Video · 2026-05-29 · 33m · 24 moments

The New Rule for Picking AI Winners | The a16z Show

✦ AI generated

GUEST Interviewer GUEST David GUEST David George

timeline · colored by role

Prediction◆

Anthropic and OpenAI are already adding more revenue per month than Meta, Google, or Microsoft, and the combined pair could be at a $200 billion revenue run rate by the end of this year.

David argues the two leading AI labs are now adding revenue faster per month than the biggest hyperscalers, and projects their combined run rate could hit $200B by year end.

transcript

David: Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft. And I wouldn't be surprised if the combination of those two companies is doing 200 billion of revenue run rate.

extends · 4rebuts · 1supports · 3

Data◆

Anthropic and OpenAI are already adding more revenue per month than Meta, Google, or Microsoft, yet AI's actual diffusion into the real economy is still under 5%, implying outcomes ahead will be extraordinary.

David argues that despite Anthropic and OpenAI already outpacing hyperscalers in monthly revenue growth, AI has diffused into less than 5% of the real economy, meaning the eventual outcomes will be extraordinary.

transcript

David: Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft. They are already at that scale of revenue getting added and actual diffusion of this technology into the real economy is tiny. It's like less than 5%.

Claim◆

Anthropic and OpenAI are already adding more revenue per month than Meta, Google, or Microsoft, yet AI's actual diffusion into the real economy is still under 5%, implying enormous unrealized upside.

David George argues that despite Anthropic and OpenAI outpacing hyperscaler revenue growth, AI adoption across most enterprise functions remains under 5%, suggesting massive headroom for outcomes.

transcript

David George: Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft. They are already at that scale of revenue getting added and actual diffusion of this technology into the real economy is tiny. It's like less than 5%.

rebuts · 1

Data

Despite the scale of revenue AI labs are adding, real-world diffusion of AI into the broader economy remains tiny, at less than 5%, outside of coding and tech-forward companies.

David contrasts massive revenue growth at AI labs with the fact that actual enterprise adoption/diffusion across most business functions is still under 5%.

transcript

David: actual diffusion of this technology into the real economy is tiny. It's like less than 5%. Now, within coding and in tech-forward companies, yes, it's it's much more advanced. Um but as it relates to every other function in the enterprise, um you know, full sort of utilization of the capabilities, we're nowhere right now.

supports · 2

Data◆

Diffusion of AI into the real economy is still under 5%, even though Anthropic and OpenAI are already adding more monthly revenue than Meta, Google, or Microsoft.

David argues that despite Anthropic and OpenAI already out-adding revenue growth of the biggest hyperscalers, actual enterprise adoption of AI outside coding remains under 5%, implying massive headroom.

transcript

extends · 2

Prediction

The Fortune 500's roughly $2 trillion in annual collective profit is the ceiling enterprises can draw from to pay for AI, and OpenAI plus Anthropic alone could hit a $200 billion revenue run rate by year-end, already representing about 10% of that profit pool.

David frames enterprise AI spending against the ~$2 trillion in collective Fortune 500 profit, noting OpenAI and Anthropic could reach $200B in run-rate revenue by year end—already ~10% of that pool—raising the question of where the money will keep coming from.

transcript

David: if you just look at the Fortune 500 or the S&P 500, they generate like 2 trillion of profit per year at the collective. And I wouldn't be surprised if the combination of those two companies is doing 200 billion of revenue run rate by the end of this year... So, we're already talking about like a 10% profit, you know, into the Fortune 500.

Example

AI-native companies operate fundamentally differently from prior SaaS-era companies, running lean and constantly by deploying swarms of autonomous agents rather than employees typing at software.

David George describes visiting cutting-edge AI companies where researchers whisper instructions to running swarms of agents instead of typing, contrasting this lean, always-on model with the inefficiently-run SaaS companies of the previous generation.

transcript

David George: It's fun to see like the most cutting-edge companies when you go in, you know, all their researchers are sitting there and they're whispering in... they're not even typing. Like they're efficient, they're whispering in and they're running, you know, swarms of agents. The new companies are very lean, very aggressive, and they work all the time.

extends · 1provides context · 1

Data

The bar for a top 1% startup exit has grown roughly 10x in about 24 months, from $10 billion to $32 billion, and could exceed $100 billion by September with OpenAI and Anthropic potentially factored in.

The interviewer cites internal data showing the top 1% exit threshold jumped from $10B (2020-2024) to $20B (early 2026) to $32B as of the latest update, and could top $100B by September if OpenAI and Anthropic exit.

transcript

Interviewer: between 2020 and 2024, top 1% exit started at $10 billion. We updated those numbers in February this year... a top 1% exit for 25 in the first 2 months of 26 was then $20 billion. We just updated them yesterday... if you look at just the exits that have closed, it's now at $32 billion.

Data

The bar for a top 1% startup exit has roughly tripled from $10 billion to $32 billion in just 24 months, and could exceed $100 billion by September once OpenAI and Anthropic are counted.

