Which AI Models Can Run in Your Country on Amazon Bedrock? Data Residency Explained, Region by Region (2026)

The bigger question for readers outside India is simple: which models stay in my country, or at least my region? Here is the answer, region by region, using only AWS documentation.

What "in-country inference" means, in one paragraph

Inference is the moment a model reads your prompt and writes its answer. With in-country inference, that processing happens only in data centres inside one country. On Bedrock it works through a geographic cross-Region inference profile. You send the request to one AWS Region (the "source"), and Bedrock may run it in another Region on a fixed list (the "destinations"), all inside the same geography. For India, that list is exactly two Regions: Mumbai (ap-south-1) and Hyderabad (ap-south-2). Your prompt never leaves India for processing. (New to AWS Regions? Start with our plain-English AWS explainer.)

Why data residency matters

Many laws and regulators care about where data is handled, not only how.

• India. The Digital Personal Data Protection Act, 2023 lets the central government restrict transfers of personal data to certain countries. Sector rules go further. The Reserve Bank of India's 2018 directive on storage of payment system data requires payment data to be stored only in India. Banks, insurers and fintechs often treat "keep it in India" as the safe default.
• European Union. The GDPR (Chapter V) limits transfers of personal data outside the EU/EEA unless certain safeguards apply. For many EU teams, an EU-only processing path removes a whole layer of legal review.
• Healthcare and the public sector. Hospitals, insurers and government bodies in many countries have contracts or rules saying patient and citizen data must stay onshore.

One honest caveat. In-country processing supports compliance; it does not guarantee it. Your contracts, logging, retention settings and lawful basis for processing still matter. Treat this as one important control, not the whole answer.

The region-by-region table (as of 22 October 2026)

This table comes from AWS's model cards for Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5, plus the AWS posts on GPT-5.6 in India. It covers the bedrock-runtime endpoint. Where we could not find a primary AWS source, the cell says not confirmed.

Three things jump out.

First, Canada is not in-country. A request sent from ca-central-1 using the US profile can run in Canada or in the United States. If your rule says "Canada only", this profile does not meet it.

Second, the UK sits with the EU. London callers can be routed to EU Regions. That may be fine under UK adequacy rules for many teams, but it is not UK-only. Sonnet 5's In-Region option in London is the UK-only path in this table.

Third, "global" means global. In Mumbai, Bahrain or São Paulo, the global profile can send your prompt to any commercial AWS Region. It is good for capacity, not for residency.

AWS also warns not to trust the prefix alone. Its inference profile docs say destination lists can differ by source Region, and anyone with residency rules should check the real list from each source Region they use.

India in detail: Mumbai, Hyderabad, three Claude models and GPT-5.6

According to AWS's Claude-in-India announcement, Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 can now be called through India geographic cross-Region inference. Requests route only between Mumbai and Hyderabad. This sits alongside the global profile that Indian customers could already use.

The model cards confirm the profile IDs:

• in.anthropic.claude-opus-5
• in.anthropic.claude-sonnet-5
• in.anthropic.claude-haiku-4-5-20251001-v1:0

You can call them from either Mumbai or Hyderabad, and both route only to those two Regions. One limit to note: AWS's Haiku 4.5 model card says structured outputs are not supported through the India Haiku profile.

On OpenAI, the claim from AWS India's president, on LinkedIn, that GPT-5.6 Terra and Luna have run locally since mid-August matches AWS's own records. The What's New post (18 August 2026) lists two India profiles, in.openai.gpt-5.6-terra and in.openai.gpt-5.6-luna, which "route requests only within the India geography". AWS's follow-up blog calls Terra the balanced, general-purpose model and Luna the fastest and cheapest. Both have a 1-million-token context window.

So the headline holds: an Indian team can now pick between two leading model families on Bedrock without data leaving India for processing. Want a cheaper, faster option? See our guide to Claude Haiku on Bedrock. Prefer Amazon's own models? Read our Amazon Nova explainer.

For developers: picking the right profile

Geographic vs global profiles

Bedrock gives you three choices, as described on each model card:

• In-Region: the request runs only in the Region you call. Strictest, but available for few model/Region pairs.
• Geo (geographic) cross-Region: the request may run in any Region on a fixed list in one geography. AWS says the destination list of a geography-tied profile "will never change". New Regions arrive as new profiles.
• Global cross-Region: the request may run in any commercial AWS Region. Most capacity, no residency.

The profile ID tells you which you are using. The prefix (in., eu., us., au., jp., or global.) comes before the model name.

