Sarvam AI
Sarvam AI was founded in August 2023 by Vivek Raghavan and Pratyush Kumar. It develops AI models and services for Indian languages. The company is based in Bengaluru and names Axonwise Private Limited as its legal business on its website.
Company & leadership
- CEO
- Pratyush Kumar, identified in NVIDIA’s March 2026 partnership announcement
- Base
- Bengaluru, Karnataka, India
- Legal business
- Axonwise Private Limited
- Investors named by Sarvam
- Lightspeed, Peak XV Partners, Khosla Ventures, HCLTech, Bessemer Venture Partners and NVIDIA
Founders
- Vivek Raghavan and Pratyush Kumar
Founders and company focus
Sarvam’s account of its founding links Raghavan’s experience in India’s digital public infrastructure with Kumar’s work on open AI for Indian languages. The company develops its own models as well as services for enterprises, developers and government.
In March 2026, NVIDIA named Sarvam as an inaugural member of the Nemotron Coalition. The announcement identifies Pratyush Kumar as Sarvam’s cofounder and CEO. This provides a direct partner source for the leadership title.
Funding milestones
Sarvam announced a $41 million Series A on 7 December 2023. Lightspeed led it, with Peak XV Partners and Khosla Ventures participating.
On 15 June 2026, it announced a $234 million first close toward a $300 million Series B, at a $1.5 billion post-money valuation. HCLTech and Bessemer participated alongside Khosla Ventures and Peak XV Partners. The announcement specified HCLTech as lead strategic investor with a planned $150 million investment. The headline round target and completed first close are different figures.
Research and product milestones
The company’s September 2026 releases included Vision 2.1 on 24 September and Saaras V4 on 25 September. Vision added form and handwriting capabilities. Saaras introduced an in-house hybrid state-space decoder for speech recognition.
Sarvam also publishes weights. These include its 105B reasoning model and Prism, which returns probabilities over fixed answer choices. The firm therefore supplies both generative models and a separate decision-model interface.