Chinese Open-Source Models Now Power 80% of U.S. AI Startups, Routing Data Shows
OpenRouter usage data reveals Chinese open-source models surged from roughly 1% of developer traffic in 2024 to over 60% in May 2026 — with 80% of American AI startups reportedly building on top of them. The implications for OpenAI, Anthropic, and U.S. AI policy are enormous.
The most consequential shift in AI infrastructure isn't happening at a frontier lab or inside a government agency. It's happening in the routing layer. According to data surfaced by @ReadFuturist, Chinese open-source AI models have gone from approximately 1% of developer usage in 2024 to more than 60% as of May 2026, with an estimated 80% of U.S. AI startups now building on Chinese open-source foundations. The figures, reportedly drawn from OpenRouter traffic data, represent perhaps the starkest illustration yet of how open-weight economics have rewritten the competitive map.
The trajectory is jarring when you consider where things stood just 18 months ago. In early 2025, DeepSeek's R1 and Qwen's 2.5 series were curiosities — impressive benchmarks from labs most American developers couldn't name. By mid-2025, Qwen, DeepSeek, and Yi had become default choices for cost-sensitive inference workloads. By 2026, they are arguably the backbone of the American startup ecosystem. The reason is straightforward: these models are free to run, free to fine-tune, and increasingly competitive with proprietary alternatives on the tasks that matter most in production — structured extraction, code generation, and agentic tool use.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.