Sarvam AI
India's first sovereign-AI champion — foundation models trained on 22 Indian languages.
sarvam.ai ↗Sarvam is India's most serious AI company by a wide margin — and the only Indian startup that's credibly positioned to be an AI platform rather than just an AI application. The sovereign-AI thesis for India is simple: 1.4 billion people, 22 languages, a government that wants indigenous AI infrastructure, and no incumbent national champion. Sarvam is the one company with the team, the government relationships, and now the capital to own that position. For investors looking for Indian AI exposure with real defensibility, Sarvam is the clean pick.
What's going for them.
- 01Sarvam-105B (Feb 2026) is the first Indian-built foundation model to reach 105 billion parameters — trained entirely on Yotta's Indian GPU infrastructure, on 16 trillion tokens across 22 Indian languages + English. Sovereign-AI credibility other emerging-market labs can't match.
- 02Raising $300–350M at a $1.5B valuation — India's largest private AI funding round ever. Unicorn status achieved within 30 months of founding, unprecedented pace for Indian deep-tech.
- 03Cap table blends global institutional investors (Bessemer, NVIDIA, Lightspeed, Khosla, Amazon) with Indian strategic partners (HCLTech) — the structure Indian AI has needed for a decade and finally has.
- 04Government of India's IndiaAI mission has effectively anointed Sarvam as the national AI champion — which provides a pipeline of domestic deployments that no foreign lab can access.
- 05Vivek Raghavan is the ex-Aadhaar tech architect; Pratyush Kumar is ex-Microsoft Research and a deep-learning researcher. The co-founder pairing combines India-scale infrastructure experience with frontier ML research — rare for any emerging-market AI lab.
What they built
Sarvam AI builds foundation models specifically for Indian languages and Indian enterprise workflows. The product portfolio includes Sarvam-1, Sarvam-2B (smaller efficient models), Sarvam-30B (MoE), and the new Sarvam-105B flagship (launched February 2026) — trained end-to-end on Indian infrastructure, covering 22 Indian official languages alongside English. The commercial stack spans an API (enterprise and developer tiers), deployments for the Government of India’s IndiaAI mission, financial-services-specific deployments for Indian banks and insurers, and telecom partnerships for regional-language conversational AI.
How they got here
Vivek Raghavan (Aadhaar technology chief architect) and Pratyush Kumar (Microsoft Research, deep-learning researcher) founded Sarvam in 2023. The co-founder combination was unusually well-suited to the task: India’s single biggest AI challenge is not model quality (the models work) but deployment at population scale across linguistic diversity — exactly the problem space that Aadhaar solved at identity level. The initial $41M seed + Series A in late 2023 was led by Lightspeed with Peak XV and Khosla Ventures participating.
The February 2026 release of Sarvam-105B — trained on Yotta’s GPU clusters in India, on 16 trillion tokens across 22 Indian languages plus English — was the capability inflection. By March 2026, Sarvam was in advanced talks for a $300–350M Series B at a $1.5B valuation, led by Bessemer with NVIDIA, Accel, HCLTech, Amazon, and Prosperity7 all participating. That’s the broadest-based Indian deep-tech round in history, and makes Sarvam the first Indian AI startup with both unicorn valuation and international institutional credibility.
What’s ahead
Three questions define Sarvam’s trajectory over the next 18 months. First, government deployment: the IndiaAI mission and various state-government AI initiatives represent a multi-hundred-million-dollar addressable market for Sarvam specifically, if commercial terms can be negotiated without getting stuck in procurement bureaucracy. Second, enterprise revenue: Indian banks, insurers, telcos, and retailers all have concrete use cases for Indic-language AI; the question is how fast Sarvam can commercialize beyond lighthouse pilots. Third, model leadership: Sarvam-105B is competitive on Indian-language benchmarks with anything in the world, but maintaining that lead as Mistral, Cohere, and Chinese labs also push into regional-language models will require continued training investment.
Why it matters
Sarvam is the clearest proof point that sovereign AI is a real category, not a buzzword — and the single most important non-Chinese, non-US AI company in Asia-Pacific outside Japan’s specialized labs. For AI founders in emerging markets, Sarvam’s playbook (start with a specific national linguistic need, build government + enterprise dual distribution, attract strategic + institutional capital) is the reference case. For investors, Sarvam is the one Indian AI story that’s now investable at institutional scale with real defensibility.
Founder interview coming soon.
We'll be sitting down with the founders and operators of the companies we profile — on fundraising, product decisions, and what they're building next. If you're part of the Sarvam AI team and want to share a perspective, get in touch.
Thinking about fundraising or M&A?
Amafi Advisory works with AI companies on strategic, fundraising, M&A, and technical advisory. Even if you're just exploring.