Exa Launches EU-Hosted AI Search in Stockholm, Cutting Latency by 100ms

Exa has launched a European deployment of its AI search engine, allowing enterprise customers to process queries, fetch page contents, and store results within the EU. The service runs in AWS Stockholm ( eu-north-1 ) and is available through the new api.eu.exa.ai endpoint, according to the EU launch post .

The release targets organizations whose procurement, security, or GDPR requirements prevent queries from reaching a US-hosted service. Europe already accounts for a substantial share of Exa’s API traffic, and customers had requested a regional deployment.

The index lives in Stockholm

Exa copied its main web index into the EU and says it maintains petabytes of data across both regions in real time. API routing, vector databases, model inference, reranking, highlights, and page retrieval all run through infrastructure in Stockholm.

Keeping two search regions aligned requires more than copying stored pages. Crawlers must feed both indexes, embeddings must produce consistent numerical representations of content, and rerankers must order results similarly. Drift in any component can change which pages users receive.

EU quality stays close to US

Exa evaluated both regions with the same benchmark suite. The largest published difference was 2.8 percentage points on ATLAS for Exa Auto, while other benchmarks showed differences ranging from 0 to 2.4 percentage points. Exa also reports that European users reduced request latency by more than 100 milliseconds when querying the Stockholm endpoint.

Keys cannot cross regions

Exa enforces residency through account, endpoint, and response-level controls designed to prevent requests from reaching another region:

• Region-bound keys. API keys assigned to an EU team work only with the EU endpoint. A misconfigured client receives an error.
• Rejected unsupported calls. The endpoint refuses unavailable features before metering them and does not retry requests in another region.
• Auditable responses. Every response includes the x-exa-processing-region: eu header, which identifies where the request was processed.
• Zero retention by default. EU teams start with request-content retention disabled.

The residency boundary has a documented exception: account and identity metadata, along with aggregated service metrics, may be stored outside the EU. Organizations whose policies cover user identifiers should include that exception in their security review and data-processing agreement.

The first release has limits

The EU region currently supports the Search and Contents APIs, with auto and lite search types. Exa plans to add more search types, followed by Agents, Websets, and Monitors. Applications that depend on those products cannot yet move their full workload into the region.

After Exa provisions an EU team, Python clients need the regional API key and base URL:

Residency clears blocked workloads

Exa has a direct commercial incentive to expand in Europe after reporting a $100M run rate . EU data residency has also become a standard procurement requirement for AI vendors selling to banks, legal firms, healthcare companies, and other organizations that handle personal or regulated data. OpenAI, Anthropic, and major cloud providers have introduced similar regional controls.

Developers building agents and retrieval-augmented generation systems can now keep European search traffic within the region while retaining Exa’s existing API shape and comparable benchmark performance. The regional endpoint can replace separately maintained EU search indexes and US-bound query paths, provided the current feature set and metadata exception meet the organization’s requirements.