Claude Now Runs Inside India: What In-Country Inference on Amazon Bedrock Means and How to Try It

Anthropic just made Claude Opus 5, Sonnet 5, and Haiku 4.5 available with in-country inference in India through Amazon Bedrock. Requests sent to the India endpoint are processed on servers inside the country. If you work at a bank, a government body, or any company that gets awkward questions about where your data goes, that one sentence changes what you can approve. Here is what shipped, what it does not change, and how to test it yourself today.

What Anthropic Announced on October 5

On October 5, 2026, Anthropic said its latest Claude models are now live with in-country inference in India via Amazon Bedrock. The lineup covers Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5.

The pitch is simple. Send a request to the India endpoint, and it gets processed on servers based in India. Anthropic also says Bedrock deployments in India come with the audit trails and access controls that risk and compliance teams ask for before anything goes into production.

This was not a surprise drop. Anthropic committed to local inference in India back in August, and early customers including Reliance and CRED tested it in private preview. AWS published the technical documentation on September 29, which is where the useful details live.

Why Data Residency Is the Real Story

Here is the thing. Most companies do not get stuck choosing between AI models. They get stuck at the compliance review, when someone asks where the prompts are processed and nobody has a clean answer.

AWS India’s president put it bluntly in the announcement: data residency is often the factor that decides whether AI stays a pilot or reaches full production. Mahindra’s CTO pointed at the same pressure from the legal side, citing India’s data protection law.

That law is real and the clock is running. India’s Digital Personal Data Protection Rules were notified on November 13, 2025, with obligations phasing in over 12 to 18 months. Most of the substantive requirements land by May 13, 2027. Companies that have been waiting for a cleaner answer on AI hosting now have about seven months to sort out their approach.

There is also a market signal here. NewsBytes reports that India is Anthropic’s second-largest market for Claude.ai, with software development making up 45.2% of work-related tasks. When a vendor builds local infrastructure for a market, it usually means the demand is already there.

How It Works Under the Hood

Let’s break it down. Amazon Bedrock serves the India endpoint through something called geographic cross-Region inference. In plain terms, your request can be handled by either of two AWS Regions, and both are in India: Mumbai (ap-south-1) and Hyderabad (ap-south-2).

Two details matter for your risk team. Your prompts and outputs may move between those two Regions. And according to AWS, customer data is not stored in the destination Region, it stays in the Region where you made the call. Bedrock also follows a zero data retention model by default, so it does not store model inputs or outputs.

The choice you make is the inference profile. Pick the wrong one and your requests can leave the country.

AWS had already offered Claude to Indian customers through global cross-Region inference. That route works, but it does not keep processing inside India. The new profile is what closes that gap.

Step-by-Step: Try Claude With India-Only Inference

You do not need to write code to test this. Start in the console, then move to the API once you like what you see.

Option 1: Test it in the Bedrock console

1. Open the Amazon Bedrock console in the Region you want as your source Region (Mumbai or Hyderabad).
2. Go to Playground. In the left navigation, look under Test.
3. Click Select model. It sits in the middle of the page.
4. Search for the Claude model you want. For Opus 5, choose the option labeled IN Anthropic Claude Opus 5 under Inference. That label is your confirmation that you picked the India profile. Click Apply.
5. Type a prompt and click Run. Start with something harmless, like asking it to summarize a public document.

Option 2: Call it from code

Before you start, make sure you have an AWS account with Bedrock access, the AWS CLI configured, Python 3.8 or later, and Boto3 installed. Your IAM role also needs permission to use a geographic inference profile, which AWS documents in its setup guide.

1. Install the SDK. Run pip install boto3 in your terminal.
2. Create a Bedrock Runtime client pointed at ap-south-1.
3. Use the India profile ID as your model ID. AWS’s example for Opus 5 is in.anthropic.claude-opus-5. Check the console for the exact ID of the model you want before you hard-code anything.
4. Send the request and print the reply.

