AI data residency has quietly blocked a lot of AI projects in India. Banks, insurers and government teams liked what Claude could do, but they could not send customer data abroad to get it. As of October 5, 2026, that blocker is gone for one major route.

Anthropic says Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 are now live with in-country inference in India through Amazon Bedrock. Requests sent through the India endpoint are processed inside the country.

I'm Nitin Kumar, Lead Project Coordinator at Debugged Pro. We build software and AI features for clients in India and the USA, and "where does our data go?" is now the first question in almost every AI kickoff. So I read the announcement, the AWS technical post and the press coverage closely. Here is what changed and what agencies should do about it.

What Anthropic and AWS Announced

The news came from Bengaluru on Monday morning. As The Times of India reported on Claude's in-country inference launch, a prompt sent to Claude through the India endpoint, and the model's reply, are both processed on servers located in India.

Here are the key facts:

Models: Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5.
Platform: Amazon Bedrock, AWS's service for building generative AI apps.
Where it runs: only between the AWS Mumbai (ap-south-1) and Hyderabad (ap-south-2) Regions.
Who asked for it: financial institutions, government bodies and large enterprises with strict data rules.
Early testers: Reliance, CRED and others used it in a private preview.

Irina Ghose, Managing Director for India at Anthropic, said, "In August, we committed to bringing Claude inference to India, and we are delivering on that promise." She added that customers in regulated sectors and government "want to use Claude on their most critical work, with requests processed in the country."

Sandeep Dutta, President of AWS India and South Asia, put the stakes plainly. He said data residency "is the parameter that decides whether AI stays confined to pilots or reaches full production."

That line matches what we see. Many Indian clients have run Claude pilots on test data for months. Moving to real customer records was where the projects stalled.

Irina Ghose, Managing Director India at Anthropic, who announced AI data residency for Claude in India — Irina Ghose, Managing Director, India, Anthropic. Image credit: The Times of India

How In-Country Inference Works on Amazon Bedrock

The technical details sit in the AWS guide to Claude in-country inferencing in India. It is worth reading in full before you plan a build. Here is the short version.

Geographic cross-Region inference

Bedrock uses something called a cross-Region inference profile. Your app calls Bedrock in a source Region, and Bedrock routes the request to one of the Regions in the profile. The India profile only includes Mumbai and Hyderabad, so prompts and outputs may move between those two, but they stay in India.

AWS says this spreads requests over a bigger pool of computing power. That helps keep speed steady when traffic spikes.

What stays where

Customer data is not stored in the destination Region. It stays in the source Region.
Bedrock does not store prompts or outputs by default. AWS calls this a zero data retention model. It notes one exception: some models need human review by AWS if automatic safety checks flag content.
Billing and quotas are tracked in your source Region, whichever Region did the work.
CloudWatch and CloudTrail logs are written only in the source Region, so monitoring stays in one place.

Which APIs work

The India profile works on the bedrock-runtime endpoint. It supports Anthropic's Messages API and Bedrock's own InvokeModel and Converse APIs. Bedrock Guardrails also work with it. In code, you point to a model ID with an "in." prefix, such as in.anthropic.claude-sonnet-5, from the Mumbai Region.

Picking the India (IN) inference profile for Claude Opus 5 in the Bedrock console. Image credit: AWS

One thing to watch on the direct Claude API

If your team calls Claude through Anthropic's own API, check the settings carefully. Anthropic's data residency documentation currently lists only "us" and "global" for its inference_geo option. So for processing inside India, the Bedrock India profile is the route to use today.

Why AI Data Residency Matters So Much in India

India is a big Claude market. Ghose told The Times of India that India is Anthropic's second-largest market for Claude.ai. Software development makes up 45.2% of work-related tasks Indian users bring to Claude, the highest share of any market.

But usage by developers is one thing. Putting a model inside a bank's core workflow is another. Three things make AI data residency a deal breaker for many Indian clients:

Sector rules. The Reserve Bank of India has required since 2018 that payment system data be stored only in India. Banks and fintechs treat anything near that data with great care.
The DPDP Act. India's Digital Personal Data Protection Act, 2023 sets rules for handling personal data. It also lets the central government restrict transfers of personal data to certain countries. You can read the official material on MeitY's data protection framework page.
Board-level risk. Even where the law allows a foreign transfer, many risk teams simply say no. Local processing removes a long debate.

Mohit Kapoor, CTO of the Mahindra Group, said in the launch release that "with the DPDP Act making data sovereignty a business imperative," keeping data and AI workloads in India lets the group scale Claude with confidence.

Who Is Already Using Claude in India

The customer list shows how far adoption has spread. According to the launch release and coverage in The Hindu BusinessLine and The Times of India:

Banking and payments: Kotak Bank, Axis Bank, IndusInd Bank and NPCI use Claude. NPCI is building its agentic AI platform, AiNxt, on Claude.
IT services: TCS is rolling Claude out to 50,000 associates. Infosys has an Anthropic Center of Excellence and combines Claude with its Topaz platform. Cognizant, LTM and LTTS are also building with it.
Large groups: Reliance, Mahindra and Mahindra, and Godrej Industries are deploying Claude across their businesses.
Digital companies: CRED, Razorpay, Swiggy, Zomato, Invideo, Freshworks and Emergent build Claude into products and engineering.

