AI model marketplace OpenRouter has officially launched U.S. in-region routing into general availability for business and enterprise customers, promising that requests sent through its U.S. endpoint are decrypted, processed and served entirely inside the country. The feature addresses a growing concern for companies using open-weight AI models: models from Chinese developers accounted for 41% of downloads on Hugging Face in the 12 months through February, ahead of the U.S. at 36.5%, while on OpenRouter specifically, open-weight models now represent roughly 60% of tokens consumed by U.S.-originating requests, with Chinese models constituting the majority of that volume.
The new routing capability works by filtering provider endpoints at the infrastructure level. When a company sends a request through OpenRouter's us.openrouter.ai endpoint, the request is decrypted on OpenRouter infrastructure inside the U.S. and the pool of providers is restricted to endpoints OpenRouter has verified as operating domestically. If no compliant U.S. provider can serve the requested model, OpenRouter returns a 404 error rather than routing the request elsewhere. Chinese models like DeepSeek V4 Pro, Kimi K3, and GLM 5.2 remain accessible through the service because U.S.-based providers including Baseten, Fireworks, and Azure host them from American data centers. Companies can also enforce the regional restriction through OpenRouter's Guardrails feature at the workspace, team, or API-key level.
According to Cailee Moberg from OpenRouter's product team, ["models from Chinese labs are still most of the [open-weight model] volume, and procurement approval for those models can be difficult."](https://thenewstack.io/openrouter-us-region-routing/) The service maintains a live list of models eligible for U.S. in-region routing, spanning proprietary frontier models from OpenAI and Anthropic alongside open-weight models from major Chinese labs. Moberg writes that "in-Region Routing allows teams with data residency requirements to get the price and performance gains from Chinese open-weight models," noting that when a U.S. or EU provider hosts a model, requests go to that provider and the lab itself isn't involved. The feature had been quietly available in limited form since early August for enterprise customers by request, while European in-region routing has been available since October 2025.
The launch comes as data sovereignty concerns climb the corporate priority list. In its 2026 State of AI in the Enterprise report, Deloitte found that 77% of companies now factor country of origin into their vendor selection, while nearly 60% construct their AI stacks primarily with local vendors. Model routers like OpenRouter have become valuable precisely because they sit between applications and models, deciding which provider serves each request and enabling developers to send simpler tasks to cheaper models while reserving expensive frontier systems for complex work. That intermediary position now lets OpenRouter guarantee where data is processed without requiring customers to manage deployments themselves. Companies could already keep these models inside the U.S. by self-hosting or contracting directly with a domestic provider, but OpenRouter's routing offers that residency guarantee through a single API. While the feature doesn't change where DeepSeek, Kimi, or GLM models are developed, it does control which copies of those models U.S. customers can access and where their prompts are handled. For enterprises balancing cost savings from open-weight models against procurement policies and data residency mandates, geographic routing may prove essential infrastructure rather than optional protection. The question is no longer whether companies will use models built abroad, but whether they can keep the data those models touch inside national borders.