Sarvam AI's $1.5 Billion Bet on India's Own AI Path

2026-06-28 — business growth strategy India

Sarvam AI closed a $234 million Series B last month. HCLTech led it, and suddenly we're talking about India building its own AI stack instead of just using someone else's.

I'll be honest. We talk about this stuff at Tanjore—sovereign AI, homegrown models, not routing everything through OpenAI's servers. Most consultants get this wrong, including us sometimes. You can say the words without understanding what's actually happening in Bengaluru.

What's actually happening is this: Sarvam isn't a typical enterprise software company that bolted on some LLM features and called itself AI. It's trying to build everything from the ground up. Foundation models. Inference. Deployed applications. The whole stack. For Indian languages. For Indian regulatory requirements. For banking and insurance and government services that need to work in Hindi or Tamil or Marathi, not just English.

The HCLTech piece matters more than the headline numbers.

HCLTech put $150 million in. That's 64 percent of the total raise. And they took just over 10 percent of the company, which means they're valuing Sarvam at around $1.5 billion. That's not investor FOMO. That's India's third-largest IT services firm saying they're betting the business on this. They bring a few things money can't buy: relationships with 500 banks, relationships with large manufacturers, and a workforce of several hundred thousand engineers. Actually, that's not quite right—it's not that money can't buy those things. It's that HCLTech already owns them, and that changes the equation entirely.

Sarvam started with a practical problem.

Vivek Raghavan and Pratyush Kumar built AI4Bharat at IIT Madras before launching Sarvam. They kept running into the same wall: American foundation models are built for English. India has 22 scheduled languages and hundreds of dialects. You can't build a banking app on a model that barely understands regional languages. You just can't.

Their Indus model is a 105-billion parameter foundation trained for Indian linguistic contexts. On top of that sits Bulbul V3, a voice model, and Bhashini-v2, a translation system they updated earlier this year. The translation part is doing something like 40 percent better on regional and tribal language variants than it was before. You can measure that improvement. It's not vague.

But here's where it gets interesting—the government connection isn't separate from this raise. It's the foundation for it.

Sarvam won a competitive tender to build India's first homegrown LLM under the IndiaAI Mission. That's a $1.25 billion national initiative. In February, the IT Minister announced 20,000 GPUs for the national AI compute pool. Odisha signed an MOU with Sarvam the same month to build a sovereign AI capacity hub. Tamil Nadu went further. They're co-building something called Digital Sangam—a 20-megawatt AI-optimized data center built specifically for this, anchored by Sarvam and IIT Madras.

That's not just funding.

That's government policy plus private capital plus physical infrastructure plus university research, all pointing at one startup. You can't fake that combination. Either the state is serious about this, or it's the most expensive theater production in Indian business. As far as I can tell, it's real.

The original plan was a $300 million raise. This $234 million is the first close. Bessemer Venture Partners came in alongside Khosla Ventures and Peak XV Partners. The goal is to combine Sarvam's models with HCLTech's existing software products and sell them to banks and governments at scale.

I think what matters now is what actually gets built and deployed.

The honest question—and I don't think anyone on the call was asking it—is whether this is real strategy or elaborate political theater. Whether Sarvam's models can keep pace with the frontier models as OpenAI and Anthropic keep scaling. Whether the government tender actually becomes government deployment, or stays theoretical.

But $234 million. A government mandate. Two state governments building physical infrastructure. HCLTech's enterprise distribution channel. That's not a press release. That's an ecosystem starting to form.

The bet is simple: countries that run their own foundation models will have more control over AI deployment inside their borders. More control over auditing. More control over training data. Whether Sarvam can execute at the speed and scale the frontier is moving is the real test ahead.