Sarvam AI's $234 million first close gives India a new AI unicorn, but the more important part is who led it: HCLTech, a services giant with deep access to the regulated customers Sarvam needs.
The round's first close was announced on June 15, with the Economic Times reporting that Sarvam AI raised $234 million at a valuation of about $1.5 billion. HCLTech, India's third-largest IT services firm, led the financing with a $150 million anchor investment for a 10.46% stake, roughly Rs 1,427 crore. Existing backers Khosla Ventures and Peak XV Partners participated again, and the round is expected to target as much as $300 million in total.
That makes Sarvam a unicorn. It also gives HCLTech something more useful than a line on a cap table. The Bengaluru startup has been selected under the IndiaAI Mission to help build indigenous foundation models, and its current stack is aimed squarely at Indian languages, government services and enterprise deployments where data location, compliance and language support are not side issues. For HCLTech, the investment is less like a passive venture bet and more like an early position in a domestic AI infrastructure layer.
Sarvam was founded in 2023 by Vivek Raghavan and Pratyush Kumar, both previously associated with AI4Bharat at IIT Madras. Raghavan has worked on digital public infrastructure at population scale, while Kumar has led multilingual AI research. In February 2026, the company introduced Sarvam 105B and Sarvam 30B at the India AI Impact Summit in New Delhi, and later released the models as open-source under the Apache 2.0 license. The 105B model is built for heavier reasoning tasks, while the 30B model is meant for lower-latency deployment.
Business Standard has reported that Sarvam 105B compared favorably with DeepSeek R1 on some benchmarks despite using a smaller active parameter footprint, and that it performed strongly against Gemini Flash in selected tests. Benchmark claims should always be treated carefully, especially when they come from a young model lab trying to prove itself. Still, Sarvam does not need to beat every frontier model on every global test to matter. It needs to be good enough in Indian languages, cheap enough to deploy and trusted enough for public-sector and regulated-industry work.
That is where the structural argument becomes interesting. OpenAI, Google and Anthropic have far larger research budgets, but their general-purpose systems were built around a global market where English has carried much of the training and product design. India has 22 official languages and a population of more than 1.4 billion people. A ministry in Odisha, a cooperative bank in Tamil Nadu or a health program serving rural users cannot treat Hindi, Tamil, Telugu, Bengali or Marathi as a localization layer added after the model is built. For many use cases, language is the product.
Sarvam's advantage is that it started with that problem instead of arriving at it later. Its models, speech systems and agent products are being built around Indian language data, code-mixing and voice-led interaction, which is closer to how many Indian users will actually encounter AI. HCLTech's relationships in banking, insurance, government technology and large enterprise accounts give Sarvam a channel into customers that a model lab would otherwise spend years trying to reach.
The fresh capital is expected to support more compute, stronger models and wider enterprise deployment. That matters because foundation model companies burn money quickly, and India's domestic AI effort cannot run on national ambition alone. Training larger models, serving them reliably and maintaining them for banks or government departments are expensive, slow jobs. A $234 million first close does not solve that permanently, but it changes the scale at which Sarvam can operate.
There are still obvious risks. The second close has not been completed, sovereign AI policy can shift, and US frontier labs are not ignoring India. Google and OpenAI have both spent the past year pushing India-specific initiatives, and their products already have consumer reach that Sarvam cannot match overnight. HCLTech also has to prove that a services-led distribution machine can help a foundation model startup move quickly without pulling it into slow enterprise sales cycles.
For now, Sarvam has something rare in the AI market: government relevance, open models, a clear language problem and a strategic investor that can carry it into the institutions most likely to pay. The deal is current because it is not only about another startup crossing a billion-dollar valuation. It is about whether India's AI stack will be bought from overseas labs, built at home, or assembled through partnerships that look a lot like this one.
Also read: Salesforce spends $3.6 billion on Fin to buy proof it could not build in time • PatSnap's dual IPO filing sets Hong Kong and Singapore against each other • Enflame's STAR Market IPO puts Tencent's chip strategy on public trial
