I’m in the homestretch of my dissertation on the U.S.-China AI race, so this particular rabbit hole runs pretty deep.

One question I keep coming back to is how China managed to close the gap so quickly. The obvious explanations are technological, with better models, more engineers, more money, and increasingly clever ways to squeeze performance out of available compute.

But I’m starting to wonder, and something my dissertation probably didn’t focus enough on, whether part of the answer is less about the technology itself and more about how we think about innovation.

The American version of innovation has always had a certain mythology around it. Invent something, patent it, build a moat, protect the secret sauce, and eventually sip champagne on a yacht while The Weeknd blasts in the background.

What fascinates me about China’s AI ecosystem is that it seems increasingly comfortable with a somewhat different idea. You don’t necessarily need to be first. You need to be really good at building on what already exists.

Then you can sip champagne on said yacht.

Do Chinese founders even party? That’s for another post.

DeepSeek releases a model and other researchers pull it apart. Someone improves a technique, someone makes inference cheaper, and someone else takes that improvement and pushes it a little further. Even some of DeepSeek’s own smaller R1 models were built on Alibaba’s Qwen models.

They’re all competing like hell, but they also seem to be standing on each other’s shoulders.

There is something interesting about that approach to iteration. Copying, adapting, improving, scaling, and then having someone else improve upon you doesn’t seem to carry quite the same stigma we sometimes attach to it in the West.

We tend to celebrate the person who had the original idea, but what if innovation is just as much about what the thousand people after that person do with it?

To be clear, open source is hardly a Chinese invention. Much of the modern open-source movement came out of the United States, and Chinese technology companies certainly know how to protect intellectual property when it suits them.

But watching Chinese AI over the last year or so, I keep coming back to the same thought. China seems particularly good right now at treating knowledge less like a castle to defend and more like scaffolding everyone can keep building on.

Necessity probably has something to do with it too. When access to the best chips gets harder, efficiency suddenly matters a lot more. If you can’t simply throw more compute at every problem, you get very creative about squeezing more out of what you have.

Constraints have a funny way of doing that.

Maybe China didn’t close the AI gap simply because it learned how to build better technology. Maybe it also got really good at building on what was already there.

We spend a lot of time asking who is ahead in AI. I’m increasingly wondering whether the more interesting question is who learns fastest from everyone else.

Roll credits.