The Pulse
AWS Keeps Claude Inference in India, Routing Between Mumbai and Hyderabad
AWS has made Claude Opus 5, Sonnet 5 and Haiku 4.5 available on Amazon Bedrock through India-only cross-Region inference. Requests can route between AWS’s Mumbai and Hyderabad regions, so processing stays in India but is not confined to one
AI.info Team · Written with AI ·
AWS is offering Indian customers local processing for three Claude models, but not a guarantee that an inference request stays in one data center. Announced on September 29, the Amazon Bedrock option routes requests between AWS’s Mumbai and Hyderabad regions, keeping inference within India while allowing prompts and results to move between the two locations.
The change adds an India-only route for Anthropic’s Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5. AWS says customers can also continue using the global cross-Region inference already available on Bedrock. The new geographic profile is aimed at workloads that need processing to remain within the country, while drawing on capacity in both Indian regions.
India-only routing, not single-region processing
Amazon Bedrock’s geographic inference profile accepts a request in a source region and routes it to a destination region in the profile. For India, AWS identifies the permitted regions as ap-south-1 in Mumbai and ap-south-2 in Hyderabad. The profile does not route inference outside India, but it can send a request from one of those regions to the other to use available capacity.
That distinction matters for organizations whose rules require domestic processing but do not require a workload to stay in one city or one AWS region. AWS describes the routing as a way to draw on a wider pool of compute and support throughput under heavier demand. It says traffic travels over its network with encryption in transit.
AWS’s authors write, “Your input prompts and output results might move between those two Regions.” Aamna Najmi, Ayan Ray, Eugenio Soltero and Sofian Hamiti, AWS The caveat means the announcement describes country-level residency, not a promise that data remains inside the region where an application starts its request.
What AWS says about storage and review
AWS says customer data is not stored in the destination region and remains in the source region. It also says Bedrock uses a zero-data-retention model by default, meaning it does not store model inputs or outputs. Those storage statements should be read separately from the routing description: prompts and generated results can move between Mumbai and Hyderabad while a request is being processed.
The zero-retention description has a qualification. AWS says certain models may require human review if automated safety classifiers flag content. Its announcement directs customers to Bedrock’s data-retention documentation for further detail, so organizations with strict handling requirements will need to assess that condition alongside their own policies.
Three APIs and a console playground
Customers can try the models in the Amazon Bedrock console’s text playground or call them through Anthropic’s Messages API, Bedrock’s InvokeModel API or the Converse API. The India geographic profile is available through the bedrock-runtime endpoint. AWS’s examples use the India profile IDs with the Mumbai region as the source; the profile can route to Hyderabad when its defined destination and available capacity call for it.
Billing and quota use are tracked against the account in the source region, AWS says, and CloudWatch and CloudTrail logs are recorded there. The setup offers Indian customers a domestic processing path across two AWS regions, with the boundary set at India’s borders rather than at a single regional endpoint.