Anthropic also opened a Bengaluru office in February, its second in Asia after Tokyo. It certified around 5,000 people at its first partner certification event in Bengaluru in August. It is working with Karya to improve Claude across 10 Indian languages.

We covered the public sector side of this story earlier in our look at Claude for government projects. The India launch makes that guide far more practical for Indian clients.

What AI Data Residency Means for Agencies in India and the USA

This is not only news for big enterprises. It changes how small and mid-sized agencies scope, price and pitch AI work.

For agencies in India

Regulated clients are now open for business. Projects for NBFCs, insurers, hospitals and state departments that stalled at the compliance review can move again. Expect old proposals to come back with "can we do this now?"

Your architecture has to prove it. Saying "we use the India endpoint" is not enough. Risk teams will want a data flow diagram that shows every hop.

For agencies in the USA

Indian clients and offshore teams need a different setup. If you build for a client in India, a US-hosted stack may not pass their review. Plan for an India deployment from day one.

Mirror the pattern at home. US clients ask the same question about keeping data in the US. The same habits apply: pin the Region, log everything, and document it.

For everyone

Residency is only one layer. Your vector database, file storage, logs, backups and any other AI services in the flow also need to stay in the right place. One stray call to a global endpoint can undo the whole design.

A Practical AI Data Residency Checklist for Client Builds

Here is the checklist we now use when an Indian client asks for Claude with local processing:

Map the data first. List every type of personal or financial data the app touches, and every system it passes through.
Pick the source Region. Run your Bedrock calls from Mumbai or Hyderabad and use the "in." inference profile IDs.
Block the global profile. Use IAM policies so your app roles cannot call global inference profiles by mistake.
Keep the rest of the stack local. Put databases, embeddings, file storage and backups in Indian Regions too.
Turn on logging. Use CloudTrail and CloudWatch in the source Region, and keep the logs for your client's audit needs.
Add guardrails and human review. Use Bedrock Guardrails and send risky actions, like refunds or account changes, to a person.
Remove what you do not need. Mask or drop personal data from prompts when the task does not require it.
Write it down. Give the client a short residency note covering Regions, data flows, retention and who has access.

If you plan to run several agents for one client, our guide on how to scale enterprise AI agents covers access control and monitoring in more depth.

Quick comparison table

Question	Bedrock India profile	What agencies should check
Where is inference processed?	Mumbai and Hyderabad only	App calls the "in." profile from an India Region
Are prompts stored?	Not by default	Your own logs and databases
Where are logs kept?	Source Region only	Retention period and access
Which models?	Opus 5, Sonnet 5, Haiku 4.5	Right model for cost and speed
Is the app compliant?	Not on its own	Consent, contracts and security under the DPDP Act

A short example

Picture a mid-sized NBFC in Pune that wants an AI assistant for loan officers. It reads customer documents, summarises them and drafts notes. Last year, its risk team blocked the pilot because prompts would leave India.

Now the agency can call Claude Sonnet 5 through the India profile from Mumbai. It keeps documents in an S3 bucket in the same Region, blocks global endpoints and logs every call. A loan officer still signs off on every decision. That is a design a risk team can actually approve.

Limits and Open Questions

This launch is a real step, but it is not the end of the work. Keep these points in mind:

It covers Bedrock, not every product. Check each tool in your stack separately, including Claude apps and the direct API.
Model choice is limited to three models on the India profile for now.
Safety review is an exception to zero retention. AWS notes that flagged content may need human review for some models, so read the Bedrock data retention terms.
Compliance is still your job. In-country processing helps, but it does not replace consent, security and breach processes.

If you want help planning an AI build that your client's compliance team will sign off, our software development consulting team works with businesses in India and the USA on exactly this kind of project.

Key Takeaways

Claude Opus 5, Sonnet 5 and Haiku 4.5 now support AI data residency in India through Amazon Bedrock.
Requests on the India profile are processed only in the Mumbai and Hyderabad Regions.
Bedrock does not store prompts or outputs by default, and logs stay in the source Region.
Regulated Indian clients, like banks and insurers, can now move many AI pilots toward production.
Agencies must keep the whole stack local and document it, because residency alone is not compliance.

FAQ: AI Data Residency and Claude in India

What does in-country inference for Claude in India mean?

It means that when you call Claude through the India endpoint on Amazon Bedrock, your prompt and the model's response are processed on servers in India. AWS says requests route only between its Mumbai and Hyderabad Regions.

Which Claude models are available with AI data residency in India?

Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 are available through the India geographic cross-Region inference profile on Amazon Bedrock.

Can I keep inference in India using Anthropic's own Claude API?

Not at the time of writing. Anthropic's data residency documentation lists only US and global options for its inference_geo setting. For processing inside India, the route announced on October 5, 2026 is Amazon Bedrock with the India inference profile.

Does in-country inference make my app DPDP Act compliant?

No. It helps with one part of the picture, which is where the AI processing happens. You still need consent, purpose limits, security controls, breach handling and clear contracts under the Digital Personal Data Protection Act, 2023.

How can an agency help a client move an AI pilot to production in India?

An agency can map where every piece of data flows, set up the Bedrock India profile with locked-down access, keep logs and storage in Indian Regions, add guardrails and human review, and write it all up for the client's risk and compliance team.

About the Author

Nitin Kumar is Lead Project Coordinator at Debugged Pro. He plans software and AI projects for clients in India and the USA, and works with delivery teams on scope, data handling and the safe rollout of AI features in production.