India’s bet on homegrown AI just got significantly bigger. On June 15, Sarvam AI announced $234 million in new funding at a $1.5 billion valuation, making it India’s newest AI unicorn and one of the clearest signals yet that sovereign AI is moving from political ambition to funded reality.

The round is the first close of a targeted $300 million Series B. HCLTech led with a $150 million strategic investment, taking a 10.46% stake in the company. Bessemer Venture Partners joined alongside existing backers Khosla Ventures and Peak XV Partners.

What Sarvam Is Actually Building

Founded in August 2023 by Vivek Raghavan and Pratyush Kumar, Sarvam is building what it calls a full-stack sovereign AI for India. The phrase “full stack” matters here: they are not fine-tuning a Western model on Indian languages. They are training frontier models from the ground up, with every component developed, deployed, and governed entirely in India.

Their current model lineup includes:

Sarvam-30B — A mixture-of-experts architecture that activates approximately 1 billion parameters per token with a 32,000-token context window. Designed for efficient inference at scale.

Sarvam-105B — Their most capable model, activating around 9 billion parameters per token with a 128,000-token context window. Built for complex reasoning and enterprise deployments.

Sarvam Arya, Bulbul V3, and Sarvam Vision — Specialized models for Indian languages, voice, and multimodal tasks, designed to handle local linguistic nuances that general-purpose Western models regularly get wrong.

The new capital will fund compute access for Sarvam’s next generation of frontier models, with particular focus on agentic reasoning, software coding, and cybersecurity capabilities.

Why HCLTech Made the Bet

HCLTech is not a passive investor here. The IT services giant brings 220,000-plus enterprise clients, deep domain expertise across manufacturing, financial services, healthcare, and government, and years of data and software IP. The stated goal is to combine Sarvam’s model development with HCLTech’s enterprise transformation reach to accelerate AI deployment across Indian and global organizations.

For HCLTech, this is a play to own the AI services stack in a market that will see enormous AI-led transformation over the next decade. For Sarvam, it is access to real enterprise distribution from day one.

The Sovereign AI Argument

Sarvam’s founders were previously at AI4Bharat, the research initiative at IIT Madras that produced some of the first serious Indian-language AI benchmarks. Their founding thesis was simple: AI trained primarily on English and Western data makes systematic errors when applied to India’s linguistic and cultural context. If you are building a business intelligence tool, a customer-facing AI, or a government service that needs to work in Hindi, Tamil, Bengali, or Telugu, a sovereign model trained on that context outperforms a transplanted one.

India’s government reached the same conclusion. In April 2025, the Ministry of Electronics and Information Technology selected Sarvam under the IndiaAI Mission to develop an indigenous foundational model, providing government-backed GPU access for training.

The $234 million round shows that private capital now agrees.

What This Means for Business

The Sarvam story is not just about India. It is a signal about how the global AI market is maturing. The first wave assumed a handful of frontier labs would serve the world. The second wave, now underway, is about regional and domain-specific AI that is genuinely good at the tasks that matter to local businesses.

For data and AI leaders, the practical implication is this: model selection is becoming a real decision. A large US-based language model may be the default, but it is not automatically the best tool for every market, language, or workflow. Companies building AI-powered products for diverse geographies will increasingly need to evaluate sovereign and regional models alongside the global frontier players.

Sarvam is a direct example of that trend, but similar movements are underway across Europe, the Middle East, and Southeast Asia.

For Enterprise DNA’s community of data professionals and business leaders, the message is straightforward: AI literacy now includes understanding where a model comes from, what data it was trained on, and whether that training actually covers the context where you are deploying it.

The era of assuming one model fits all is closing. That is a good thing.