🧭 Anthropic Opens Bengaluru Office — India Is Its #2 Market Globally

Anthropic officially opened its Bengaluru office on 16 February 2026 — its second in Asia after Tokyo and its fourth international location overall. Heading the operation is Irina Ghose, a former Microsoft India managing director, who joins as Anthropic's India MD. The announcement confirmed what internal usage data had already shown: India is now Claude.ai's second-largest market worldwide, with nearly half of all Indian usage concentrated in computer and mathematical tasks — a profile that skews sharply towards developers and technical professionals.

Launch partners and sectors

• Air India — AI-powered customer service and operational workflows
• CRED — financial product personalisation and fraud detection
• Razorpay — payment intelligence and developer tooling
• Pratham — education nonprofit deploying Claude for learning support across underserved communities
• Ministry of Statistics — an MCP server connecting Claude to national statistical data for policy and research use

Language and sector priorities

Anthropic has committed to improving Claude's fluency across 10 major Indian languages — including Hindi, Tamil, Telugu, Bengali, and Kannada. The company has flagged three sectors of particular national impact: agriculture (advisory and market pricing tools for smallholder farmers), healthcare (diagnostic support and clinical documentation), and judicial access (India has approximately 50 million pending court cases; AI-assisted legal drafting and case summarisation is an explicit near-term goal).

🧭 Anthropic's $50B US Infrastructure Buildout — First Data Centres Now Online

Announced in November 2025, Anthropic's $50 billion American computing infrastructure programme — developed in partnership with data-centre builder Fluidstack — is no longer just a pledge on paper. The first custom facilities in Texas and New York have come online during Q1 2026, marking Anthropic's first meaningful move away from pure reliance on AWS and Google Cloud for compute capacity. The scale of the build reflects a clear strategic intent: to own the infrastructure layer as AI workloads become more complex, longer-running, and latency-sensitive.

What makes these facilities different

• Custom-designed for Anthropic workloads — rack density, cooling architecture, and networking are optimised specifically for the token-processing and long-context inference patterns Claude uses, rather than being general-purpose cloud hardware
• Direct power agreements — the Texas facilities are built adjacent to renewable generation sites, reducing both cost and carbon intensity relative to shared-tenancy cloud regions
• Edge inference readiness — the architecture supports lower-latency serving for enterprise customers who cannot tolerate the variable round-trip times of a public cloud hop

By the numbers

• ~800 permanent jobs and ~2,400 construction roles created
• Anthropic now serves over 300,000 business customers, with large accounts (>$100K annual revenue) growing roughly 7× year-over-year — the infrastructure investment is a direct response to that trajectory
• The $50B commitment is spread over several years; Fluidstack is the primary build-and-operate partner, with Anthropic retaining ownership of the hardware layer

🧭 Anthropic Interviewer — How to Run AI-Powered Qualitative Research Studies

Yesterday's entry covered what Anthropic's 81,000-person study found. But the tool that made it possible — Anthropic Interviewer — is itself worth understanding as a reusable research methodology. Interviewer is an AI-powered conversational research agent built on Claude: it conducts open-ended interviews at scale, follows up on interesting threads, probes for nuance, and synthesises the responses into structured thematic data. The 81K study ran it across 159 countries in 70 languages without any human interviewers — a previously impossible scope for qualitative work.

How Interviewer works

• Conversational depth, survey scale — unlike a fixed-question survey, Interviewer adapts its follow-up questions based on each participant's previous answer. A respondent who mentions "job anxiety" will receive a different next question than one who mentions "time savings".
• Multilingual by default — participants respond in their own language; Claude handles translation and thematic tagging in the same pipeline without losing nuance to literal translation artefacts
• Structured output — responses are tagged into a hierarchical theme taxonomy, enabling quantitative analysis of qualitative data (e.g. "26.7% mentioned reliability concerns")
• Bias mitigation — the system prompt is carefully designed to avoid leading questions; Anthropic published its full interview protocol alongside the study findings

Building your own research study with Claude

You don't need access to Anthropic Interviewer directly to apply the same pattern in your own product. The core methodology translates into a Claude API workflow:

system: """You are a neutral research interviewer. Your goal is to understand
the participant's experience with [TOPIC] in depth.

Rules:
- Ask one question at a time
- Follow up on specific points the participant raises
- Never lead or suggest answers
- After 5-7 exchanges, summarise the key themes you've heard
- Output a JSON summary at the end with detected themes and sentiment"""

user: [opening question]