Nvidia's China risk is becoming a developer story. The question is not only whether Huawei chips can run a model. It is whether builders start arranging their work around another stack.

This paper said Sunday that DeepSeek V4 timing belonged to Huawei Ascend integration, not ordinary launch gossip. It also said Kimi held the second Chinese frontier seat. Monday adds distribution.

Reuters reported DeepSeek's new model push in the context of Huawei chip adaptation. [1] Cloudflare's changelog says Kimi K2.6 is now available on Workers AI with day-zero support, usable through bindings, REST, OpenAI-compatible endpoints, and AI Gateway. [2]

That is the quiet threat. Benchmarks persuade analysts. Distribution changes defaults. A developer who can call a Chinese open-weights model through a Western edge platform is not making a geopolitical speech. He is writing code.

OpenSourceForU's benchmark treatment keeps the leaderboard story alive. [3] But the ecosystem story sits under the leaderboard: CUDA is powerful because people build around it. The alternative becomes dangerous when people stop treating it as alternative.

That is why Cloudflare belongs in an Nvidia brief. The moat erodes first as a habit, not as a press release.

Habits are cheaper than chips, and harder to sanction once they spread.

-- THEO KAPLAN, San Francisco