Claude AI can now process eligible Amazon Bedrock requests within India through a geographic inference profile that routes work between AWS’s Mumbai and Hyderabad Regions. AWS announced the availability on 29 September 2026; Anthropic described the deployment to Indian media on 5 October. The distinction matters: this is an enterprise Bedrock option, not a claim that every Claude.ai chat or Anthropic API request stays in India.
Key takeaways
- AWS says the India profile supports Claude Opus 5, Sonnet 5 and Haiku 4.5.
- Requests may move between Mumbai and Hyderabad, but the selected geographic profile keeps inference in India.
- A company still has to check its own configuration, contracts and legal duties; local inference does not automatically make a workload compliant.
- The launch was first announced on 29 September, before the 5 October India media coverage.
Claude AI in India: what actually launched
The first-party account is AWS’s 29 September technical announcement. It says customers can call Anthropic’s Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 through an India geographic cross-Region inference profile on Amazon Bedrock. AWS identifies Mumbai, ap-south-1, and Hyderabad, ap-south-2, as the two destination Regions. This is a specific routing choice on a cloud service, not a blanket change to every way people use Claude AI.
On 5 October, The Times of India interviewed Anthropic India managing director Irina Ghose and an AWS India solutions architect about the service. Business Standard reported a company statement and partner comments the same day. News9 separately examined the routing and cautioned that customers still need to assess their own compliance settings. These reports confirm the India story, while the AWS release fixes the launch date.
The practical change is straightforward. A team using Bedrock chooses the India geographic inference profile rather than a global profile. AWS can then send a model request to capacity in either of its two Indian Regions. The prompt and response may cross between Mumbai and Hyderabad, but AWS says this selected profile does not send inference outside India. An organization should verify that it has actually chosen the India profile; simply working from an office in India does not establish local processing.
Why the two-region design matters
A large AI model needs substantial computing capacity. If a provider ties every request to one data centre, a surge in demand can make that location a bottleneck. AWS’s geographic cross-Region design gives a customer access to a broader pool across two Indian Regions. AWS describes this as a way to maintain throughput and performance during busy periods. That is a vendor explanation of the design, not an independently measured guarantee about response times for any particular customer.
A request begins in the customer’s source Region. AWS says it can then route the request to a destination Region defined by the India inference profile. The input and output may move between the two Regions. AWS says cross-Region transit uses its network with encryption in transit. It also says billing, quotas and CloudWatch and CloudTrail entries remain associated with the source Region. This separation between where the model runs and where the account records activity is useful for architects preparing audit documentation.
The exact choice of profile is central. AWS also offers global cross-Region inference for certain Claude models. That is a different routing option. In its March explanation of global availability, AWS described worldwide capacity sharing. The September India geographic option was added so customers with a need for local processing could select a bounded geography. Enterprises should record the profile they invoke and test that their production configuration matches their stated policy.
What data residency does and does not establish
Local inference answers one narrow question: where does this selected model execution take place? It does not answer every question about an application. A developer can still send data to other APIs, export logs, configure overseas storage, or permit staff in another country to access records. An end-to-end privacy review must trace the entire workflow. That includes prompts, attachments, retrieval databases, monitoring, support access, retention, and the response after it returns to the application.
AWS says customer data remains in the source Region rather than being stored in the destination Region used for cross-Region execution. Its Bedrock announcement also describes a default zero-data-retention model for inputs and outputs. Yet AWS flags an exception: some models may require human review when automated safety classifiers mark content. A procurement or compliance team should inspect the applicable Bedrock terms and the relevant model documentation instead of treating “zero retention” as an unconditional promise.
The regulatory language in the earlier version of this article was too broad. India’s Digital Personal Data Protection framework is not a simple blanket command that every category of personal data must be processed domestically. Some sectors impose their own requirements, and a particular institution’s obligations also depend on the records and services involved. For example, RBI’s payment-data storage direction concerns payment-system data; it should not be presented as a universal rule for all bank AI use. The India inference profile may help an organization satisfy a requirement it has identified, but it does not certify legal compliance on its own.
News9 made the same practical distinction in its independent report: organizations need to check their own logging, retention, encryption and review policies. This is particularly important for banking, health, government and other sensitive applications. A team should get counsel and security reviewers to approve a documented data flow before moving live customer information to a model, regardless of how the service is marketed.
The confirmed models and the model-name error
The models AWS names for this rollout are Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5. The earlier duplicate post on this site named Claude 3.5 Sonnet and Claude 3 Opus as the launch lineup. Those model names do not match AWS’s first-party September announcement. Availability can change, so buyers should consult the current AWS regional model list when implementing a workload.
