Editorial illustration for Sarvam's Token Factory Shifts India's AI Sovereignty Debate
Sarvam's Token Factory Shifts India's AI Sovereignty Debate
Sarvam AI picked up ₹32 crore from the IndiaAI Mission in 2024 on the promise of building India's own foundation model. On November 16, it launched something else entirely: Token Factory, an inference platform that routes developers to eight different AI models through one API, with all processing done on servers inside India. The catalogue includes Sarvam's own 105B model, but also DeepSeek, Kimi and GLM, all built by Chinese labs, plus a model from Google. Pricing for DeepSeek V4.1 starts at ₹14 per million input tokens.
The launch has put Sarvam back in a position it has occupied before: at the center of an argument about what India actually means when it says "sovereign AI." Is it enough to keep data processing within Indian borders, running on Indian GPUs, even if the underlying models come from Beijing? Or does a company that took government money to build indigenous AI owe the country something closer to home-grown intelligence, not just home-hosted infrastructure for someone else's? The answer isn't settled, and Sarvam's own choices have made it harder to avoid asking.
Ultimately, Sarvam’s Token Factory may be a sensible business decision. But it also raises a question that extends beyond the company: should a company receiving public support to build India’s AI capabilities be judged primarily by its commercial success, or by how much original AI technology it contributes to the country?
Why this matters
Sarvam received government backing to build sovereign AI for India, not to become a reseller for Chinese model makers. Token Factory's pitch, keep data processing local while giving developers one API for many models, sounds practical. But practicality isn't the mandate taxpayers funded.
Hemant Mohapatra's defense of the move tells us the sovereign AI label in India has gotten elastic enough to cover data residency alone, divorced from who actually trains the models running on that infrastructure. For developers building on Token Factory, the convenience is real: fewer integrations, data that stays within Indian borders, one bill. For India's broader AI strategy, it's a tell.
If a flagship sovereign AI company's answer to "why host foreign models" is "because customers want convenience," the term has stopped meaning model ownership and started meaning server location. Worth watching: whether Sarvam's own models gain share on this platform over time, or whether Token Factory becomes permanent infrastructure for other countries' AI, wrapped in Indian compliance. That distinction will say more about sovereignty than any press release does.
Further Reading
- Papers with Code - Latest NLP Research - Papers with Code
- Hugging Face Daily Papers - Hugging Face
- ArXiv CS.CL (Computation and Language) - ArXiv