Why $1B Exits are Dead

The conversation opens with David George arguing that the scale of AI revenue growth is unprecedented: Anthropic and OpenAI are adding more monthly revenue than any hyperscaler, and he projects a combined $200 billion revenue run rate for those two companies by end of year. Critically, he notes this is happening when enterprise AI diffusion is below 5%, suggesting the growth trajectory is extraordinary. He frames the upper bound of AI's economic capture as a function of Fortune 500 profits — roughly $2 trillion annually — implying AI could eventually absorb a substantial share of that.

The discussion then turns to where value is being captured in the AI stack. George describes a shifting narrative: early assumptions that model companies would dominate gave way to beliefs that application companies would thrive on top of APIs, and now model companies are again moving up the stack into applications to drive stickiness. He argues that being 'in the token path' is the most important criterion for evaluating companies right now, because enterprise IT budgets are being squeezed by AI costs and won't expand for legacy software.

George describes how native AI companies operate fundamentally differently from SaaS-era companies — leaner, more aggressive, and already using agents rather than keyboards. He contrasts this with mature enterprises, which are mostly still in a 'documentation phase,' converting internal knowledge to markdown files before attempting automation. He argues the real agentic transformation of enterprise operations is still very early.

On exit sizes, Clark notes that the top 1% exit threshold has gone from $10 billion (2020–2024) to $20 billion (early 2026) to $32 billion (current), representing a 10x increase in roughly 24 months. The combination of large expected IPOs (SpaceX, OpenAI, Anthropic) could represent $4–5 trillion in value creation, dwarfing the cumulative value of all VC-backed IPOs over the last six years.

The pair discuss whether AI is in a bubble. George argues confidently it is not currently a bubble, citing supply constraints across compute, power, data centers, and talent as a limiting factor. He notes new data center capacity isn't available until late 2028 or early 2029, and the US is already roughly a year behind expected build-out schedules. The one scenario that could shift this, he suggests, is an unexpected algorithmic breakthrough that drastically reduces token consumption.

On defensibility and competitive dynamics, Clark raises the Forbes AI 50 data showing 40% of companies dropped off the list year-over-year, suggesting very short half-lives for AI startups. George acknowledges the difficulty of predicting winners and notes that the biggest unknown is the market structure of frontier model companies — specifically how many players remain at the frontier and whether token prices stay high or compress. He argues that lower token prices, while better for the broader economy, could accelerate labor restructuring.

The Chinese AI competitive dynamic is briefly discussed: Chinese LLMs are reportedly six months behind US frontier models but 10x cheaper, raising classic innovator's dilemma questions. George notes that appetite for frontier intelligence has so far exceeded expectations, even as per-token costs drop more than 10x year-over-year in absolute dollar terms.

On VC portfolio construction, George argues against optimizing for low loss ratios, saying a prominent VC who has 'never lost money' is actually demonstrating insufficient risk-taking. A16Z's philosophy is to back the best founders in spaces with strong tailwinds, accepting that some spaces won't work out. He distinguishes this from the worse outcome: a space succeeds and they backed the wrong company.

Finally, both speakers express optimism about AI's societal impact and the opportunity for venture capital to be at the center of a generational shift. George notes that the last decade of consumer internet was characterized by big tech capturing attention with little room for challengers, and he believes AI will cause a meaningful shift in consumer time-spent, potentially creating the next wave of extraordinary consumer companies.