Chinese AI models dominate OpenRouter’s US token consumption. It can now guarantee that traffic stays entirely in the US.
The New Stack Paul Sawers
Chinese AI models now make up most of OpenRouter’s US traffic. Now US customers can force that traffic to stay inside the country, or it gets blocked.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Chinese open-weight models are no longer some niche side bet. On OpenRouter, they now account for the majority of tokens consumed by US-originating requests, and the company is responding by giving customers a hard geographic boundary around that traffic.
OpenRouter has moved its US in-region routing into general availability for business and enterprise users. Requests sent through its US endpoint are decrypted, processed and served entirely inside the country, and if that cannot happen, the request is rejected. The company says the feature had been available quietly in limited form before this rollout, and it already offers a separate European in-region option.
That matters because the routing layer is where OpenRouter earns its keep. The company sits between developers and a long list of model providers, choosing which model serves each request. That makes it possible to send cheaper prompts to cheaper systems and reserve the pricier frontier models for the jobs that need them. It also makes residency controls possible. With the US endpoint, OpenRouter filters the provider pool down to approved US-based endpoints before a request leaves its infrastructure.
Cailee Moberg, who works on OpenRouter’s product team, says the company is seeing strong demand from customers who need to know where their data goes, especially because Chinese labs still make up most of the open-weight volume. She points to DeepSeek V4 Pro, Kimi K3 and GLM 5.2 as examples available through the US routing setup because they are served by Baseten, Fireworks and Azure from US data centers. If no compliant US provider can serve the request, OpenRouter returns a 404. That is blunt, but it is also the point.
The broader backdrop is easy to see. Hugging Face data from February showed Chinese developers ahead of the US in downloads over the previous 12 months, and OpenRouter says Chinese models dominate its own US token flow. Companies can already self-host or use a US provider directly. OpenRouter is betting that many would rather keep the convenience of a router and still draw a border around the prompts.
My take — AI-written commentary, not fact-checked reporting
This is the kind of boring product feature that actually matters. Everyone loves talking about model capability; far fewer want to talk about where the prompt went after procurement signed off. OpenRouter is reading the room correctly: data residency is becoming the adult in the building, and the open-weight boom doesn’t get a free pass just because it’s cheaper.
Read more about this at: The New Stack
Related stories
Faced with less compute and fewer tokens, Chinese AI labs are tightening the gap with the US by just being more efficient
Fortune ·
24