At a glance
WHAT IT’S REALLY ABOUT
How to pick AI winners amid scale, costs, and churn
- Anthropic and OpenAI are portrayed as adding revenue at a pace exceeding hyperscalers while AI penetration into the broader economy remains very low, implying large upside if diffusion accelerates.
- Enterprise AI is still largely “skeuomorphic” (making existing workflows faster), but the next shift is toward native, proactive, agentic systems and new ways companies are run.
- The magnitude of venture outcomes is compressing in time, with “top 1% exits” rapidly re-benchmarking upward, increasing both opportunity and the cost of picking wrong.
- Predicting durable winners is getting harder as technology shifts faster and “half-life” appears short, with a large fraction of prominent AI startups falling off leader lists year over year.
- Value capture hinges on token economics and model-market structure (number of frontier competitors, open source viability, and cost declines), while compute and data-center scarcity reduces near-term bubble risk.
IDEAS WORTH REMEMBERING
5 ideasScale is racing ahead of real-economy adoption.
They argue leading model companies are already adding revenue faster than hyperscalers even though AI diffusion into most enterprise functions is cited as under ~5%, implying the next adoption wave could be much larger than current usage suggests.
Enterprise spend will be bounded by where budget can come from—profits and labor.
A practical “upper bound” framing is that enterprises must fund AI from a finite pool (e.g., aggregate profits), forcing tradeoffs like replacing legacy software spend, raising prices, and/or restructuring labor to cover AI run costs.
“Be in the token path” is emerging as the rule of thumb for picking winners.
Because AI costs hit buyers directly, companies that sit directly in the flow of token consumption (or tightly govern/optimize it) are positioned to remain essential when budgets tighten and procurement becomes cost-driven.
Model-market structure will decide who captures most of the economics.
If only a few frontier labs exist, token prices stay higher and value accrues upstream; if many credible frontier providers and open source options exist, token prices fall, enabling more value to accrue to downstream applications and the broader economy.
Open source’s fate may depend on distillation feasibility and countermeasures.
They note distillation can be far cheaper than full pretraining, which would favor open ecosystems—unless frontier labs successfully prevent/limit distillation, shifting advantage back to closed providers.
WORDS WORTH SAVING
5 quotesBasically Anthropic and OpenAI are adding more revenue per month than Meta, Google, or Microsoft.
— David George
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%.
— David George
The new companies are very lean, very aggressive, and they work all the time.
— David George
Between 2020 and 2024, top 1% exits started at $10 billion... We just updated them yesterday. Um, and if you look at just the exits that have closed, it's now at $32 billion. So Wiz is the, is the threshold for the, the top 1%.
— David Clark
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 three years from now.
— David George
High quality AI-generated summary created from speaker-labeled transcript.
