Amazon Bedrock now exposes Anthropic’s Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5 through an India geographic inference profile. The models run in the ap‑south‑1 and ap‑south‑2 regions, keeping data processing inside India while still leveraging Bedrock’s cross‑Region inference capabilities.
India geographic inference in practice
Bedrock’s cross‑Region inference profiles let a request originate from a source region and be automatically routed to one of the destination regions defined in the profile. The new India profile restricts routing to the two Indian regions, so prompts and responses may travel between ap‑south‑1 and ap‑south‑2 but never leave the country. This expands the compute pool available to a single request, helping maintain throughput during traffic spikes without the need to provision capacity in each region individually.
Operational impact
- Capacity planning: Engineers can rely on the broader pool of compute across both Indian regions, reducing the need for per‑region capacity reservations.
- Billing and quotas: All usage is billed and counted against the account in the source region, regardless of which Indian region actually performed the inference.
- Observability: CloudWatch metrics and CloudTrail logs are emitted only in the source region, simplifying monitoring and alerting configurations.
Security and data‑handling considerations
Data in transit between the source and destination Indian regions travels over the AWS private network with end‑to‑end encryption. The service follows a zero‑data‑retention model: inputs and outputs are not persisted by default. The only exception noted is that some models may trigger human review if automatic safety classifiers flag content, which could introduce limited data exposure. Customer data never resides in the destination region; it remains in the source region for the duration of the request.
Getting started
Practitioners can experiment with the new models directly in the Bedrock console’s Playground, selecting the appropriate inference profile (e.g., “IN Anthropic Claude Opus 5”). The same profile can be referenced when calling the bedrock-runtime endpoint via the Anthropic Messages API, the InvokeModel API, or the Converse API. All three APIs continue to support Bedrock Guardrails and intelligent prompt routing, now with the added locality guarantee.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
For AI engineers and platform teams, the India geographic inference profile removes a common compliance hurdle by ensuring data never leaves the country, while still offering the elasticity of multi‑region scaling. DevOps and SRE staff gain a simpler observability model because logs and metrics stay in a single region, and capacity can be balanced automatically across two Indian regions. Security engineers should verify that any downstream processes respect the zero‑data‑retention expectations and review the limited human‑review exception for flagged content. The next step is to validate the latency and throughput characteristics of the Indian profile against existing workloads and to update any automation that assumes a single‑region endpoint.