Choosing the India profile

Create your Bedrock client in ap-south-1 or ap-south-2, then pass an in. profile ID as the model ID. This Converse API sketch follows the pattern in AWS's own model-card samples, with the Region and ID swapped for India:

import boto3

client = boto3.client("bedrock-runtime", region_name="ap-south-1")
response = client.converse(
    modelId="in.anthropic.claude-sonnet-5",
    messages=[{"role": "user", "content": [{"text": "Summarise the DPDP Act in three lines."}]}],
)
print(response["output"]["message"]["content"][0]["text"])

To see exactly where a profile can route from your source Region, AWS documents two calls: ListInferenceProfiles (to see what is available) and GetInferenceProfile (the models field lists one model ARN per destination Region).

Watch your SCPs

If an organisation-wide Service Control Policy blocks any destination Region in a profile, AWS says the request fails, even if other destinations are allowed. For India, allow Bedrock inference in both Mumbai and Hyderabad.

Logging, billing and quotas

According to AWS's India posts:

• CloudTrail and CloudWatch log entries only in the source Region. For GPT-5.6, AWS says CloudTrail's additionalEventData includes an inferenceRegion key showing which of the two Indian Regions served the call.
• Billing and quotas are tracked against your account in the source Region.
• Storage: AWS says customer data "is not stored in a destination Region" when using cross-Region inference, and Bedrock does not store inputs or outputs by default. One exception: content flagged by abuse-detection classifiers can be kept for review for certain models, including GPT-5.6.

Pricing

Neither India post lists rates. Both point to the Amazon Bedrock pricing page. Claude models are billed through AWS Marketplace, so they show under the model provider in Cost Explorer. Check the pricing page for your Region and profile type before you assume geo and global cost the same.

Editor's take

This is a big deal for India and a useful moment for everyone else to re-check their setup.

For Indian banks, insurers and hospitals, "we can't use a frontier model because the data would leave India" has been a real blocker. With Claude and GPT-5.6 both on in. profiles, that blocker is mostly gone for inference. The remaining work is paperwork, logging and access control. Those are solvable.

For everyone else, the table is the warning. Canada still leans on US capacity. The UK still shares with the EU. Most of the Middle East, Africa and Latin America only get global routing for these models. If your compliance team assumed "we're in our local Region, so we're in-country", the model card may say otherwise. Check it, and run GetInferenceProfile from every Region you use.

FAQ

Is Claude available in India on Bedrock?

Yes. AWS says Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 are available through India geographic cross-Region inference, routing only between Mumbai and Hyderabad.

What is Amazon Bedrock data residency?

It means controlling where your prompts and outputs are processed. On Bedrock you do this by choosing an In-Region model, a geographic profile (such as in. or eu.), or a global profile, which has no residency guarantee.

How does Bedrock cross-Region inference work in India?

You call Bedrock in Mumbai or Hyderabad with an in. profile ID. Bedrock runs the request in whichever of those two Regions has capacity. Logs and billing stay in the Region you called.

Does Bedrock store my data?

AWS says Bedrock does not store model inputs or outputs by default, and does not store customer data in destination Regions during cross-Region inference. Content flagged by abuse-detection systems may be kept for review for some models.

Is GPT-5.6 available in India on Bedrock?

Yes. AWS announced India profiles for GPT-5.6 Terra and GPT-5.6 Luna in August 2026, routing only within Mumbai and Hyderabad.

Can I keep AI processing inside Canada or the UK?

Not fully with the geo profiles for these models. AWS's model cards show Canadian calls on the US profile can run in the US, and London calls on the EU profile can run in EU Regions. Sonnet 5 lists an In-Region option in London.

Which Bedrock profile is best for GDPR?

For EU teams, the eu. profiles keep processing within EU Regions, plus Zurich or London when you call from there. Whether that satisfies your obligations is a legal question for your data protection officer.

Is in-country inference more expensive?

AWS's India posts don't give rates. Check the Bedrock pricing page for your model, Region and profile type.

Sources

• AWS: Amazon Bedrock expands Claude model availability to India with cross-Region inference
• AWS What's New: Amazon Bedrock now supports OpenAI models in India
• AWS: Introducing India cross-Region inference for OpenAI GPT-5.6 models on Amazon Bedrock
• AWS docs: Supported Regions and models for inference profiles
• AWS docs: Claude Opus 5 model card
• AWS docs: Claude Sonnet 5 model card
• AWS docs: Claude Haiku 4.5 model card
• AWS: Amazon Bedrock pricing
• Government of India, MeitY: Data protection framework (DPDP Act, 2023)
• Reserve Bank of India: Storage of Payment System Data (2018)
• EUR-Lex: Regulation (EU) 2016/679 (GDPR)
• Sandeep Dutta, President of AWS India and South Asia, on LinkedIn (post around 22 October 2026)

Featured image: Sandeep Dutta, President of AWS India and South Asia, via LinkedIn. Source