Sarvam raised $234 million at a $1.5 billion valuation, making the Bengaluru-based startup India's newest AI unicorn. HCLTech, the IT subsidiary of HCL Group, led the round with a $150 million strategic check, joined by Bessemer Venture Partners alongside existing backers Khosla Ventures and Peak XV Partners. The company is aiming for $300 million in total Series B commitments.
The valuation marks a sharp step-up from Sarvam's prior capitalization. The startup had raised just $41 million across its seed and Series A rounds over the previous two years, meaning this single round delivers roughly 5.7x the lifetime capital the company had taken in before.
Sarvam released open-source models at 30 billion and 105 billion parameters earlier this year, tuned for Indian languages and local enterprise use cases. The company is one of a small group of Indian startups trying to operate full-stack — model training, inference infrastructure, and enterprise applications — rather than wrapping a foreign provider's API.
Key facts
- 01Sarvam raised $234M at a $1.5B valuation, with $150M coming from lead investor HCLTech.
- 02The Series B targets $300M total and follows just $41M raised across the startup's seed and Series A rounds.
- 03Sarvam released open-source models at 30B and 105B parameters earlier this year, aimed at Indian languages.
- 04Its conversational platform handles 2M interactions per day; its inference platform processes 10M API calls daily.
- 05Bessemer Venture Partners joined existing backers Khosla Ventures and Peak XV Partners in the round.
Deployment numbers are already non-trivial. Sarvam said its conversational AI platform handles more than 2 million interactions a day, while its inference platform processes roughly 10 million API calls daily. Its speech models transcribe more than 500,000 hours of audio each month, and its document AI digitizes more than 35 million pages of records.
The customer footprint skews toward scale-government and financial-services work. Sarvam's multilingual voice agents collected data from 17 million farmers for India's Ministry of Agriculture and Farmers Welfare, and a voice campaign for a leading insurer supported policy renewals across 45 million policyholders. A large fintech is using Sarvam's agentic platform to support a sales force of more than 350,000 people.
Co-founder Vivek Raghavan framed the company's strategy as horizontal diffusion rather than a single flagship model.
The HCLTech tie-up is the operational core of the deal. Sarvam plans to combine its models with HCLTech's enterprise relationships, engineering workforce, and software assets to package AI products for businesses and governments, giving the startup a distribution channel it would have taken years to build alone. Raghavan and co-founder Pratyush Kumar previously worked at AI4Bharat, the Indian-language AI initiative at IIT Madras backed by Nandan Nilekani.
The round lands during a sharp shift in how governments are thinking about model access. Anthropic disabled access to its latest Fable 5 and Mythos 5 models last week after the U.S. government ordered the company to suspend use by foreign nationals on national-security grounds, a move covered here at the time. That decision underscored how concentrated frontier model access remains — and why countries are funding domestic alternatives.
India is simultaneously one of the largest AI consumer markets and one of the smallest contributors to frontier model development. Both OpenAI and Anthropic describe India as their second-largest market after the U.S., yet high compute costs and limited domestic capital have left few Indian labs competing with U.S. and Chinese players. Sarvam is positioning itself in the gap.
The risks are real. A $1.5 billion valuation puts Sarvam in the same conversational range as much better-resourced labs whose pretraining budgets dwarf its total capital raised. Building sovereign infrastructure that can hold its own against models trained on hundreds of millions of GPU-hours is an open question, and the company has said it will use the fresh funds to expand compute access and develop next-generation agentic, coding, and cybersecurity models — categories where U.S. labs are iterating monthly.
Sarvam's bet is that distribution beats parameter count in regulated, language-specific verticals. If sovereign AI continues to harden as a procurement category — and the Anthropic export episode suggests it will — domestic full-stack providers with government-scale deployments and a deep-pocketed enterprise channel like HCLTech may end up owning the segments that matter most to local buyers, even without matching frontier benchmark scores. That is a narrower thesis than competing globally, and a much more defensible one.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
