Sarvam AI Hits $1.5 Billion Valuation as HCLTech Bets $150 Million on India Sovereign AI

India-based Sarvam AI has been valued at $1.5 billion after a Series B first close that brought in approximately $234 million, led by a strategic $150 million investment from HCLTech. The Bengaluru startup sold a 10.46% stake to HCLTech for ₹1,427.25 crore as part of the round; returning investors include Khosla Ventures and Peak XV Partners alongside new participation from Bessemer Venture Partners. Sarvam is targeting a $300 million total for the Series B, with roughly $66 million still to close.

"Our investment in Sarvam marks a significant step toward building India's trusted and globally competitive AI ecosystem," said C. Vijayakumar, CEO of HCLTech. "By bringing together Sarvam's research in AI models with HCLTech's global presence, we are creating a differentiated full-stack AI platform for enterprises and governments."

The deal is notable both for its size and for HCLTech's strategic posture. The IT services giant, which reports roughly 227,000 employees and $14.7 billion in annual revenue, has existing partnerships with OpenAI and Google Cloud but chose a structural equity position in a three-year-old domestic startup rather than a client-supplier tie. Company executives say the plan is for HCLTech to build industry-specific solutions on top of Sarvam's foundational models, leveraging HCLTech's global enterprise relationships while Sarvam retains responsibility for model development and infrastructure.

What Sarvam has built

• Founders: Pratyush Kumar and Vivek Raghavan, both formerly with AI4Bharat at IIT Madras.
• Production scale: more than 2 million daily interactions and 10 million API calls.
• Customer deployments: a nationwide insurance voice campaign servicing 45 million policyholders; Sarvam Vision has digitized over 35 million pages of handwriting and Indian-language records.
• Valuation trajectory: pre-Series B valuation of roughly $196 million in 2025 to $1.5 billion post-money in 2026.

Sarvam's technical pivot underpins the investor enthusiasm. After criticism of its May 2025 Sarvam-M release — Deedy Das of Menlo Ventures labeled that model "embarrassing" for being fine-tuned from a French model rather than trained from scratch — Sarvam released two models in February 2026 trained entirely in India. Sarvam 30B and Sarvam 105B were trained "from first principles" using compute provisioned through the IndiaAI Mission and employ a Mixture-of-Experts (MoE) transformer backbone with 128 sparse expert feedforward networks.

• Sarvam 30B: 30 billion total parameters, ~2.4 billion active parameters per token, Grouped Query Attention (GQA), 32,000-token context window, trained on 16 trillion tokens, in production in Samvaad.
• Sarvam 105B: ~105 billion total parameters, ~9–10.3 billion active parameters per token, Multi-head Latent Attention (MLA), 128,000-token context window, trained on 12 trillion tokens, powers Indus.
• Tokenizer: Indic-optimized tokenizer achieves token fertility of 1.4–2.1 across supported Indian languages versus 4–8 tokens per word with standard multilingual tokenizers.

Outlook

The investment positions HCLTech to commercialize Sarvam’s models across industry verticals and geographies while preserving its existing vendor relationships. For Sarvam, the cash infusion accelerates deployment at population scale where per-token economics matter — a central rationale for its MoE architecture and custom Indic tokenizer. With $66 million left to reach the Series B target and institutional backing from both domestic and international investors, Sarvam is now squarely in the category of startups seeking to translate frontier model research into large-scale enterprise and government use cases.