Sarvam, India's full-stack sovereign AI startup, has raised $234 million in the first close of a $300 million Series B round, at a post-money valuation of $1.5 billion.
HCL Technologies Ltd led the round, committing $150 million as the lead investor and picking up a 10.46% stake, according to regulatory filings. Bessemer Venture Partners joined the company's cap table alongside the IT major. Existing investors Khosla Ventures and Peak XV Partners also participated in the round.
“The funding allocation is across layers. The first one is building frontier models and conducting research and development to train better, more competitive models. That's where we'll be investing quite a bit,” said Pratyush Kumar, co-founder at Sarvam, in an interview with Mint. “Second, we're investing in compute. We're going to be investing a lot of money to deploy full frontier applications in India.”
The announcement comes just after the US government directed US foundational model company Anthropic to restrict use of its latest models, Fable 5 and cybersecurity-focused Mythos, citing export-control rules.
The Bengaluru-based company is yet to close out the remaining $66 million of its $300 million target. Kumar said a new set of investors would be joining the startup's cap table and that an announcement would happen “very soon”, but declined to provide a timeline.
Sovereign AI ambition
The capital will fund Sarvam's next frontier model—targeting agentic, coding, and cybersecurity use cases—and deepen its forward-deployed motion across banking, insurance, government technology, and defence.
Sarvam believes that, because all its data is sourced within the country, its sovereign nature positions the company well to leverage that advantage, even within the armed forces. “The armed forces need to deploy technology in sovereign settings in air-gapped conditions. As a full-stack company, we can work across every layer. Our unique selling point is our ability to build on air-gapped systems, which is a horizontal requirement across the range of things the armed forces would need them for,” Kumar said.
The startup positions itself as India's answer to sovereign AI—a full-stack company that trains its own models from scratch in India, rather than fine-tuning foreign models on Indian data.
India's sovereign AI ambitions are reaching an inflexion point. As global AI development concentrates among a handful of American and Chinese labs, a growing cluster of governments and enterprises is seeking AI infrastructure that keeps data, models, and inference loops within national borders.
The company believes that, even though the majority of its revenue at the moment comes from enterprise customers, there's significant value in having the government as a client, given the nature of its technology.
“Even if you do work for the government, which is a natural consequence of what we're building, we are building in India for India, and the distribution and the kind of size of the government it has is indeed a large opportunity. But, we are a technology provider. We neither have any specific, business advantage with respect to the government, nor is that what we are optimizing for.”
Simplifying AI for everyone
In February, at the India AI Impact Summit in New Delhi, the company announced a line of smartglasses called Kaze. The product is more business-to-business-focused than directly available to consumers. Sarvam has partnered with a company to manufacture the glasses, which will run on its technology. “This is something that enterprises buy from us along with our AI agents, which are deployed on the glasses, and come with a subscription,” Kumar said. “Right now they're being manufactured. The first set of companies buying them has been identified.”
The overarching logic behind the company's decision to launch the smartglasses was to target workers who can't be on their phones or at computers at all times. This includes frontline workers, such as delivery drivers, as well as those who have to collect documents. “The solutions in the market are very expensive for something like that to be done from a productivity perspective. You need to make AI work for everybody, and so if you can magnify the productivity of someone earning a ₹25,000 salary a month, right, with cost-effective AI, that's exactly what we are fitting for,” Kumar said.
Sarvam AI models
Sarvam's model portfolio has expanded considerably in recent months. Its models include the 105B model, which claims to match or outperform larger reasoning models on knowledge, reasoning, and agentic benchmarks. Its Sarvam 30B variant is optimized for edge deployment on consumer hardware.
It has also released Sarvam Vision, a model built for handwriting and Indian-language document recognition, which is being used to digitize over 35 million pages of insurance forms and land records.
This is the company's latest round of funding, coming two-and-a-half years after Sarvam first raised capital from investors. Back in 2023, the startup raised $41 million in a combined seed and Series A round from Lightspeed, Khosla Ventures and PeakXV Partners.