Uncapped with Jack AltmanDavid George on Growth Investing, AI, and Why the Power Law Is Stronger Than Ever | Ep. 58
At a glance
WHAT IT’S REALLY ABOUT
Why AI investing isn’t zero-sum: everything scales, power laws intensify
- David George argues AI investing is unusually non-zero-sum because demand is massive, supply is constrained, and many layers of the stack can scale simultaneously.
- He claims current enterprise revenue is highly concentrated among coders (a steep power law), implying we are early in diffusion across the broader 1.5B knowledge-worker market.
- He explains why both frontier labs and open-source/N-1 models should succeed, with frontier winning high-value tasks and cheaper models expanding as cost optimization becomes important.
- He outlines why vertical application companies (e.g., legal via Harvey) can build durable businesses through workflow depth and go-to-market execution that frontier labs are unlikely to replicate broadly.
- He connects these product-cycle dynamics to capital allocation, arguing growth-stage investing is increasingly where private-market returns are made and that narrative (“vibes”) materially affects valuation, hiring, and customer trust.
IDEAS WORTH REMEMBERING
5 ideasIn AI, the right answer is usually “and,” not “or.”
George argues the core mistake is framing AI outcomes as mutually exclusive (frontier vs. open source, incumbents vs. startups). With demand far ahead of supply and diffusion still early, multiple layers and competitors can all grow dramatically at once.
AI monetization is a power law: a tiny set of heavy users funds the boom.
He notes today’s AI revenue is disproportionately driven by a small cohort of paying enterprise users—especially coders—despite massive top-of-funnel usage. That concentration implies both a strong current wedge (coding) and huge remaining whitespace across broader knowledge work.
Compute buildout is likely to be absorbed, not stranded—absent a major efficiency breakthrough.
George believes compute/token demand will remain insatiable because (1) enterprise diffusion is still very low, (2) bottlenecks constrain supply for years, and (3) better reasoning at inference time tends to consume more compute to produce higher-quality outcomes.
Frontier and open source both win: users will segment by value and cost.
He expects frontier labs to dominate high-value workloads (users paying ~10x more for better models) while open source/N-1 models expand as usage scales and cost sensitivity rises. Over time, applications and orchestration layers will dynamically route tasks to the “right” model for value/cost.
Apps can still win against labs because distribution + workflow fit are hard to centralize.
In vertical apps like legal (Harvey), he argues durable advantage comes from workflow details and heavy go-to-market/implementation, which frontier labs are unlikely to replicate broadly without damaging ecosystem trust. This supports a broad field of successful vertical and application-layer companies even near the “coding blast radius.”
WORDS WORTH SAVING
5 quotesIf the premise of your question is, is this gonna be successful or that thing gonna be successful? The answer in AI is probably and.
— David George
So like it is barbaric to not actually allow for these things to diffuse into your economy.
— David George
We did this analysis a long time ago that basically showed, um, half of, half of private market returns get generated between the seed and the B, and then half of returns get generated from the C+.
— David George
Whereas if you put $40 billion towards training, you know, these models- you know? Like- they get much better.
— David George
I never short a messianic founder and I never short a product that people love.
— David George
High quality AI-generated summary created from speaker-labeled transcript.
