This is pretty scary. I see a lot of logical takes here, even though I canโt verify the data.
From first principles: talent, energy, regulation, compute.
China is likely on par with or ahead in talent, clearly ahead in energy, and appears more supportive on regulation. Theyโre behind in compute today, but they can add more chips over time, even if theyโre less efficient.
The only practical way for the U.S. to beat China is to make energy abundant and cheap, with fewer regulatory barriers.
Just look at the open-weighted models from China; theyโre incredibly good. And since NVIDIA is banned there, they can be running entirely on the Chinese stack.
Soon, a large share of the worldโs AI workloads could be running on Chinese open models from Chinese clouds powered by Huawei chips.
What would stop, say, an Indian developer from choosing cheaper Deepseek-like Chinese cloud API alternatives over say OpenAIโs API?
To be honest, I think the U.S. actually accelerated Chinese AI progress by blocking the H20s earlier (China banned them only after the export controls were lifted) which seems to have triggered a new sense of urgency and national pride around AI development there.