They are making the same mistake they did with semiconductor — emphasis on only the cutting edge tech and ignoring everything else.The problem of trying to ban Chinese models is it's trivially easy to post train fine-tune an open weight model to become a new (American) model. Just example, Cursor's Composer 2.5 is a post-trained fine-tune version of Kimi 2.5. Cursor never touts this but this is well known. It takes a Chinese model and re-makes it into another model. In what universe can you regulated/ban it? It's no longer Chinese anymore as it's post trained.
So think about this, they literally cannot ban it, so what can they do? Anthropic better IPO soon to cash in because long term they have no enduring moat unless they make a breakthrough in AGI which is unlikely with LLMs. There is no justification of all these enormous compute spend if China just releases a free version 6 months later. The moat isn't there if token costs are sky high because you are basically brute forcing it with compute. At some point, open weight models will be good enough for the price point to do most of the routine agentic or coding tasks, and closed source frontier will be left with planning or debugging stubborn problem.