Also distinguish a Claude model from the channel used to reach it. Claude.ai is the consumer and team-facing product. Amazon Bedrock is a cloud platform through which developers invoke models for their own applications. Anthropic’s direct API is another access path. The India geographic profile documented here is a Bedrock feature. Neither AWS nor the three original reports establishes that all Claude.ai consumer traffic or all direct Anthropic API calls now remain in India.
| Question | What the sources establish |
|---|---|
| First availability announcement | AWS published it on 29 September 2026. |
| Models named | Claude Opus 5, Sonnet 5 and Haiku 4.5. |
| Processing geography | Selected India profile routes between Mumbai and Hyderabad Regions. |
| Consumer Claude.ai traffic | Not established by the Bedrock announcement. |
| Automatic legal compliance | Not established; each organization must review its own workload. |
Why Indian enterprises asked for local processing
Anthropic told Business Standard that financial institutions, public bodies and large enterprises had repeatedly asked for local inference. That is a company characterization of customer demand. AWS India president Sandeep Dutta said data residency can decide whether regulated organizations advance AI work beyond pilots. His assessment is useful context, but it does not prove that any specific deployment is approved or in production.
In the Times of India interview, Ghose said the company heard the same request from regulated customers. She also described India as Anthropic’s second-largest Claude.ai market and said software development represents 45.2% of work-related tasks users bring to Claude in India. These are Anthropic’s own usage figures and definitions, reported by TOI, rather than an independent census of India’s whole AI market. Readers should not interpret the share as 45.2% of all Indian software work or all national AI usage.
Business Standard reported that Anthropic identified Reliance and CRED as early private-preview customers. The article also carried comments from TCS, Mahindra Group and NPCI about enterprise uses and governance. Those comments show interest and collaboration, but the report does not independently verify broad production use, measured latency savings, or a specific return on investment. This article therefore avoids turning a partner quote into a performance result.
The India angle: capacity choice, not a sovereignty certificate
India’s opportunity is a more usable enterprise option for teams that already run applications on AWS. The design gives them a documented way to call current Claude models while keeping the selected inference workload inside two Indian Regions. That is more specific than a general promise to “serve India.” It also gives architects a concrete item to test: the profile identifier in a production request.
The limitation is equally important. A regional model endpoint is one component in a bigger system. If an application retrieves documents from an overseas database, exports transcripts to another service, or sends support data abroad, the AI model’s local execution cannot repair the whole chain. Teams need an inventory of those flows before they can make a truthful data-residency claim to customers.
This is why the value of local inference is most likely to show up first in procurement and risk reviews. It gives teams a specific control they can document. Whether that translates into faster deployment, lower latency or stronger customer trust depends on the workload and configuration. AWS and Anthropic argue that the design improves resilience and removes a barrier for regulated sectors; those are vendor claims, not universal measured outcomes.
How the rollout fits Anthropic’s wider India push
Anthropic’s India expansion did not begin with this Bedrock feature. The company announced plans for a Bengaluru office in October 2025. It later discussed partnerships and Indian enterprise applications. Local model execution is a separate infrastructure step that connects those commercial plans to a specific cloud deployment route.
Lapaas Voice has also covered Claude’s rupee subscription pricing in India. That consumer-facing pricing change should not be confused with Bedrock’s enterprise inference geography. For developers following the product itself, our Claude Code Mods report describes a different update to Anthropic’s software tooling. Each development serves a different user and deserves its own scope.
The earlier duplicate versions of this story blurred those boundaries. They also treated possible future company deals, automatic regulatory clearance and precise latency as established facts. None of the checked primary or original reporting supplies those claims. The corrected account keeps the confirmed infrastructure mechanics and leaves future adoption to be reported when companies disclose it.
What companies should verify next
Before a live deployment, an Indian enterprise should record which model and inference profile its code calls, then confirm the set of destination Regions in the current AWS documentation. It should review Bedrock data-handling terms, safety-review exceptions, IAM permissions, CloudTrail settings, application logs and any outside services that receive prompt or response data. Its legal team should map the actual data category to the applicable rule, rather than assuming every Indian record has the same localization requirement.
Teams should also run workload-specific tests for latency, cost and capacity. AWS says the two-region pool is designed to help with throughput, but the public announcement does not publish a universal millisecond improvement or a guaranteed price comparison. The useful benchmark is an organization’s own application under normal and peak load, measured against its prior routing choice.
Finally, the team should separate an internal pilot from a claim of production adoption. Anthropic and partners may describe trials and customer interest; a bank or public body should only be listed as live on the relevant use case when that institution or a reliable independent report confirms it. This avoids overstating what an infrastructure launch has already achieved.
Frequently asked questions
Did Claude AI launch in India on 5 October?
AWS announced the India geographic Bedrock inference profile on 29 September 2026. Anthropic and Indian publishers discussed the availability again on 5 October. The later media coverage should not be mistaken for the first date the feature existed.
Does every Claude AI request from India stay in India now?
No such blanket claim follows from the sources. The confirmed capability is for customers selecting the India geographic inference profile on Amazon Bedrock. Consumer Claude.ai and other API pathways need separate verification.
Can a request move between Mumbai and Hyderabad?
Yes. AWS says the profile can route work between ap-south-1 and ap-south-2. That is cross-Region processing inside India, so “in-country” does not mean “one data centre.”
Does using the profile guarantee compliance with Indian law?
No. It can help satisfy a local-processing requirement where one applies, but a company must assess its own data, application architecture, provider terms and sector-specific duties.
Sources: AWS technical announcement, 29 September; Times of India original interview, 5 October; Business Standard company and partner report, 5 October; News9 technical and compliance report, 5 October.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
