AWS Consulting ·
Generative AI on AWS with data kept in India: Bedrock in-country inference
AWS announced Claude models on Amazon Bedrock in India with geographic cross-Region inference across Mumbai and Hyderabad. Here is what in-country inference covers for regulated buyers — and what Region choice still does not settle.
By Upvantix team
Amazon Bedrock can keep generative AI inference inside India. Since AWS’s announcement of 29 September 2026, Anthropic’s Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 are available on Amazon Bedrock in India, with the regional endpoint served through geographic cross-Region inference across the Mumbai and Hyderabad Regions, so data is processed in India Regions. The global cross-Region inference option remains available as well.
In plain terms: if you call the India geographic path, Bedrock can route inference across Mumbai and Hyderabad and keep that processing inside India. That is different from global cross-Region inference, which can send requests to other commercial Regions. Choosing an India Region or an India geographic profile is one important control for GenAI residency — it does not, by itself, finish a DPDP or contractual residency analysis.
What AWS announced on 29 September 2026
AWS What’s New (29 September 2026) says Anthropic’s Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 are available on Amazon Bedrock in India. The announcement highlights customers in financial services, healthcare and the public sector who need to process data within a specific geography and keep inference in-country.
In India, the regional endpoint is served through geographic cross-Region inference across Mumbai and Hyderabad. Customers can access these models while processing data in India Regions, and they can still use global cross-Region inference if that fits their design.
What in-country inference covers — and what it does not
According to AWS, India geographic cross-Region inference keeps inference processing in India Regions (Mumbai and Hyderabad). That is the control to evaluate when the buyer question is whether prompts and model outputs for this Claude-on-Bedrock path can stay inside India.
Region or geography choice does not automatically answer every residency, contractual or DPDP question. Logging, monitoring, retrieval stores, fine-tuning data, human review tools, third-party connectors and support access can still leave India unless you design them not to. Treat the Bedrock announcement as one control in a wider design review.
Geography-based versus global cross-Region inference
Amazon Bedrock documents cross-Region (system-defined) inference profiles. Geography-tied profiles keep a fixed destination Region list. Global profiles can route more broadly across commercial Regions.
For India buyers with an in-country requirement, the India geographic path is the one to evaluate against the global option — and to verify with current AWS documentation (including GetInferenceProfile) for the source Region you will call from. Available profiles and destination Regions can differ by source Region.
Checklist: decide before production prompts land on Bedrock
Use this as a buyer checklist. It is not legal advice, and it does not give latency, cost or throughput figures.
| Criterion | Check | Why it matters | Owner |
|---|---|---|---|
| Models and profile IDs | Confirm Claude models and the India geographic inference profile you will call | Availability and routing differ by profile | Cloud / platform architect |
| In-country vs global routing | Prefer India geographic cross-Region when in-country inference is mandatory | Global profiles can leave India | Security / compliance |
| Prompt and output stores | Map every store that holds prompts, outputs, embeddings and evaluation data | Inference residency ≠ storage residency | Application owner |
| Logs and observability | Decide where invocation logs, traces and guardrail events are retained | Support and monitoring paths often cross borders by default | SRE / SecOps |
| DPDP and contract wording | Align the design with DPDP and customer contract language, then re-check AWS docs before go-live | Model availability can change | Legal + cloud |
How this relates to DPDP and other Upvantix guides
For the broader India residency and DPDP design questions beyond GenAI, see the AWS data residency and DPDP guide. For Upvantix GenAI services, see the AI/ML & Agentic AI Services page. Neither page replaces a formal legal review.
Where Upvantix fits
Upvantix is an AWS Select Tier Services Partner. We help India mid-market teams read current AWS Bedrock residency options, map GenAI data flows, and plan consulting work based on what AWS documents today.
Common questions about Bedrock GenAI residency in India
Can Claude models on Amazon Bedrock keep inference in India?
Yes for the India path AWS announced on 29 September 2026: geographic cross-Region inference across Mumbai and Hyderabad, so processing stays in India Regions. Always confirm the profile and models you will call in current AWS documentation.
Which Claude models did AWS list for India on that date?
Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 on Amazon Bedrock in India.
Is global cross-Region inference the same as India geographic inference?
No. Global cross-Region inference can route more broadly across commercial Regions. The India geographic path is aimed at in-country processing across Mumbai and Hyderabad.
Does choosing Mumbai alone finish a DPDP analysis?
No. Region choice is one control. Logs, stores, connectors and people access still need design review.
What AWS partner status does Upvantix hold?
Upvantix is an AWS Select Tier Services Partner.