Anthropic Launches Claude in India via AWS

Anthropic has started processing requests for Claude within India as of Monday (5), through Amazon Bedrock. Claude Opus 5, Sonnet 5, and Haiku 4.5 are available on an Indian endpoint that routes traffic between AWS regions in Mumbai and Hyderabad, ensuring that inference does not leave the country, according to the company's announcement. This promise was made in August; delivery came two months later.

The target audience is explicit: banks, governmental agencies, and large enterprises subject to data handling regulations. Reliance and fintech CRED have tested the feature in a private preview. In the announcement, TCS, Kotak Mahindra Bank, Mahindra Group, CRED, NPCI, the Data Security Council of India, and Godrej Industries commented on the launch, all citing data residency, governance, or regulatory requirements.

Why India Comes First

India is the second-largest market for Claude.ai, according to Anthropic itself, which opened an office in Bengaluru in February under the leadership of Irina Ghose, former managing director of Microsoft in India. The clients announced at that time already indicated the profile of demand: Cognizant is bringing Claude to 350,000 employees globally, and CRED reported functionality delivery that is twice as fast and a 10% greater testing coverage.

The bottleneck was different. Generative AI pilots in banks, NBFCs, healthcare providers, and the Indian public sector faced the requirement to keep sensitive data within national territory. The most recognized rule is that of the Reserve Bank of India, which since 2018 has mandated payment system operators to store this data only in the country, and the risks and compliance areas of banks extend the same reasoning to any customer data sent to an external model. With the local endpoint, the discussion shifts from whether data can leave the country to which use case goes into production first. The fact that NPCI, which operates UPI, India's instant payment system, commented on the launch indicates where the appetite lies.

The Standard Behind the Announcement

India is not the first country to receive this arrangement. In October 2025, AWS launched geographic inference of Claude Sonnet 4.5 and Haiku 4.5 limited to Japan, between Tokyo and Osaka, and to Australia, between Sydney and Melbourne. The logic is the same: instead of a single region, a pair of regions within the national border, which preserves capability and meets location requirements. The difference lies in the catalog. The Japanese and Australian launch only covered Sonnet 4.5 and Haiku 4.5; India receives the Opus family from day one.

For Japanese and Australian banks, which have been operating under this arrangement since 2025, the relevant question becomes catalog parity: Anthropic and AWS did not detail whether the generation 5 models are already running under the same geographic restriction in these countries. For buyers, a local endpoint only resolves the issue if the model required by the project is present.

What Changes for Integrators and Consultancies

TCS, Infosys, and Cognizant, partners of Anthropic, have until now been the main routes for Claude within regulated Indian clients. With local inference, these firms can sell production projects—not just proofs of concept—to banks and insurers in the country. The presence of Kotak Mahindra Bank among those commenting on the launch shows that at least one large private bank is already treating the topic as a production agenda.

The Counterpoint

Data residency is not sovereignty. Inference operates on AWS infrastructure, an American company, and the model remains Anthropic's. For more stringent regulators, the physical location of processing solves part of the equation, but not all: questions regarding access to logs, retention, and jurisdiction over the service provider remain open, and this is where local vendors, such as those operating models trained in the country, will focus their business argument. Anthropic and AWS did not disclose differentiated pricing for the Indian endpoint or capacity limits per client.

The competition for the Indian corporate market will be decided less by model performance and more by where it runs and who signs the contract. Anthropic took two months to turn a promise into an endpoint; the question for competitors is how long they take to offer the same level of model within the same borders.