Here is the pattern AWS shows for the Converse API:

import boto3
bedrock_runtime = boto3.client(
    service_name="bedrock-runtime",
    region_name="ap-south-1"
)
response = bedrock_runtime.converse(
    modelId="in.anthropic.claude-opus-5",
    messages=[
        {
            "role": "user",
            "content": [{"text": "Summarize the key risks of storing customer data offshore."}]
        }
    ],
    inferenceConfig={"maxTokens": 1024}
)
blocks = response["output"]["message"]["content"]
print("\n".join(b.get("text", "") for b in blocks if "text" in b))

5. Check your logs. CloudWatch and CloudTrail record entries in your source Region only, which keeps monitoring in one place. Confirm the calls show up where you expect.

A 5-Question Residency Checklist for Any AI Vendor

This one is not about India. Whatever country you operate in, the same questions will come up in your next security review. Ask them before you sign anything.

1. Where is the prompt processed? Not where the company is headquartered. Where the GPU actually runs.
2. Where is anything stored? Logs, caches, and fine-tuning data count.
3. Can requests fail over to another country? Resilience features are great until they route your data abroad.
4. Does a human ever review flagged content, and where? Get this in writing.
5. Which logs prove it? If you cannot show an auditor evidence, the answer does not count.

Screenshot this list. It works for every vendor, not just Claude.

Who Is Already Using Claude in India

According to Anthropic’s announcement, the adoption list is already long. Indian IT services firms including TCS, Infosys, and Cognizant are building Claude into client delivery, with TCS rolling it out to 50,000 associates. NPCI is building its agentic AI platform, AiNxt, on Claude. Groups like Reliance, Mahindra, and Godrej are deploying it across their businesses, and digital companies such as CRED, Swiggy, and Freshworks use it in products and engineering.

These are Anthropic’s own figures and customer statements, so treat them as vendor-reported. Still, the names tell you which buyers were waiting for local processing.

Honest Limitations

A few things worth knowing before you tell your boss this solves everything.

Residency is not compliance. Processing in India helps with one requirement. Consent, purpose limits, breach reporting, and vendor contracts under the data protection law are still your job. Talk to your legal team. This post is not legal advice.

It covers the Bedrock India endpoint. The announcement is about Claude on Amazon Bedrock. It says nothing about the Claude.ai app or Anthropic’s direct API, so do not assume those are processed in India.

Data can move between two Indian Regions. If your policy says a specific city, this setup may not satisfy it.

Human review can still apply. AWS notes that certain models require human review by AWS if content is flagged by automatic safety classifiers. Read the data retention page before you send sensitive material.

The performance claims are unverified. Customers cited lower latency and better resilience, but the sources include no published benchmarks. Test with your own workload. I also did not find pricing in the announcement, so check Bedrock’s pricing page for your Region.

The Bottom Line

This is infrastructure news, which means it is unglamorous and important. For regulated teams in India, the blocker was never model quality. It was the compliance conversation. A Claude endpoint that processes requests inside the country removes one of the hardest items from that conversation.

If you are in that position, spend 20 minutes in the playground this week. If you are elsewhere, notice the pattern: AI vendors are starting to compete on where your data runs, not just how smart the model is. Expect your own region to be next, and start asking the five questions now.

For more on getting practical value from Claude, browse the Claude section of this blog.

Know someone in banking, insurance, or government who keeps hearing “we can’t use AI because of data residency”? Send them this post before their next compliance meeting.

Sources

• Business Standard: Anthropic brings in-country Claude inference to India via Amazon Bedrock (October 5, 2026)
• AWS Machine Learning Blog: Amazon Bedrock expands Claude model availability to in-country inferencing in India (September 29, 2026)
• AWS Documentation: Geographic cross-Region inference
• AWS Machine Learning Blog: Access Claude models in India with global cross-Region inference
• NewsBytes: Anthropic starts processing Claude requests within India
• Hogan Lovells: India’s Digital Personal Data Protection Act 2023 brought into force
• Anthropic: Opens Bengaluru office and announces new partnerships across India (February 16, 2026)