Anthropic has picked Bengaluru for its first India office, and the choice says a lot about where the AI company thinks the next wave of growth will come from. India is already the second-largest market for Claude.ai, and a huge chunk of that usage isn’t just people asking fun questions—it’s developers quietly using Claude to write code, modernise old systems, and ship production software. Nearly half of Claude’s usage in India is tied to computer and mathematical tasks, which is basically Anthropic’s way of saying: this is where serious AI-building is happening.

The new Bengaluru office is Anthropic’s second in Asia after Tokyo, and it will be led by Irina Ghose, the company’s Managing Director for India and a longtime enterprise and startup tech leader. The plan is straightforward: hire local talent across roles and plug directly into India’s startup ecosystem, IT service giants, and digital-first companies that are already betting heavily on AI. In parallel, Anthropic is rolling out a network of partnerships across enterprise, education, agriculture, and the public sector, effectively turning India into one of its most important testbeds for “responsible AI at scale.”

One big theme running through Anthropic’s India push is language. India has more than a billion people speaking over a dozen officially recognised languages, yet most powerful AI models still behave as if the world is mostly English. About six months ago, Anthropic kicked off a company-wide effort to fix at least part of that: curating higher-quality, more representative training data for 10 major Indian languages—Hindi, Bengali, Marathi, Telugu, Tamil, Punjabi, Gujarati, Kannada, Malayalam, and Urdu. The company says this has already improved Claude’s fluency in these languages, and it’s still tuning those capabilities.

But beyond training data, Anthropic is now working with Karya and the Collective Intelligence Project to build proper evaluation benchmarks for India-specific tasks—things like helping a farmer understand an agronomy recommendation, or assisting with legal information in regional languages. These evaluations are being shaped with domain experts from nonprofits such as Digital Green and Adalat AI, and Anthropic intends to make them publicly available so other players can also test models on Indian use cases instead of generic English benchmarks. It’s a subtle but important shift: India isn’t just a big user base, it’s shaping the standard for how AI should behave in multilingual, high-stakes environments.

On the commercial side, Anthropic’s India story is already well underway. The company says its run-rate revenue in India has doubled since it first announced an expansion in October 2025, and the customer list spans incumbents and upstarts. Air India is using Claude Code to help its developers build custom software faster and at lower cost, as part of a broader ambition to become what its leadership calls an “agentic airline”. CRED reports that using Claude Code has helped it ship features twice as fast while improving test coverage by 10%, which, in startup terms, is a meaningful edge. Cognizant, meanwhile, is rolling out Claude to 350,000 employees globally to modernise legacy systems and accelerate software development for its enterprise clients.

Among startups, the use of Anthropic’s stack is even more embedded. Razorpay is weaving AI into its risk systems, decision-making, and day-to-day operations. Enterpret leans on Claude both for its customer-facing assistant and internally through Claude Code; the company has even built an MCP integration that pipes customer insights directly into Claude for richer context. Then there’s Emergent, a platform that lets anyone describe the software they want in plain language and have it built with the help of AI—Anthropic says Emergent hit 2 million users and 25 million dollars in annual recurring revenue in under five months, powered entirely by Claude. For Anthropic, these stories are proof that India’s developer and startup ecosystem isn’t just experimenting with AI—it’s building real products and real businesses on top of it.

Education is another pillar where Anthropic is trying to show that AI can do more than draft emails. Around 12% of Claude usage in India is already tied to educational and instructional tasks, and Anthropic is now formalising that with partnerships. One of the most notable is with Pratham, one of India’s largest education nonprofits, which has picked Anthropic as its first strategic AI lab partner. Pratham’s “Anytime Testing Machine,” powered by Claude, is being piloted with 1,500 students across 20 schools and is expected to expand to 100 schools by the end of 2026. The same tool has already been adapted for over 5,000 learners in Pratham’s Second Chance programme, which helps women who dropped out of school prepare for exams and certifications.

Anthropic is also teaming up with Central Square Foundation to support organisations building AI-enabled tools for children in underserved communities—think personalised tutors, teacher coaching tools, and assessment-linked instruction. The company says it will offer technical expertise, mentorship, and API credits to these projects, with the goal of reaching more primary school students across the country. The subtext here is clear: if AI is going to be used in Indian classrooms, Anthropic wants Claude to be the engine powering that shift, especially in environments where resources are tight and teacher-student ratios are high.

The public sector angle is equally interesting. India has a global reputation for building large-scale digital public infrastructure—Aadhaar, UPI, and DigiLocker being textbook examples. Anthropic is partnering with the EkStep Foundation to explore how Claude can plug into similar population-scale systems, starting with agriculture. Agriculture accounts for nearly one-sixth of India’s GDP and employs almost half of its workforce, yet expertise often doesn’t reach small farmers in time. Using the OpenAgriNet effort, Anthropic and partners are working towards deployments of Claude that can expand access to expert agricultural knowledge across regions and languages.

The company is also experimenting with AI inside nonprofits that work in critical areas like health. Noora Health, which provides health coaching to millions of families, and Intelehealth, which connects patients in remote communities to doctors, are using Claude Code and Cowork to streamline internal workflows and software development. It’s less flashy than a big consumer launch, but it’s the kind of plumbing work that decides whether AI tools actually show up in everyday services or stay stuck in pilots.

Justice is another sector where Anthropic is leaning in. India currently has roughly 50 million pending court cases, and case updates can take months to reach litigants, often requiring repeated court visits or an intermediary to navigate dense paperwork and legal jargon. Anthropic is supporting Adalat AI, a legal-tech nonprofit, to launch a national WhatsApp helpline, powered by Claude, that goes live with this announcement. Through a chatbot on WhatsApp—a platform already used daily by hundreds of millions of Indians—people can look up their case, check status, see the next hearing date, and download orders. Claude’s job in the background is to summarise long, complex court orders and translate them into everyday language in the user’s preferred tongue, currently English, Hindi, Kannada, and Malayalam, with plans to expand to more languages. For many litigants, that means understanding their own case for the first time without depending entirely on a lawyer or middleman.

Underpinning a lot of these deployments is something quite technical but strategically important: the Model Context Protocol, or MCP. Anthropic developed MCP as an open-source standard that lets AI systems connect cleanly to external tools, databases, and services, and it recently donated MCP to the Linux Foundation, framing it as a building block for the broader “agentic AI” ecosystem. In India, that’s not just theory. The Ministry of Statistics and Programme Implementation (MoSPI), with support from nonprofit Bharat Digital, has launched what it calls the first official Indian government MCP server. This server allows AI systems to query authoritative national statistics in an open, interoperable way, which means developers and companies can build agents that tap into reliable government data instead of scraping random PDFs.

On the consumer side, Swiggy is already using MCP so people can order groceries or book dining reservations directly through Claude. It’s a neat example of what Anthropic wants MCP to enable: instead of AI living inside one app, it becomes a thin, smart layer that can talk to many services behind the scenes. For India’s digital-native population, this is the kind of integration that can quietly normalise AI agents in daily life.

Put together, the Bengaluru office is less a simple real-estate move and more a signal of how Anthropic sees India: as a massive, multilingual lab for building and deploying AI that has to work in messy, real-world conditions. From enterprise codebases and fintech risk engines to government stats, school exams, farm advice, and court orders on WhatsApp, the company is trying to show that Claude can embed itself into systems that matter at scale. For Indian developers and organisations, it also means that one of the world’s biggest AI players is now physically in the country, hiring, partnering, and, crucially, tuning its models around local realities rather than treating India as just another line on a global usage chart.