Sarvam AI: Building sovereign AI for India

Every major economy today faces a key question: whose AI will power national infrastructure?  For India, the answer has implications for strategic long-term autonomy. With the shifting geopolitical landscape, applications critical to national security and economic well-being, defense systems, national intelligence, citizen-scale voice services, and regulated financial and healthcare infrastructure, need control of their own destiny. India’s “atmanirbhar” (self-reliance) policy framework, and the India AI Mission’s goal to create a national AI ecosystem, are propelling a generation of entrepreneurs building sovereign AI for India.

Developing sovereign AI for India means serving 1.4 billion citizens, mission-critical sectors, and enterprises and MSMEs forming the lifeblood of the world’s fastest-growing major economy. It’s AI that must be voice-first (connecting hundreds of languages and thousands of local Indic dialects) and should democratize thousands of agentic and reasoning use cases spanning customer-facing and back-office workflows across organizations. Above all, it needs to be built, trained, and deployed in India to be genuinely world-class. Building this kind of stack requires a rare combination of top-tier research talent, deep institutional relationships, and the engineering discipline to build indigenous technology that scales at indigenous price points.

Enter Sarvam: India’s full-stack sovereign AI platform

Sarvam AI was founded in 2023 by Dr. Pratyush Kumar and Dr. Vivek Raghavan, two of the most accomplished digital infrastructure builders in India. Pratyush co-founded AI4Bharat at IIT Madras, the country’s foremost open-source Indic language AI initiative. Vivek spent more than a decade as Chief Biometric Architect at UIDAI, where he helped design Aadhaar, the world’s largest biometric identity system, and was a key architect of UPI, India’s dominant payments rail. 

Sarvam has built a full-stack sovereign AI platform: from foundational models trained from scratch in India, to commercially deployed applications already handling billions of enterprise interactions. Sarvam’s flagship mixture-of-expert models, Sarvam 30B and Sarvam 105B, are competitive with global open and closed source models across a range of knowledge, reasoning, and agentic benchmarks. Sarvam’s Bulbul text-to-speech and Saaras speech-to-text models are state-of-the-art in Indic voice use cases. Platforms ranging from Samvaad to Arya power the full spectrum of conversational and agentic workflows. There has been a significant acceleration of Sarvam’s solutions in the past few months, both for the government and enterprises in banking, financial services, insurance and healthcare. Sarvam’s AI stack opens up sovereign use cases for enterprises, empowering them to leverage their private data, fine-tune models, and develop custom agents catering to their bespoke requirements.

At Bessemer, we believe that this is India’s AI moment, and Sarvam’s sovereign AI stack will be a foundational driver. This is why we’re proud to invest in Sarvam’s Series B and partner with Pratyush, Vivek, and the entire Sarvam team for the journey ahead.