There is a very interesting phenomenon in the US-China AI game:

As super chips like BlackWell become mass-produced, the gap between top Chinese and American laboratories is widening rapidly. The Deepseek -V3.2 paper mentions that they are currently bottlenecked by computing power. Meanwhile, China's open-source ecosystem is at the top of the world; it is precisely because of open source that it has helped American companies like Meta. Meta is using Qwen model data to train a new generation of models.

Meta is clearly behind other major companies in AI models, but now Meta is apparently leveraging Chinese open-source models as their own checkpoints, building upon them and further optimizing.

You see, the US sanctions Chinese chips, forcing Deepseek to create large models with only 1/10 of the computing power, while China's open-source approach has overtaken on the bend. However, this open-source advantage has also been used to feed companies like Meta.

In the world of closed-source and open-source, the competition may be more about strategic intelligence.