96% of the World, 12% of the Money
U.S. startups took nearly 88% of global AI funding this year. Here's the mental model for reading a number that extreme.
Here’s a number that should stop you. Of every dollar of AI startup funding raised globally this year, nearly 88% went to companies headquartered in the United States. That’s about $319 billion. Most of it landed with just two names, OpenAI and Anthropic. Pull the lens back to all venture funding and U.S. companies still took roughly 80% of global seed-through-growth money.
For context, in the decade before the AI boom, American startups usually pulled less than half of global investment. We’ve gone from under 50% to nearly 80% in a few years. Crunchbase, whose data this is, is openly calling that concentration a reason for serious bubble consideration.
Meanwhile the rest of the world isn’t dead. China’s startups have raised over $33 billion this year, already past their full 2025 total. U.K. startups are at $16.5 billion and on pace to beat last year. The talent and the companies exist. They’re just not where the capital’s pooling.
The mental model: Regression to the Mean
Regression to the mean is one of the most underused ideas in business. When a measurement lands at an extreme, the next one tends to move back toward the long-run average, because extremes are usually part skill and part temporary conditions, and the temporary part fades. The tallest fathers have shorter sons. The hottest fund has a cooler next year.
A few caveats before anyone over-reads this. Regression doesn’t promise when, and it doesn’t say the average itself can’t shift upward. The U.S. genuinely builds more category-defining tech companies than anywhere else, and that’s not luck. But a jump from under 50% to nearly 88% is exactly the kind of extreme the model tells you to treat with suspicion instead of extending in a straight line.
That’s the opposite of how most people read a hot streak. They see 88% and assume it’s the new normal and only heading higher. The model says the unusual part of that number is the part most likely to soften.
Spence’s take
None of this is a prediction about any stock or a reason to do anything with your money. It’s a thinking tool. When you see a number sitting way out past its own history, the useful question isn’t “how do I ride it.” It’s “what’s temporary here, and what’s permanent.”
For builders, the contrarian read is the interesting one. If almost all the capital and attention is crowding into a handful of U.S. labs, the underserved ground is everywhere else: other geographies, other markets, the places good companies get built without a megaround behind them. Concentration creates blind spots, and blind spots are where the next thing usually starts.
I think about this with /mkt too. Building inside Reg A+ with tZERO’s infrastructure is partly about widening who gets access to markets in the first place, instead of concentrating it further. Where everyone’s already crowding is rarely where the opportunity’s biggest.
Don’t extrapolate the extreme. Ask what regresses and what doesn’t.
———————————————————————————————
If this was useful, share it with someone who builds things. And if you want the full toolkit of 50 mental models, you can grab my book, Mental Models: How to Think, Act, and Win, on Amazon right now.
This post is for informational and educational purposes only. It is not investment advice, an offer to sell, or a solicitation of an offer to buy any security.