The timing tells you something about how seriously India is taking artificial intelligence.
On June 15, less than three years after its founding in August 2023, Bengaluru-based Sarvam AI closed $234 million in fresh capital—the first tranche of what it hopes will become a $300 million Series B—and landed a $1.5 billion post-money valuation at the first close. For a startup that raised $41 million just eighteen months earlier, the arithmetic is startling: a thirty-fold jump in perceived worth, fueled by a single conviction. That the future of enterprise AI in India, maybe across much of the Global South, belongs to homegrown infrastructure.
HCLTech, the Indian IT services giant, led the round with roughly $150 million for a 10.46 percent stake. Bessemer Venture Partners came in as a new backer. Khosla Ventures and Peak XV, both early believers, doubled down.
It's the kind of valuation that raises eyebrows even in an industry accustomed to frothy numbers. But Sarvam's founders—Vivek Raghavan and Pratyush Kumar, both alumni of AI Pure, a Bengaluru research lab they co-founded before pivoting to commercialization—argue the bet isn't on hype. It's on sovereignty.
When an IT Giant Becomes a Customer and Partner
HCLTech's involvement goes deeper than most venture checks. The company isn't just providing capital; it's embedding Sarvam into its enterprise sales motion, a distribution channel that touches thousands of corporate and government clients across India. The partnership formalized on June 25, when HCLTech completed its investment in Axonwise Private Limited, Sarvam's parent entity, according to investor disclosures reviewed by regulatory filings.
A month later, the contours of the relationship came into sharper focus. On July 24, HCLTech and Sarvam announced plans for a Rs 14,257 crore data center—call it $1.48 billion—in Bhubaneswar, the capital of Odisha. The facility is designed to serve as the computational backbone for the state's sovereign AI ambitions, part of a broader strategy that's seen Odisha and Tamil Nadu compete to become India's AI heartland. (Tamil Nadu inked its own partnership with Sarvam back in February.)
The arrangement suits both parties. HCLTech gets access to cutting-edge language models and agentic AI capabilities it can bundle into client engagements. Sarvam gets enterprise distribution and infrastructure at a scale few startups can muster on their own.
Though whether the arrangement will generate returns commensurate with the valuation—well, that's the wager.
Building the Stack, One Layer at a Time
Sarvam's thesis centers on what it calls "full-stack sovereign AI." The idea is to control everything: model training, inference infrastructure, deployment tooling. No dependence on foreign cloud providers or pre-trained models from OpenAI or Google. In practice, that means building from scratch in a market where most companies are content to resell someone else's technology.
In April 2025, the Indian government handed Sarvam a meaningful validation. Under the IndiaAI Mission, the startup was selected to develop the country's sovereign large language model. The designation came with access to government-subsidized compute resources—a non-trivial advantage when training runs can cost millions of dollars. By March, the IndiaAI compute portal had onboarded more than 38,000 GPUs, with officials projecting that figure could hit 100,000 by year's end. (Whether those projections hold is another question.)
Sarvam moved quickly to capitalize. That same month, it released two open-source models: Sarvam-30B and Sarvam-105B, both trained entirely within India using the IndiaAI infrastructure. The larger model—105 billion parameters, employing a Mixture-of-Experts architecture with Multi-head Latent Attention—supports a 128K context window and was pre-trained on 12 trillion tokens. The training corpus spanned multilingual text, code, math problems, and web data. Both models carry Apache 2.0 licenses, a nod to the open-source ethos Sarvam says distinguishes it from proprietary rivals.
February brought a flurry of product launches, perhaps more than the founders initially expected to ship in a single quarter. Sarvam Dub, a real-time multilingual dubbing tool. Sarvam Audio, built for long-duration automatic speech recognition. Sarvam Vision, a 3-billion-parameter vision-language model optimized for optical character recognition across 22 Indian languages. And Indus, a consumer-facing agentic assistant powered by the 105B model.
The company also published developer APIs, documented on its website with transparent pricing and rate limits—an unusual move in a market where many AI startups prefer to negotiate custom contracts rather than expose standardized economics.
