AWS offers GPT access from Canada through regional routing

AWS’s October 20 announcement covers Canada Central and Calgary, with Bedrock handling destination selection.

For GPT‑6.1 Sol, the model card makes the distinction concrete. Its US geographic profile can send requests from either Canadian source to that Canadian region or to Northern Virginia, Ohio or Oregon. On bedrock-runtime, callers use an inference-profile identifier; direct in-region invocation is unsupported. The model supports Responses, Chat Completions and Converse there, so choosing the Canadian entry point does not establish a Canada-only processing boundary.

The routing guide explains how to inspect that boundary. ListInferenceProfiles returns profiles available from the intended source region. GetInferenceProfile then exposes destination regions in the returned model identifiers. For residency requirements, AWS says to check every intended source region and all returned destinations, rather than relying on the profile’s geographic prefix. An organization’s service control policy blocking any destination in the selected profile causes the request to fail.

Processing location also matters for retained content. Under AWS’s abuse-detection rules, classifier-flagged GPT‑6.1 Sol inputs and outputs can be retained for up to 30 days for automated offline checks. With cross-region inference, that retained data stays in destination regions. AWS processes it without sharing it with third-party model providers. Eligible customers can request full zero-data retention through their AWS account team.