India’s Sarvam AI nears $1.6b valuation in funding round: sources

Sarvam AI, an India-based startup building language-focused AI models, is nearing a US$300 million to US$350 million funding round valuing it at US$1.5 billion to US$1.6 billion, people familiar with the talks said.

Bessemer Venture Partners is expected to lead the round, with Nvidia, Amazon, and Prosperity7 Ventures also participating, the people said, adding the deal could close as soon as next week.

Founded in 2023 by AI researchers Vivek Raghavan and Pratyush Kumar, the Bengaluru company builds voice-first models that support 22 Indian languages, and it has said this could help adoption in India.

Sarvam also offers agentic AI tools for tasks such as coding and meeting planning, while co-founder Kumar said at a February event that it wants to bring AI to more Indians.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Government backing and Nvidia collaboration fuel Sarvam’s valuation

The funding round rests on more than technology. Sarvam AI plays a central role in the Indian government’s IndiaAI Mission, a national program that aims to build India’s domestic AI capabilities 1.
That tie-up brings direct support. Sarvam received nearly Rs 99 crore (about $12 million) in subsidies to secure a cluster of 4,096 Nvidia H100 SXM GPUs, high-end AI chips used to train and run large models, through Yotta Data Services, an Indian data-center and cloud provider 2.
Nvidia also contributed at the engineering level. Sarvam worked with Nvidia on hardware-software co-design that delivered a 4x inference speedup, faster model responses at run time, for Sarvam’s Sovereign 30B model on Nvidia Blackwell versus baseline Nvidia H100 GPUs 3.

Sarvam’s rise signals a new focus on sovereign AI and serving-cost efficiency

Sarvam’s momentum sketches a template for “sovereign AI.” It prioritizes performance in Indic languages and Indian workflows where global models often fall short, including OCR (optical character recognition, which converts images of text into machine-readable text) and speech benchmarks 4.
The results also support a more multi-polar AI market. Regional champions can meet local enterprise and government needs rather than relying on a small set of dominant players 5.
The company puts sustained effort into lowering inference costs, which cover the expense of running models to generate answers for users 3.
As AI adoption matures, advantage may move from raw benchmark scores to the economics of serving models at scale, especially in price-sensitive markets 3.