Sarvam Hits $1.5B as India's Sovereign-AI Unicorn

Sarvam, a Bengaluru-based AI startup, raised $234 million at a $1.5 billion valuation, making it India's newest AI unicorn, TechCrunch reported June 15. The round was led by IT-services giant HCLTech, which put in $150 million, alongside Bessemer Venture Partners, Khosla Ventures and Peak XV Partners. It's part of a Series B targeting $300 million in total, and it dwarfs the roughly $41 million Sarvam had raised across its seed and Series A rounds.

Sarvam was founded by Vivek Raghavan and Pratyush Kumar, who both previously worked on AI4Bharat at IIT Madras. The company builds AI models tailored to Indian languages and use cases, and it runs the full stack — model development, the inference infrastructure to serve them, and the enterprise apps on top. That last part matters: Sarvam isn't a research demo. It says it already handles 2 million conversational AI interactions and 10 million API calls a day, transcribes 500,000 hours of audio a month, and has digitized 35 million pages with its document AI.

The deployments are unusually concrete for an AI startup. Sarvam says its systems serve 17 million farmers for India's Ministry of Agriculture, support 45 million insurance policyholders through voice campaigns, and power a 350,000-person sales force at a large fintech. In other words, the money is following models that already do real work in Indian languages, not a promise to build them.

Why 'sovereign AI' is suddenly hot

The timing isn't an accident. This raise lands right after Anthropic suspended access to its most advanced models for foreign nationals — a vivid reminder that a tool your business runs on can be switched off by someone else's government overnight. India has been pushing hard on "sovereign AI": models built, hosted and controlled inside the country, on local languages and data, so the lights don't go out because of a decision made in another capital.

For most people this won't change which chatbot you open tomorrow. But the bigger shift is real and worth understanding: the AI map is fracturing along national lines. Instead of a few global models everyone shares, countries are funding their own — for language coverage, yes, but mostly for control. A model trained on your language and rules, that can't be cut off from abroad, is becoming a strategic asset like a power grid.

My take: the language angle is the part I'd actually celebrate. Most AI is still quietly English-first, and tools that work well in Hindi, Tamil or Telugu open the technology to hundreds of millions of people who've been second-class users until now. The geopolitics is the headline; better-served languages is the part that touches real lives.