Update: Nvidia, Meta, Microsoft sign on to save Chinese open weights
25 US firms, minus OpenAI, Anthropic and Google, warn a ban kills American startups, not the models — which stay one download away on Hugging Face.
Copy markdown
The coalition — and the labs that skipped it
Twenty-five US firms — Nvidia, Microsoft, Meta, Dell, IBM, Palantir, Mistral, Mozilla, the Linux Foundation, Hugging Face, a16z and Y Combinator among them — signed a joint letter urging Washington to avoid “premature restrictions” on open-weight models. OpenAI, Anthropic and Google all sat it out, splitting the industry cleanly along the open-vs-proprietary line.
What Washington is actually weighing
The push was reportedly revived after Moonshot's Kimi K3 topped US models on benchmarks: the administration is looking at restricting Chinese open-weight models — Kimi, Qwen, DeepSeek, GLM — on cybersecurity grounds. Nothing is law yet, but the letter lands as that review heats up.
A ban you can't actually enforce
The signatories' sharpest point: the weights are already downloaded, mirrored and running everywhere, so a ban can't un-publish them. It would just make US developers the only ones who legally can't use them, while the models stay one click away on Hugging Face for everyone else.
Why it lands on your stack
These are the exact models this audience self-hosts for cheap, frontier-tier work — Kimi K3's weights drop tomorrow, GLM-5.2 ships under MIT, Qwen and DeepSeek V4 are staples. A restriction wouldn't scrub them from the internet; it would pull the sanctioned rug out from under the cheapest tier in your toolkit.