OpenRouter Launches US Data Residency Controls as Chinese AI Models Dominate Usage

During August, open-weight models represented roughly 60% of the tokens consumed by US-originating requests on OpenRouter, with Chinese-developed systems making up the bulk of that traffic. The company has now rolled out a new capability designed to address data residency concerns: in-region routing that guarantees requests are decrypted, processed and served only within the United States.

The appeal of open-weight models is well established in the industry: organizations can obtain the model weights, adapt them to their needs, deploy them on their own infrastructure, and maintain tighter control over data processing locations—typically at substantially lower costs than proprietary alternatives. Open-weight systems are now estimated to lag leading frontier models by approximately four to five months in capability.

The sector's momentum is evident in major corporate moves. Nvidia, currently the world's most valuable company, announced in early September that it would acquire Hugging Face—the platform hosting over three million AI models—for $12.9 billion, committing to preserve openness across different model architectures, cloud providers and computing environments. Additionally, Nvidia disclosed this week how its proprietary Nemotron open-weight model is being deployed alongside Palantir technology to optimize its global supply chain operations.

Yet this expansion raises critical security considerations. Greg Brockman, president of OpenAI, recently cautioned that increasingly sophisticated open-weight models—citing China's GLM-5.3 specifically—could "significantly accelerate the threat landscape" by making models with advanced cyber capabilities freely available for modification and deployment.

For enterprises using these models through intermediary platforms, a more immediate concern emerges: understanding where their data travels when they access models originating from Chinese developers.

China and the open-weight factor

Hugging Face reported in February that models created by Chinese developers captured 41% of downloads over the preceding year, surpassing the United States at 36.5%. On OpenRouter's platform, open-weight models now represent approximately 60% of tokens consumed through US-based requests, with Chinese-developed systems constituting the dominant share.

Responding to this dynamic, OpenRouter has transitioned its US in-region routing feature from limited enterprise availability to general availability for business and enterprise customers. The service ensures that requests routed through the company's US endpoint are decrypted, processed and delivered entirely within American territory—or rejected if domestic processing cannot be guaranteed.

The capability existed in preliminary form previously, with OpenRouter updating its documentation in early August to indicate US in-region routing was accessible to enterprise customers upon request. The company also maintains European in-region routing, which has been operational since October 2025.

Founded in early 2023 by Alex Atallah, formerly CTO at OpenSea, OpenRouter functions as a unified interface to the fragmented AI model marketplace, enabling developers to switch among hundreds of models from different providers through a single API. Payments processor Stripe recently disclosed plans to acquire OpenRouter in a reported $8 billion transaction, joining other companies such as Cursor, Ramp and Meta in building proprietary model routing infrastructure.

Model routers have become highly sought after primarily for economic reasons. Developers have historically locked applications into specific models, whereas a router can make routing decisions per request, directing simpler tasks to less expensive models while preserving costly frontier systems for work requiring their capabilities.

OpenRouter's intermediary position enables its residency controls: because the platform already determines which provider handles each request, it can now constrain that selection to US-based provider endpoints.

Keeping Chinese models inside the US

In announcing the feature Wednesday, Cailee Moberg from OpenRouter's product team acknowledged that while US-developed models from Nvidia and Thinking Machines are contributing to open-weight expansion, Chinese models dominate consumption patterns and create procurement challenges for data-conscious organizations.

Models from Chinese labs are still most of the [open-weight model] volume, and procurement approval for those models can be difficult.

Cailee Moberg, OpenRouter

Deloitte's 2026 State of AI in the Enterprise report underscores this concern, finding that sovereign AI considerations are gaining prominence: 77% of companies now incorporate country of origin into vendor decisions, while nearly 60% construct their AI infrastructure "primarily with local vendors."

This context clarifies OpenRouter's expansion of in-region routing for US customers. Moberg highlights DeepSeek V4 Pro, Kimi K3 and GLM 5.2 as concrete examples. All three are accessible through US In-Region Routing because Baseten, Fireworks and Azure host them from US data centers. While companies could previously maintain these models domestically through self-hosting or direct engagement with US providers, OpenRouter's routing capability delivers that residency assurance without requiring customers to manage their own deployments.

OpenRouter publishes a current list of models qualifying for US in-region routing, spanning proprietary frontier systems from OpenAI and Anthropic alongside open-weight offerings from major Chinese laboratories.

In-Region Routing allows teams with data residency requirements to get the price and performance gains from Chinese open-weight models.

Cailee Moberg, OpenRouter

Moberg adds: "When a US or EU provider hosts a model, requests go to that provider and the lab is not involved."

The technical implementation operates at the routing layer. Through OpenRouter's standard global endpoint, requests can be fulfilled by any eligible provider across regions, meaning even US-developed models do not guarantee domestic request processing. The us.openrouter.ai endpoint decrypts requests on OpenRouter infrastructure located in the US and restricts the provider pool to endpoints OpenRouter has verified as operating domestically.

If no compliant US provider can deliver the requested model, OpenRouter returns a 404 error. Organizations can enforce regional constraints through OpenRouter's Guardrails at the workspace, team or API-key level, with tools that would transmit prompt data outside the US disabled on the regional endpoint.

While in-region routing does not alter where DeepSeek, Kimi or GLM models are developed, it does determine which geographic instances US customers access and where their prompts are handled throughout the request lifecycle.