The interviewer cites internal data showing the top 1% exit threshold jumped from $10B (2020-2024) to $20B (early 2026) to $32B as of the most recent update, with a potential leap past $100B once OpenAI and Anthropic exits are included.

transcript

Interviewer: Between 2020 and 2024, top 1% exit started at $10 billion. We updated those numbers in February this year, and a top 1% exit for 25 in the first 2 months of 26 was then $20 billion. We just updated them yesterday, and if you look at just the exits that have closed, it's now at $32 billion.

Data◆

The bar for a top 1% startup exit has roughly tripled in about two years, rising from $10 billion (2020-2024) to $20 billion (early 2026) to $32 billion based on already-closed deals, and could exceed $100 billion by September.

The interviewer cites their firm's data showing the top 1% exit threshold jumped from $10B to $32B in roughly 24 months, potentially topping $100B by September with OpenAI/Anthropic exits.

transcript

Interviewer: between 2020 and 2024, top 1% exit started at $10 billion. Um we updated those numbers in uh in February this year. Um and a top 1% exit for 25 in the first 2 months of 26 was then $20 billion. We just updated them yesterday. Um and if you look at just the exits that have closed, it's now at $32 billion.

Data◆

The bar for a top 1% startup exit has roughly tripled in the last year and 10x'd over 24 months, going from $10 billion to $32 billion, and could top $100 billion by September given OpenAI and Anthropic.

The interviewer cites internally tracked data showing the threshold for a top 1% exit has exploded from $10B to $32B in two years, and may exceed $100B once OpenAI and Anthropic potentially exit.

transcript

Interviewer: So, between 2020 and 2024, top 1% exit started at $10 billion. Um we updated those numbers in February this year. And a top 1% exit for 25 in the first 2 months of 26 was then $20 billion. We just updated them yesterday. And if you look at just the exits that have closed, it's now at $32 billion.

Data

Predicting which AI companies will ultimately capture value is getting much harder, as shown by roughly 40% of last year's Forbes AI 50 list dropping off this year's list.

Citing the high year-over-year churn in Forbes' AI 50 list, the interviewer argues that AI company leadership is unusually short-lived and unpredictable compared to prior tech cycles.

transcript

Interviewer: One of the things we track is, you know, every year Forbes comes out with their AI 50 startups list. And what was really interesting was, you know, from last year to this year, 40% of the companies that were on that list last year dropped off. So, like the half-life of these companies feels kind of incredibly short.

explains mechanism · 1rebuts · 1

Data

Company defensibility in AI is unusually fragile: 40% of the companies on last year's Forbes AI 50 list dropped off this year, showing that being an early leader doesn't guarantee staying power, much like Google wasn't the first search engine and Facebook wasn't the first social network.

The interviewer notes that 40% of companies on the Forbes AI 50 list fell off within a year, illustrating that predicting which AI company ultimately captures value is getting much harder, echoing history where first movers like early search or social networks didn't win.

transcript

Interviewer: every year Forbes comes out with their AI 50 startups list. And what was really interesting was, you know, from from last year to this year, 40% of the companies that were on that list last year dropped off... So, like the half-life of these companies feels kind of incredibly short.

Fact

Predicting which companies will capture value in AI is getting much harder, as shown by 40% of Forbes' AI 50 list dropping off year over year, indicating an incredibly short half-life for leading AI companies.

The interviewer notes that historically first movers haven't always won, and that 40% turnover on the Forbes AI 50 list in one year shows how quickly current AI leaders can fall away.

transcript

Interviewer: every year Forbes comes out with their AI 50 startups list. And what was really interesting was, you know, from from last year to this year, 40% of the companies that were on that list last year dropped off. So, like the half-life of these companies feels kind of incredibly short.

rebuts · 1

Mechanism◆

The biggest unknown driving where AI's economic value gets captured is the market structure of frontier model labs — fewer competitors at the frontier likely means higher token prices, while more competitors means lower prices.

David frames being 'in the token path' as the top criterion for AI investing today, and argues the number of frontier labs competing is the key unknowable that will set token prices and determine who captures value.

transcript

David: The biggest driver of where value is going to get captured right now is I would say something that is totally unknowable, which is what is the market structure of the model companies? How much competition is there? If there's a couple at the frontier, token prices will probably be higher. If there are five at the frontier, token prices will probably be lower.

Mechanism

Being 'in the token path' is now the single most important factor for assessing a company's staying power, because the market structure of frontier model companies (how much competition exists) will determine token prices and thus who captures economic value.

David explains that being positioned in the 'token path' is now the top criterion for evaluating a company, since the unknowable competitive structure among frontier model labs will set token prices and dictate where value accrues.

transcript

David: right now, you have to be in the token path. Like that is the number one thing that we're looking to for our companies... The biggest driver of where value is going to get captured right now is I would say something that is totally unknowable, which is what is the market structure of the model companies? How much competition is there? If there's a couple at the frontier, token prices will probably be higher. If there are five at the frontier, token prices will probably be lower.

extends · 1provides context · 1supports · 2

Claim◆

Being in the direct path of AI token spend is now the single most important criterion for a company's ability to capture value, and that value capture will ultimately hinge on the unknowable competitive structure of the frontier model labs.

