Sarvam, the Bengaluru company building India’s sovereign AI stack, has become the country’s newest AI unicorn. It has raised $234m in the first close of a $300m Series B, at a $1.5bn valuation, led by IT services giant HCLTech, which is putting in $150m.
The round is a sovereign-AI bet with a strategic spine.
HCLTech is the lead investor; Bessemer Venture Partners joined, with existing backers Khosla Ventures and Peak XV. Sarvam was founded by Vivek Raghavan and Pratyush Kumar, both veterans of AI4Bharat, the Indian-language AI effort at IIT Madras backed by Nandan Nilekani.
Earlier this year it released open-source models trained from scratch in India: a 105-billion-parameter system it says rivals larger models, and a 30-billion one tuned to run on consumer hardware.
Why HCLTech is betting on Sarvam
For HCLTech, this is not a passive cheque. The IT giant sells software and services to banks, insurers and governments, and a homegrown, India-controlled AI model is a powerful thing to put in front of customers nervous about routing sensitive data through American clouds.
The plan is to fuse Sarvam’s models with HCLTech’s enterprise relationships and engineering muscle. Sarvam’s stated focus, banking, insurance, government technology and defence, lines up almost exactly with where HCLTech already does business.
India’s sovereign-AI moment
The timing is not an accident. India is the second-largest market for both OpenAI and Anthropic, yet it has produced few serious frontier-model contenders, and one of the highest-profile, Krutrim, recently pivoted to cloud services. The debate about that dependence sharpened last week, when the US forced Anthropic to pull its most powerful models from non-US users.
Sarvam is the flagship of the “build our own” camp.
This is not a slideware startup. Sarvam says its voice agents have gathered data from 17 million farmers for India’s agriculture ministry, a campaign for one insurer touched 45 million policyholders, and its platforms now handle millions of interactions and API calls a day.
The open question is the hard one: whether it can train a genuinely frontier model, against OpenAI, Google and cheap Chinese open weights, on a fraction of their compute. India wants a national champion; Sarvam has just been handed the money, and the expectations, to be it.
