OpenRouter In-Region Routing (2026): Keep AI Data in US/EU

Short answer: On September 9, 2026, OpenRouter launched US In-Region Routing, joining the EU routing it shipped in October 2025. Point your requests at us.openrouter.ai or eu.openrouter.ai and your prompts are decrypted inside that region, served only by providers running there, with no fallback outside it. The catch: it requires a Business plan (8% fee on credits, up from 5.5%) or Enterprise, so it matters to teams with compliance reviews, not to most solo chat users.

If your company has ever asked "where exactly does our AI data get processed," you know how short the usual answers fall. A model from a US lab can still run in a data center on another continent when a gateway load-balances for price. OpenRouter's announcement this week gives teams a concrete control for that problem. Here is what changed, how it works under the hood, what it costs, and whether it matters for your setup.

New to the one-key-for-every-model idea? Our OpenRouter key setup walkthrough covers creating a key and connecting it first.

What OpenRouter launched on September 9

Two regional endpoints now exist alongside the global one:

Endpoint	Behavior
https://openrouter.ai/api/v1 (global)	Requests load-balanced across providers anywhere, price-first
https://us.openrouter.ai/api/v1 (new, Sep 9 2026)	Decrypted inside the US, served only by US providers
https://eu.openrouter.ai/api/v1 (since Oct 2025)	Decrypted inside the EU, served only by EU providers

Three details from the announcement are worth your attention. First, your API key, request body shape, and model IDs are unchanged. You switch regions by changing the base URL, nothing else. Provider preferences, fallbacks, and privacy settings carry over from your account, and you can point one service at a regional endpoint while everything else stays global.

Second, the routing is fail-closed. If no in-region provider serves the model you asked for, the request fails with a 404 (No endpoints found supporting your data region) instead of quietly routing outside the region. For compliance purposes that failure mode is the whole point: a guarantee that leaks would be louder than a silent fallback.

Third, the regional catalogs are subsets of the global one. A model appears for a region only when at least one provider can serve it from inside that region. Named in the launch post: GPT-5.6, Claude Opus 5, and Gemini 3.6 Flash on the closed side, plus DeepSeek V4 Pro, Kimi K3, and GLM 5.2 served from US data centers by Baseten, Fireworks, and Azure. GLM 5.2 is also available in the EU via Mistral's data centers. Check the live, region-filtered model list before you plan around a specific model; the OpenRouter setup guide covers connecting your key to a chat app.

End-to-end routing versus inference-only routing

The most useful part of the announcement is a distinction OpenRouter draws explicitly, because it gives you two questions to ask any gateway that claims "regional routing":

1. Where is the request decrypted and processed? Some gateways pin inference to your region but decrypt and handle the request wherever their own servers run, which means your prompt exists in plaintext outside your region for part of the trip. Server-side tools get overlooked the same way: web search on a "regional" endpoint sometimes runs on global infrastructure.
2. Where do its tools run? OpenRouter says it evaluates the data jurisdiction of server tools before offering them regionally, and disables tools that would send data outside the region rather than falling back to global infrastructure.

OpenRouter's version is end-to-end: decrypted inside the region, processed there at every step, served only by in-region providers. When you evaluate a competing claim, ask those two questions and see whether you get a straight answer. Vague wording around "processing locations" usually means inference-only.

What it costs and who can actually use it

In-Region Routing is available on Business and Enterprise plans only. The public Business terms: an 8% platform fee on credit purchases instead of the standard 5.5%, with no monthly minimum, no seat license, and no contract. You upgrade with a credit card from plan settings, then send traffic to the regional endpoint. Enterprise adds custom terms through their sales team.

That pricing shapes who this is for. The fee is on credit purchases, not on inference, so a team spending $500 a month on credits pays about $40 in platform fees on Business versus $27.50 on standard. That is a small premium for a residency guarantee that would otherwise mean negotiating provider contracts per region yourself.

Enforcement tooling comes with it. OpenRouter's Guardrails let an admin restrict a workspace, team, or single API key to one region, so any covered request arriving on another hostname gets rejected. You can also set region pinning as an account-wide default in privacy settings. If you are the person who has to answer the residency review, that per-key enforcement is arguably more valuable than the endpoint itself.

What this means if you chat through OpenRouter with your own key

Most individual BYOK chat users are on standard credits, which means this launch changes nothing in your settings today. No action needed, no migration, no new fee. But three things are still worth knowing.

First, the price gap between open-weight and closed models keeps widening, and residency removes one objection to the cheap side. OpenRouter's numbers: the open-weight share of tokens rose from 26% to 60% in the US and 16% to 65% in the EU between September 2025 and August 2026, crossing 50% in early 2026. Chinese open-weight models served from US or EU data centers (DeepSeek, Kimi, GLM lines) now give regulated teams the cost gains without the data leaving the region, since the originating lab never touches the request.

Second, if you do move to Business later, connection is just a base URL. Any chat app that accepts a custom OpenAI-compatible endpoint can point at the regional host with your existing key. Our custom endpoint guide shows the pattern for apps that support it.

Third, key storage still matters independently of routing. Regional routing controls where your prompts run; it says nothing about where your API key lives. For that half of the picture, see what AES-256 actually protects and whether AI apps train on your data.

Which setup fits you

If you…	Best option
Must pass a US/EU data residency review	Business or Enterprise + regional endpoint + per-key Guardrails
Chat solo across many models, no compliance need	Standard credits on the global endpoint (5.5% fee, full catalog)
Want cheap open-weight models with residency	Regional endpoint + DeepSeek/Kimi/GLM served in-region
Just need one key for personal multi-model chat	Standard key, default settings; revisit if your employer asks questions

The honest summary: this is infrastructure news for teams, not a feature for individuals. Nobody needs to change anything today. But if your employer, client, or regulator has ever asked where your AI prompts go, the answer "decrypted in the US, served in the US, fails loudly otherwise" is now available for the price of a 2.5-point fee bump. That is a good deal for the teams that need it and irrelevant overhead for everyone else.

FAQ

Is OpenRouter In-Region Routing free? No. It requires a Business plan (8% fee on credit purchases, no monthly minimum) or Enterprise. Standard 5.5%-fee accounts use the global endpoint.

Does my API key change when I switch regions? No. Same key, same model IDs, same account settings. Only the base URL changes to us.openrouter.ai or eu.openrouter.ai.

What happens if my model is not available in-region? The request fails with a 404 and never leaves the region. Check the region-filtered models list before depending on a specific model.

Does in-region routing stop providers from training on my data? It controls where data is processed, not retention or training policy. Pair it with zero-retention settings and providers whose API terms exclude training to cover both halves.

---

Start your free trial → — 7 days, all providers, no credit card required.