NVIDIA, Microsoft, Meta, and 22 others just signed a letter that, read carefully, is a lobbying front to keep Chinese open-weight models legal in America.
The letter is titled Open Weights and American AI Leadership. Not once does it say “China.” It does not have to. Washington is drafting restrictions on Chinese open-weight models, citing cybersecurity, IP theft, and distillation. Twenty-five American companies replied with one message: hands off.
Open weights slash training costs, break vendor lock-in, and expand the total addressable market for AI compute. Jensen Huang does not care which flag is on the model. He cares which chips run it.
The letter calls distillation a legitimate technique, not theft. (It concedes that “unlawful extraction” deserves targeted enforcement — but that concedes nothing, because Chinese licences explicitly permit distillation. What is theft against Anthropic is a feature against K3.)
Our research team ran the numbers: an American startup can distill a competent SLM off K3 for roughly $30–50K all-in — the cost of a mid-range sedan — and own the full stack.
This is not US versus China. This is closed-model rentiers versus the open-weight industrial base.
Washington sees a geopolitical threat. Silicon Valley sees a supply chain opportunity.
The gatekeepers just do not know they have already lost.
Transcribed from Jaime Xiao’s LinkedIn feed.