trillion market cap leader. Mobile, you got Apple, $3, $3.5 trillion market cap leader. Social, you got Meta, $1.8 trillion market cap leader. AI supercycle, you've got Nvidia, leader, on to $4 trillion market cap. And this is the largest supercycle of them all. And so. Yes, it'll be short-term dislocations, but I think in the long term it makes a lot of sense and on its path to a great future.
Let's pick up on another big AI company, OpenAI, of which you're an investor in and have led Altimeter's investment in. Can you just talk to me about how you see the future direction of travel for, for OpenAI? Of course, it made waves with ChatGPT. Now we're hearing it potentially launching a browser. It bought I/O from Jony Ive's startup as well. All these different areas that the company is looking to dabble in as well. What ultimately is going to be the business model of this company?
Look, I would say the key ingredients of OpenAI started with great talent. We built great technology and great models. We built great products with ChatGPT and some of the products you're referring to. And ultimately, I would say the most defensible thing we've got is a great brand. ChatGPT now serves over 1 billion users. It has gotten there in just 2.5 years. Since it was launched. And just to give you a sense of scale, the largest consumer platforms on the planet have 2 to 3 billion users. And here's a player that has gotten there in just 2.5 years, growing fast with great usage and great retention. Why is that the case? The reason that's the case is because they've delivered a viral moment in text, the ChatGPT moment in November 2022, in voice, Advanced Voice Mode in 2023, image, Studio Ghibli. You saw the viral moments that we all enjoyed. Text, audio, video. I mean, that's— those are the three modalities in which humans communicate. And so now ChatGPT is a brand that is widely understood by everybody.
But is brand— is brand a moat? When I look at how quickly innovation is happening in this space right now, we saw what happened. DeepSeed came out, made those kind of waves. It just takes another one of those moments for a brand new product to come up so quickly in this space.
Yeah, you know, we've done some analysis on this, which maybe you can bring up the analysis. We compared the usage of all the great consumer brands that have come up like Deepseek, like Claude and Perplexity and Copilot and so on. And we looked at the impact that they've had on the growth of ChatGPT. The answer is not much so far. Look, I'm not saying it's a layup. It might happen. But so far, the ChatGPT growth has been unassailed. And from the looks of it, we think it'll continue because there's underlying factors that drive that retention. People are coming back because the feature velocity of ChatGPT is so fast, they're shipping features, keeping up with the product releases. OpenAI is a full-time job, and that ultimately is the differentiator.
And then it comes down to valuation. And look, if you do it, if you value this company on this year's or next year's earnings, you're going to come up with an astronomical figure and you're not going to— it's not going to make any sense. But if you extrapolate the exponential growth you've been talking about a couple of years out, it begins to make sense. But I mean, it's still extraordinary valuations. I mean, there's a long way to go before they hit the kind of level, you know, somewhere between now and 2030, perhaps towards the back end of that period where the forecast revenues make sense on an earnings valuation to sales level, which is what everyone in your industry looks at, all the other measures as well. Are you concerned about— not about technology, you love it clearly. Are you concerned about the valuation being attached to this company?
Look, Steve, I'll get back to what I was saying. Each of the large supercycles have created large outcomes.
Yes.
Google at $2.5 trillion. Apple at $3.5 trillion. Meta at $1.8 trillion.
Do we believe companies you mentioned have got a stake in the game too?
That's right.
To have a degree of competition.
That's right. And do we believe that the winner of this supercycle will yield a business that will be larger than $300 billion in market cap? Yeah, absolutely.
Yeah.
The question is, is OpenAI that winner.
Yes.
And look, I'm not saying it's a layup, but based on everything we know right now, it is the indisputable leader.
So here's the point. I think the key point for me, I'm trying to be a guardian for that lot out there, not wasting their money. If they are the winner, that means— and I get it, large outcomes. I love what you're saying. I understand. Then there's an awful lot of players out there that are overvalued because they are not going to be the winner.
I'm not going to be— I'm not going to deny that. Look, there's— there is— there a bubble right now? Is our valuations hot? Certainly. But ultimately, the—
you could almost argue that NVIDIA plus OpenAI, the large outcome winners, I think you might even be arguing definitely the rest of them, forget it.
I think power laws play out, right? You can do play Mag 7. 7 of the 500 in S&P 500 have driven most of the returns. What you just described, those two that you picked, if you had a basket of those two versus everything else, I suspect a similar power law will play out with the fullness of time.
You get your 10x and then a bit more.
Yeah, I agree.
So let me just ask you, therefore, as all of the recent announcements in terms of the additional spending that Meta has done trying to hire left, right and center from other competitors, basically a waste if they can't compete with these two companies that we just mentioned.
Look, Meta is betting that, you know, 1% of their market cap in exchange for having the option value to win the largest supercycle. It makes a lot of sense. They got Alex Wang from Scale, incredible leader. Nat Friedman from— and Daniel Gross, incredible leader. Now this leader from Apple. I mean, these are, these are some great leaders. And so the calculus makes a lot of sense to us.
And honestly, the question is, Will the others follow in terms of what's right down to the hiring?
Yeah.
Yeah. But it's— sorry, is talent actually what they need in order to compete with this in this race, or is it actually technology?
Absolutely. I mean, the leading indicator of value is talent, right? Talent, data, compute. And between those three, talent is now there. They've got great compute, they've got great data, and they've got great distribution.
Above—
it's like speed dating. Two quick questions. I've ever done speed dating. This is—
if you're watching, if we look at that, Tech stack right now, all the money is being made in infrastructure with— but by NVIDIA. So let's move forward a few years and where is the value going to be created? Software layer, application layer. What are you looking at?
Look, we think a lot about this question and there is a sequencing to these, to where the value accrues. And, you know, we think about this in 3 layers. You've got the semis layer, you've got the infrastructure layer, and then you've got the applications layer. And we studied the supercycles of the internet, mobile, cloud. And invariably it starts at the semis layer as the buildout happens. Think of this as the railroad moment and it moves up to infrastructure and ultimately applications. And so that's how, that's how we are thinking about the world. I'm focused, spending a lot of my time in the application layer. And, you know, ultimately that's what leads to a sustainable market.
