The US Outspent China on AI 23 to 1

The US outspent China on AI 23 to 1 last year. It didn't work.

Stanford's 2026 AI Index puts a number on it: the gap between US and Chinese AI output has closed to 2.7%.

The US poured $285.9 billion into private AI investment last year. China spent $12.4 billion. Twenty-three times more capital. Roughly the same output.

And while we're still running pilots, millions of ordinary Chinese citizens are already using personal AI agents (they call them "lobsters") to trade stocks, post content, and automate their daily lives.

This isn't a model benchmarks story. It's a capital efficiency story. And if you're a business owner watching your AI budget grow while your results don't, that gap should feel familiar.

The companies I talk to are spending real money on AI tooling, platforms, and headcount. Most of them can't tell you what they've gotten back (and I've asked). Not because AI doesn't work, but because spending more has become the strategy. Bigger model, more licenses, another vendor.

China's results say otherwise. The teams winning aren't the ones spending the most. They're the ones picking a clear problem and shipping a solution.

What gets me though is the way most companies evaluate AI: they're relying on the vendor's scorecard. Public benchmarks that try to measure everything for everyone and end up measuring nothing for you. Your business has specific problems, specific data, specific workflows. The only benchmark that matters is the one you build yourself: does this model do this task better than what we had before, and at what cost?

Skip the leaderboard. Build your own test. Run the models against your actual work. That's how you stop buying hype and start buying results.