David George says the number one thing a16z screens for now is whether a company sits 'in the token path,' and argues the biggest unknown determining value capture is how much competition exists among frontier model labs, which will set token prices.

transcript

David George: Right now, you have to be in the token path. Like that is the number one thing that we're looking to for our companies... The biggest driver of where value is going to get captured right now is I would say something that is totally unknowable, which is what is the market structure of the model companies? How much competition is there?

Claim◆

The single biggest unknowable factor determining who captures value in AI is the market structure of the model companies — how many frontier labs compete determines whether token prices stay high or fall.

David says the unknowable variable that will decide value capture across the AI stack is the competitive structure among frontier model labs — fewer competitors means higher token prices, more competitors means lower prices and a healthier broader ecosystem.

transcript

extends · 1supports · 1

Claim

A venture firm that never loses money on a deal isn't taking enough risk — the right early-stage strategy is to back the most talented founders in promising spaces, even if that means accepting real losses.

David explains a16z's early-stage philosophy: pick the best founders in high-tailwind markets and accept losses when a space doesn't work out; a near-zero loss ratio is treated as a red flag, not a badge of honor.

transcript

David: One of his big points of pride is he's never lost money on a deal. And we're like, that's not a point of pride. Like that's a horrible data point. Like that's not what you want... That's a PE firm. And so certainly you can make the case that you're not taking enough risk if that's the way you approach it.

Mechanism

A venture firm's early-stage strategy should be to back the most talented founder in any promising space regardless of outcome, because a low loss ratio actually signals insufficient risk-taking, not skill.

David George explains a16z's philosophy of backing the leading entrepreneur in any credible space with tailwinds, arguing that having zero losses is actually a bad sign, and the real failure mode is picking the wrong leader in a space that succeeds.

transcript

David George: Any major space where there are multiple very talented entrepreneurs building, where we think there's tailwinds, where we have a point of view on the technology that it's good, we should pick the best founders and try and back the leaders at the early stage. If the space happens to work out, and we've got the leader, excellent. If the space happens to not work out, and we have the leader, no harm, no foul.

Prediction

The AI industry is not currently in a bubble because it is constrained by supply — compute, power, and data centers — rather than by demand, with data center capacity essentially booked out until late 2028 or early 2029.

David argues today's AI boom differs from classic bubbles because scarcity of compute and data center capacity, not oversupply, is the binding constraint, making a bubble unlikely in the near term.

transcript

David: I feel pretty confident saying that we're not in a bubble right now. I'm less confident that we won't be in a bubble 3 years from now. But all I can speak to is where we are right now. We're massively supply constrained. You can't get data center capacity at scale until late '28, early '29 right now.

supports · 2

Prediction◆

The current AI investment climate is not a bubble because it is constrained by scarce supply (compute, power, data centers) rather than driven by excess supply chasing weak demand, though a breakthrough enabling much smaller, more efficient models could change that.

David George argues that unlike classic bubbles caused by oversupply, today's AI market is supply-constrained across compute, power, and data centers, which makes a bubble less likely for now, though a future algorithmic breakthrough toward smaller models could flip that dynamic.

transcript

David George: It's probably a healthy thing right now that we're supply constrained, only in the sense that it probably makes it less likely that we have a bubble. I feel pretty confident saying that we're not in a bubble right now. I'm less confident that we won't be in a bubble 3 years from now.

rebuts · 3supports · 2

Claim◆

The AI industry is currently supply constrained rather than demand constrained — scarce compute, memory, data centers, and power — which makes a bubble less likely right now, though that could change with an algorithmic breakthrough enabling far smaller, more efficient models.

David argues that unlike classic bubbles driven by oversupply, AI today is bottlenecked by scarce data center capacity, power, and hardware — data center capacity at scale won't be available until late 2028/early 2029 — which makes a bubble unlikely in the near term, barring a breakthrough that shrinks model compute needs.

transcript

David: We're massively supply constrained. You can't get data center capacity at scale until late '28, early '29 right now... I think we're probably a year behind schedule of what people would expect for data center buildout in the US.

rebuts · 2supports · 1

Prediction◆

We are not currently in an AI bubble, largely because the market is supply-constrained (compute, memory, data centers, power) rather than demand-constrained, though that could change within a few years.

David argues the AI market isn't in a bubble right now specifically because supply constraints (data center capacity unavailable at scale until 2028-2029) are limiting growth rather than weak demand, though he's less sure about three years out.

transcript

David: I feel pretty confident saying that we're not in a bubble right now. I'm less confident, you know, that we won't be in a bubble 3 years from now. But all all I can speak to is where we are right now... We're massively supply constrained. You can't get data center capacity at scale until late '28, early '29 right now.

provides context · 2rebuts · 1supports · 4

Highlight slides

Related episodes

Video · The Cognitive Revolution

Fable's Back, AI Engineer Recap, & SambaNova

→

Video · BG2 Pod

The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang

→

Video · 20VC

OpenRouter CEO: Why Chinese Open Models Are Beating the US | Why Enterprises Fear OpenAI & Anthropic

→