Sarvam AI Raises $234M, Plans Full-Stack AI for India
In June, Sarvam AI entered India's unicorn club, raising $234 Mn (about ₹2,210 Cr) in a $300 Mn Series B round at a $1.5 Bn valuation. A month after the raise, at its inaugural Sarvam Epoch developer conference, the startup signalled the funding was only the beginning.
It laid out plans to become a full-stack AI venture covering foundation models, infrastructure, and applications across speech, vision, coding, cybersecurity, and defence — a sharp pivot from its origins as a builder of Indian-language models like OpenHathi-v1.
What Sarvam AI builds
Sarvam provides an India-hosted inference platform for running frontier open-weight models, including its own 105-billion-parameter model alongside models such as GLM 5.2 and Gemma 4. It is also building a trillion-parameter frontier model.
On the application side, its product line spans a coding agent (Sarvam Code), voice agents (Samvaad, which has handled 325 Mn minutes of customer conversations), a speech model supporting all 22 constitutional Indian languages (Saaras V4), document intelligence, workplace agents (Sarvam Work), and even hardware — smart glasses under the brand Sarvam Kaze.
Newsletter
The corridor, every morning.
Funding rounds and cross-border capital moves, one email, five minutes.
Why this matters from outside India
For global enterprises and investors eyeing India, Sarvam is making a deliberate sovereignty play. Its inference platform allows enterprises to run open models on infrastructure inside India, targeting government departments, defence organisations, and regulated industries that require sensitive data to remain within the country.
Its Anvaya platform is purpose-built for defence and national security use cases and can run entirely on-premise. On pricing, Sarvam positions itself aggressively: Samvaad costs ₹3.5 a minute against an industry range of roughly ₹8–12, according to the startup. The full-stack approach — owning the model, infrastructure, and application layers — is aimed at enterprises that value the right model at the right price with the right security, rather than the world's best model for every task.
The hire of Devendra Singh Chaplot, a founding member at Mistral AI and former xAI pre-training lead, and a new San Francisco office, underscore that this ambition extends beyond India's borders.
More in Funding
