Sarvam faces backlash for offering Chinese AI models on its platform

This week, Sarvam unveiled Token Factory, an inference platform offering multiple AI models via a single API interface, hosted on local computing infrastructure

Sarvam AI
The controversy stems from Sarvam’s history. It built its frontier model on the backing of the government’s IndiaAI Mission, and was one of the first local startups to be given subsidized compute
Bloomberg
2 min read Last Updated : Nov 19 2026 | 10:30 AM IST
By Saritha Rai
 
Indian startup Sarvam has begun selling access to DeepSeek and other Chinese AI models, triggering an online backlash over its role as one of the main proponents of the country’s AI independence.
 
This week, Sarvam unveiled Token Factory, an inference platform offering multiple AI models via a single API interface, hosted on local computing infrastructure. Its catalog includes Sarvam’s own model, but also open-weight systems from Chinese AI labs DeepSeek, Moonshot AI and Z.AI Co.
 
Sarvam, valued at $1.5 billion in April, is the poster child for India’s push to create AI services tailored to the languages and cultures of the world’s most populous country. Prime Minister Narendra Modi’s government has made a forceful case that countries beyond the US and China need to control their own destiny with the emerging technology.
 
The controversy stems from Sarvam’s history. It built its frontier model on the backing of the government’s IndiaAI Mission, and was one of the first local startups to be given subsidized compute.
 
“We haven’t changed our motto. We continue to build AI models for India. We will be a full stack AI company,” co-founder Pratyush Kumar said in an interview following a rash of comments online questioning its mission.
 
The Token Factory offering’s commercial logic is quite straightforward, he said. Selling AI tokens brings in revenues and makes the business sustainable as building models is capital-intensive. The Bangalore-based startup, backed by Lightspeed and billionaire Vinod Khosla’s Khosla Ventures, already serves over 450 billion tokens daily on the DeepSeek V4.1 Flash model, it said in a blogpost.
 
The infrastructure hosting the new service is separate and runs on computational power financed by its investors, the founders emphasized. “Those chips are entirely funded by private money,” the second co-founder Vivek Raghavan said in the same interview.
 
“Anytime you use a model, open or proprietary, hosted on outside servers, data exfiltration is a possibility,” he said. Sarvam’s latest offering ensures India’s AI users’ data stays within its borders, said Raghavan. “Otherwise, the economics of AI for countries like India would be that Indians pay for the tokens and then double-pay with their own data that AI models amass.”
 

More From This Section

Topics :Artificial intelligenceartificial intelligence and roboticsDeepseekAI Models

First Published: Nov 19 2026 | 10:30 AM IST

Next Story