Sarvam faces backlash for selling DeepSeek and other Chinese AI models on its new Token Factory; it says no subsidy was used
India's best-known 'sovereign AI' startup launched a platform to run AI models on its GPUs in India. Critics noticed most of the catalogue is Chinese.

- Launched 18 November 2026
- 12 models listed; prices from ₹14 per million input tokens
- ₹98.68 crore GPU subsidy received for model training
What Sarvam launched
Sarvam opened Token Factory on 18 November. It lets developers and companies use many AI models through one API, running on Sarvam's GPUs in India, with prices in rupees. Launch pricing is ₹14 per million input tokens and ₹55 per million output tokens.
The list of 12 models includes Sarvam's own 105B models alongside DeepSeek V4.1 Flash, DeepSeek V4 Pro, Moonshot's Kimi K3.1 Pro, Z.ai's GLM 5.3 and Alibaba's Qwen, plus Google's Gemma.
Why people are upset
Sarvam was picked by the government in April 2025 to build India's own foundation model, and it got a ₹98.68 crore GPU subsidy. Critics such as RunAnywhere CEO Sanchit Monga said the company is now 'mostly serving Chinese models' instead of shipping frontier research. Bloomberg reported an online backlash.
Sarvam's answer
Co-founder Pratyush Kumar says Token Factory runs on GPUs Sarvam bought itself, and that its discounted GPU access under the IndiaAI Mission ended in March 2026. Requests are not sent to the model makers' servers abroad, it says.
Some disagree with the critics. MediaNama founder Nikhil Pahwa sees nothing wrong with hosting Chinese models, and Sarvam investor Lightspeed's Hemant Mohapatra called it a timely business during a global GPU shortage.