Revenue Trails Valuation, but Deals Are Coming
Here's where the narrative gets complicated.
Revenue for the fiscal year ending March 2026 came in around Rs 45 crore, or roughly $5.4 million, according to regulatory filings cited by the Economic Times in June. That's modest, even by Indian startup standards, and a fraction of what you'd typically expect from a billion-dollar-plus company. The gap between valuation and revenue isn't unusual in early-stage AI—Anthropic, for instance, raised at a $15 billion valuation before meaningful commercialization—but it does mean Sarvam is being priced on potential, not performance.
Still, the customer roster suggests traction in high-value verticals. Named clients include SBI Life, LIC, IDFC First Bank, Tata Capital, and Cred—household names in Indian finance. One large fintech, whose identity Sarvam hasn't disclosed publicly, is deploying the company's agentic AI to support a 350,000-person sales force. That's the kind of deployment that, if it works, could unlock significant follow-on business.
Government projects add another layer. The Ministry of Agriculture has engaged Sarvam's tools for agricultural data initiatives. An insurer is piloting voice AI for customer outreach. And in February of last year, the Unique Identification Authority of India—UIDAI, the agency that manages Aadhaar, the world's largest biometric database—announced a partnership to bring multilingual voice capabilities to its services.
Beyond customers, Sarvam has signed collaboration agreements with Bosch Global Software Technologies (March) and YCP India (May), both aimed at accelerating enterprise AI adoption.
The Capital Treadmill
The $234 million will go toward three priorities, according to company statements: expanding compute infrastructure within India, training larger models focused on agentic AI, coding, and cybersecurity, and hiring what Raghavan describes as "superstar" talent. There's also a plan to open a San Francisco office—a signal, perhaps, that Sarvam's ambitions extend beyond the domestic market, even if India remains the core.
In a June 16 interview with the Economic Times, cofounder Pratyush Kumar offered a candid assessment of the funding landscape. "Our capital raise is large for India but quite small in the global context," he said. "We expect a constant cycle of building bigger models, raising more capital, and scaling up."
It's a treadmill familiar to anyone watching the AI industry. OpenAI has raised billions. Anthropic, too. Even well-capitalized startups find themselves returning to investors every twelve to eighteen months, chasing the compute budgets required to stay competitive.
Raghavan struck a slightly different note in a February interview with Mint, one that suggested a touch of skepticism about the parameter-count arms race. "We could have even built a trillion-parameter model," he said. "The question is why and how we should go about it."
The comment reads less as bravado than as restraint—a recognition, maybe, that sheer scale doesn't guarantee usefulness, especially in markets where inference costs and latency matter as much as raw capability. It's a position that's gained currency across the industry as the initial euphoria around ever-larger models has given way to questions about efficiency and return on investment.
Betting on Homegrown Infrastructure
For now, Sarvam is wagering that India's AI ecosystem will eventually demand local control over the full stack. The thesis has geopolitical tailwinds: governments worldwide are rethinking their dependence on a handful of American tech giants, and enterprises are increasingly wary of sending sensitive data to foreign clouds. Whether that translates into sustainable competitive advantage—and whether customers will pay a premium for sovereignty—remains an open question.
The challenges are real. Google, Microsoft, and Amazon already have deep footholds in India's enterprise market. Their models are mature, their infrastructure proven, their sales teams entrenched. Displacing them will require more than patriotic sentiment; it will require demonstrable technical superiority, or at least parity, plus pricing that doesn't punish customers for choosing the local option.
Sarvam's leadership seems aware of the gap. Kumar's comment about the "constant cycle" of fundraising isn't just acknowledging reality—it's a warning. Competing in AI at this level demands capital on a scale that few Indian startups have ever accessed. The $234 million is a start. It won't be the last check.
Whether this bet pays off at unicorn valuations will hinge on execution in a market that's both skeptical and eager, curious about homegrown alternatives but unwilling to compromise on performance. Sarvam has the backing, the infrastructure partnerships, and a government mandate. What it still needs is proof that sovereign AI can work at scale—and that enterprises will choose it over the incumbents they already trust.
