The Rule for Picking AI Winners | The a16z Show
Anthropic and OpenAI are each adding more monthly revenue than Meta, Google, or Microsoft — yet actual enterprise adoption sits below 5%. Top-tier exit valuations have tripled from $20 billion to $32 billion in just six months, with potential to reach $100 billion by year's end. Meanwhile, 40% of last year's «AI 50» startups have already vanished from the list. As trillion-dollar IPOs loom and token costs pressure every buyer, three critical unknowns will determine who captures value: the competitive structure of frontier model companies, the viability of open-source alternatives, and whether algorithmic breakthroughs can radically shrink model size. Can venture capital navigate these shifting sands, or will the winners emerge from places no one is watching?
Key Takeaways
The two leading AI model companies are already adding more revenue per month than any hyperscaler, yet enterprise diffusion remains under 5% — suggesting extraordinary scale ahead as adoption spreads beyond coding into legal, operations, and other white-collar functions.
Top 1% venture exits have grown from $10 billion (2020–2024) to $32 billion today, with potential to exceed $100 billion by September 2025 — outcomes now larger than the entire Russell 2000 and all VC-backed IPOs of the last six years combined.
Value capture hinges on three unknowables: competitive structure among frontier model providers, the role of open source (dependent on distillation viability at ~2% of training cost), and potential algorithmic breakthroughs that could reduce token consumption — making early-stage portfolio construction both essential and perilous.
Companies must be «in the token path» to survive, as Fortune 500 buyers face cost pressures equivalent to 10% of annual profit from AI spend alone — forcing reallocation from legacy software budgets and potentially restructuring labor.
The half-life of AI leaders is collapsing (40% of last year's top startups are gone), yet supply constraints — data center capacity unavailable until late 2028 — make a bubble unlikely in the near term, with five trillion in expected capex justified by projected revenue returns.
In a Nutshell
We are witnessing the fastest wealth creation in venture history — companies reaching $30–60 billion valuations in four years — but survival requires being «in the token path» and adapting to a market where the model labs' competitive dynamics, not product features, will dictate who captures trillions in enterprise spend.
The Revenue Scale Revelation
Leading AI labs add more monthly revenue than hyperscalers at under 5% diffusion.
The enterprise AI market underwent a phase change in November 2024. Anthropic and OpenAI are each adding more revenue per month than Meta, Google, or Microsoft — a scale previously unthinkable for private companies. Yet actual diffusion of this technology into the real economy remains below 5%. Within coding and tech-forward companies, adoption is advanced, but across every other enterprise function, utilization of AI capabilities is virtually nonexistent.
The Fortune 500 collectively generates approximately two trillion dollars in annual profit. By year-end 2025, the combination of Anthropic and OpenAI could be running at a $200 billion revenue rate — roughly 10% of that profit pool, before accounting for open-source usage and other vendors. This cost pressure is already forcing enterprises to reallocate budgets away from previous-generation software, and in many cases, consider restructuring labor. The diffusion curve has barely begun, but the financial impact is already material.
Early signals of what full adoption looks like are visible in legal services, where the pattern that emerged in coding is repeating: when models reach sufficient capability and products around them mature, usage takeoff is dramatic. This pattern is expected to replicate across dozens of enterprise functions and verticals over the next twelve months, fundamentally reshaping how companies operate and where they allocate resources.
Exit Valuations in Hyperdrive
Top-tier exits have tripled in six months and may hit $100B this year.
The Three Unknowables
The Token Path Mandate
Survival requires being in the inference path as cost pressures reshape budgets.
The Token Path Mandate
The single most important filter for evaluating startups is whether they are «in the token path» — generating inference volume that customers must pay for. Cost pressure is already hitting Fortune 500 buyers, who cannot expand budgets to cover AI growth and are cannibalizing legacy software spend. Companies not embedded in the inference flow will face existential budget squeezes, while those that control token consumption will capture compounding value as usage scales.
Supply Constraints, Not Bubble Dynamics
Data center scarcity through 2028 and justified capex make oversupply unlikely.
The Changing Venture Landscape
Founders prefer scale platforms; loss ratios will normalize as market matures.
The venture market has undergone structural shifts in response to the speed and scale of AI company growth. Companies are encountering big-company problems — complex supplier negotiations, cloud deals, international expansion, sales force scaling — far earlier in their lifecycles than any previous technology wave. Cursor, for example, is generating billions in revenue while still very small and early-stage. This dynamic has driven founders toward large platform firms that can provide expertise in pricing, channel strategy, and international operations from day one.
Loss ratios in AI investing are currently in the single digits, well below the historical 60% benchmark for early-stage portfolios. This is unsustainable and reflects the extreme youth of the market, not a new paradigm. Forty percent of the companies on last year's «Forbes AI 50» list have already dropped off, signaling that the half-life of AI startups is collapsing. The philosophy remains unchanged: back the best founders in spaces with tailwinds, accept that many categories will not work out, and avoid the fatal error of picking the wrong company in a category that does work.
The near-term focus is on maintaining high win rates on the best early-stage companies and building sufficient ownership to capture power-law outcomes. The firms best positioned are those that adapted their platforms to match what hypergrowth AI companies actually need, rather than applying outdated playbooks from the SaaS era.
Securities Mentioned
People
Glossary
Disclaimer: This is an AI-generated summary of a YouTube video for educational and reference purposes. It does not constitute investment, financial, or legal advice. Always verify information with original sources before making any decisions. TubeReads is not affiliated with the content creator.