China is flooding the world with open AI weights — on purpose
Beijing treats Kimi, DeepSeek and Qwen as industrial policy: free frontier weights, developer programs abroad, and a supply that outlasts any US ban.
Copy markdown
Open weights are Beijing's industrial policy now
A July 26 analysis and Xi Jinping's mid-July AI alliance make it explicit: China is deliberately pushing Kimi K3, DeepSeek, Qwen and GLM into global circulation and funding developer-training programs across emerging markets. The free frontier weights your stack leans on aren't a fad — they're state-backed and accelerating.
The arsenal you can self-host today
The strategy's output is the open-weight lineup already on Hugging Face: Kimi K3 (2.8T-param MoE, 1M context, weights landing this week), DeepSeek V4, Qwen and GLM. Net for you: near-frontier coding and agents with no per-token bill and no data leaving your machine.
The risk to hedge: Washington could wall it off
Washington is still weighing restrictions on Chinese open weights, and the fight is unresolved. Even if a ban lands, the weights stay downloadable abroad — so the real risk to your workflow is enterprise and compliance friction, not the models disappearing. Practical move: pin the exact checkpoints you rely on locally now and keep a non-Chinese fallback wired up.