Washington's Chinese-AI panic is aimed at the wrong target, Stratechery argues

Kimi K3, the open-weight model released last week by Moonshot AI, is good enough that Wall Street and Washington are once again convinced America is losing the AI race. Ben Thompson argues that reaction is largely misplaced: the frontier-lab advantage inside the United States remains real, and claims that Chinese labs distilled Kimi K3 from American models in a matter of weeks are, on the timeline involved, close to implausible. The actual danger, he writes, is self-inflicted — a mix of heavy safety guardrails and premium pricing on US models that hands Chinese open-weight alternatives a frictionless path into enterprise pipelines and startup stacks. That dynamic is already devaluing the proprietary foundation-model layer that companies like OpenAI and Anthropic depend on. Thompson's prescription is uncomfortable for Washington: solving the cybersecurity risk of open Chinese models requires understanding and accepting the reality of that competition rather than legislating it away, even if that leaves the American labs to fend for themselves. The essay lands just as the Trump administration revives talk of banning Chinese models outright — a policy that open, downloadable weights would make nearly impossible to enforce.