Sarvam AI, the Bengaluru-based sovereign AI startup, is near-closing a $300-350 million funding round at a $1.5-1.55 billion valuation, according to Outlook Business. Bessemer Venture Partners is expected to lead the round, with Nvidia, Amazon, and Prosperity7 Ventures participating. If completed, it would be the largest private funding deal for an Indian startup in 2026.
The company was founded in 2023 by Vivek Raghavan and Pratyush Kumar, who previously collaborated on AI4Bharat at IIT Madras. Sarvam builds full-stack generative AI systems spanning foundational model research to enterprise deployment tools, all designed for India’s linguistic landscape.
The Indic Language Problem
India has over 1,600 dialects and 22 official languages. English-centric AI models fail across most of the country’s 1.4 billion population. Sarvam’s core technical advantage is building models that natively handle Hindi-English and Indic language code-switching, the fluid mixing of languages that is standard in Indian communication, according to Rest of World.
The company unveiled two large language models at the India AI Impact Summit in February: a 30 billion parameter model and a flagship 105 billion parameter model, both trained from scratch in India. Outlook Business reports the models were positioned as cost-effective alternatives with strong benchmark performance on Indian language tasks.
There is also a tokenization cost problem. A sentence in Hindi requires three to four times more tokens than the same sentence in English, making every AI interaction in an Indian language significantly more expensive. “The same question, when asked in English, costs one-fifth of what it costs in an Indian language,” Raghavan told Rest of World. Sarvam built optimized tokenizers to close this gap.
Sovereign AI as Government Strategy
Sarvam is part of a broader Indian government push. Under the IndiaAI Mission, the government is directly funding 12 AI organizations to build sovereign foundation models, including IIT Bombay’s BharatGen consortium. Sarvam has secured a significant allocation of Nvidia H100 GPUs through Yotta Data Services, per Outlook Business.
Microsoft’s $17.5 billion India data center commitment is materializing with a flagship Hyderabad facility going live this summer. Krutrim, backed by Ola founder Bhavish Aggarwal, already achieved unicorn status. The pattern across Indian AI: build infrastructure specifically designed for local languages and regulatory environments rather than renting access to Western models.
Agentic Applications at Scale
For agent builders, the relevance is direct. Sarvam’s models are designed for voice-enabled, multilingual conversational agents, according to Rest of World. The company is deploying agents that allow rural patients to access medical advice through WhatsApp and low-bandwidth interfaces, and building AI tutors that deliver instruction in students’ mother tongues. These are frugal AI agents built to run on low-end smartphones with intermittent connectivity.
India’s sovereign AI stack is the non-Western agentic infrastructure buildout most worth watching. When agents are doing work in Indic languages under Indian regulatory requirements, the foundation models they run on need to be built for that context. Sarvam is building that layer.