Analysis
The shift is the result of a deliberate Chinese strategy. Rather than competing head-on with Western labs on closed frontier models, Chinese researchers prioritised open-weight releases with permissive licenses, no rate limits, and aggressive performance benchmarks. The strategy has paid off in distribution.
For Western labs, the picture is more complicated. Open-source releases give Chinese models a halo effect that traditional paid customers now measure themselves against, and the geopolitical angle is increasingly hard to ignore — every developer downloading a Chinese model is also downloading a piece of the Chinese AI ecosystem.
Looking ahead, the open-source wave is expected to widen further. New Chinese labs are releasing more domain-specific models — for scientific research, mathematics, and code — and the volume of downloads is now growing faster than the volume of paid inference. The next round of the AI competition will be measured in downloads, not tokens.