> just because Chinese firms make open weight models available now, this could easily change very quickly in the future for a variety of reasons

China is not providing "open weight" models just because they think it's good for humanity. It is a competitive strategy to stay relevant at the forefront of the industry. The reality is that China is severely compute-constrained mostly due to Nvidia chip and ASML embargo. Nearly all available compute capacity has to go toward training, leaving very little for inference. It is why they release their models openly allowing third parties in EU and USA to host inference for them.

This will change when China reaches a critical scale of domestic AI chip production, which is inevitable.