Anthropic launches in-country Claude inference in India

Anthropic will begin processing Claude requests on servers physically located in India through Amazon Bedrock, AWS's managed GenAI service. The capability, arriving in the coming weeks, removes the data-residency barrier that has kept Indian banks, financial institutions and government agencies from deploying international frontier AI models in production.

India is Anthropic's second-largest market after the US, accounting for an estimated 5.8% to 7.2% of global Claude usage. Until now, enterprises in regulated sectors could evaluate Claude but not ship it, because prompts and outputs crossed borders. With in-country inference, that changes.

What does in-country inference actually solve?

RBI circulars and India's Digital Personal Data Protection Act require certain sensitive data to stay within national borders. When a bank sends customer records to a model hosted abroad, it may breach those rules. In-country inference keeps prompts and responses on Indian servers, closing that gap.

Latency drops too. Round-tripping a request to US or EU endpoints adds tens of milliseconds. For real-time fraud detection or trading-desk assistants, that matters.

"When that data can stay in India, AI moves from pilots into the systems that matter most," said Irina Ghose, Anthropic's India Managing Director.

Early adopters in Indian banking and IT

Anthropic named several enterprise customers already running Claude. Axis Bank and IndusInd Bank are using the model for engineering and productivity workflows. NPCI is building its agentic AI platform, AiNxt, on Claude. TCS has rolled out access to 50,000 associates, while Infosys has integrated Claude into its Topaz platform and set up an Anthropic Centre of Excellence.

Routing through Amazon Bedrock adds enterprise controls that compliance teams typically require: audit trails, identity-based access and governance dashboards. For regulated institutions, that governance layer is as important as the model itself.

The competitive picture tightens

Anthropic is not alone. Last month, Google launched in-country inference for Gemini 3.5 Flash for Indian enterprise customers. Homegrown Sarvam AI has built its entire sovereign-AI pitch around locally hosted models.

For Indian foundation-model startups, this is uncomfortable. Sovereign data residency was one of their clearest differentiators. Now that Anthropic and Google offer the same capability, domestic labs will have to compete on model performance, Indic-language quality, pricing and vertical integrations.

Anthropic also introduced rupee-denominated pricing for Claude subscriptions in India in early July, removing the friction of USD payments for individual developers and smaller teams.

What it does not answer

Anthropic did not disclose which AWS region will host the inference endpoints or the exact go-live date beyond "coming weeks." Token pricing for Bedrock-billed Claude 3.5 Sonnet and Opus in India also remains unpublished. Teams planning migrations should confirm billing rates before committing workloads.

Whether the regulator's data-localisation requirements are fully satisfied will depend on how each institution interprets the rules. Storing model outputs and audit logs may still demand additional controls beyond what Bedrock provides by default.

The sovereign-AI race in India is no longer about whether global models will adapt. It is about which vendor locks in enterprise contracts first.

Source: Inc42 Media / Ankush Das