Amazon Bedrock adds cross-region routing for AI requests for EU customers

AWS has expanded Amazon Bedrock capabilities for European companies: the Cross-Region Inference (CRIS) mechanism automatically routes AI requests across multiple AWS regions while maintaining compliance with data confidentiality requirements. This capability is particularly relevant for organizations operating in heavily regulated EU industries.

Why European businesses need this

Access to the most powerful generative AI models is not a guaranteed option for European companies. Leading-edge models and high-performance GPU accelerators are unevenly distributed across AWS regions: new releases often appear first in American zones and only reach the EU months later. During peak load periods, individual regions become overloaded, leading to delays and API call errors.

In parallel, a strict regulatory environment is in effect: GDPR restricts cross-border transfer of personal data, and the recently enacted AI Act adds requirements for transparency and auditability of processing. The result is that businesses are forced to choose: either powerful models outside Europe, or compliance with legislation on a limited set of tools. CRIS eliminates this conflict.

The system automatically balances load across approved EU regions, providing access to the latest models without violating requirements.

How routing works

Amazon Bedrock with CRIS introduces the concept of "cross-region inference profiles"—a set of regions within which request processing is permitted. The developer specifies an allowed zone (for example, only EU regions: Frankfurt, Ireland, Stockholm), and Bedrock independently distributes traffic between them taking into account current load and model availability.

• Requests are routed automatically—no changes to application code are required
• Data is processed strictly within the specified geographic zone
• Latest models are available, not yet deployed in a specific local region
• The probability of failures due to individual region overload is reduced
• Billing is standard per Amazon Bedrock—no additional markup

All major model families on the platform are supported: Claude from Anthropic, Llama, Mistral, and others available through Bedrock. Switching between regions is invisible to the end application—the API interface remains unified.

Practice and limitations

Cross-region routing works at the Amazon Bedrock API level: companies already using the platform can enable CRIS without redesigning their architecture. For new customers, this means GDPR compliance is built into the infrastructure rather than provided through legal disclaimers in the contract. An important nuance: CRIS guarantees data remains within the selected cluster of regions, but does not exclude processing outside the customer's "home" region within that cluster. For regulated industries—finance, healthcare, insurance—this will require additional review with the legal department and a DPA (Data Processing Agreement). A separate issue is latency. Cross-region routing adds milliseconds of network delays. For most GenAI applications this is not critical, but real-time tasks—voice bots, interactive chats—should be tested separately.

"AWS customers need tools to leverage model availability and compute resources across multiple regions while maintaining security and confidentiality requirements,"

AWS explains in its official blog.

What this means

The emergence of CRIS removes one of the major operational barriers to adopting generative AI in the European corporate sector. Companies that delayed scaling AI projects due to regulatory uncertainty now have a ready-made infrastructure solution: full access to Bedrock's model portfolio within the EU. For the competitive landscape, this means the technical gap between American and European teams is beginning to narrow.