Stanford’s 2026 AI index just dropped: the US spends 23x more than China on AI, but the performance gap is down to 2.7%

I went through Stanford HAI's 2026 AI Index Report this weekend and one number keeps bugging me. The top US model and the top Chinese model are now separated by 2.7% on the Arena leaderboard. In 2023 that gap was north of 17 percentage points. It basically evaporated in under three years.

The part that really got me thinking was the spending side. US private AI investment last year was $285.9 billion. China's was $12.4 billion. That's a 23 to 1 ratio. And yet the actual output, measured by the one benchmark the industry watches most closely, is nearly identical. I keep trying to square those two numbers and I can't.

China also leads in AI patent filings by a wide margin and installed nearly nine times as many industrial robots as the US in 2024. Meanwhile Stanford flags that the number of AI researchers coming into the US has dropped 89% since 2017. The talent pipeline is going the wrong direction at the exact moment it matters most.

None of this means China has "won" anything. The US still puts out more top-tier models and has way more data center capacity. But if you're someone who cares about capital efficiency, and in this sub I assume most of us do, the fact that one side is getting 97% of the way there on 4% of the budget is worth paying attention to.

Curious if anyone here has started rethinking their China tech exposure because of this, or if the geopolitical risk still outweighs the efficiency story for you. I've been poking around beyond KWEB and noticed CNQQ weights its holdings by R&D intensity, which seems like it would naturally tilt toward the kind of companies doing more with less. Haven't pulled the trigger on anything yet but the framing fits what this report is describing.