Skip to content
MukhyamantriMukhyamantri
TechnologyDelhi

A made-in-India AI platform is now serving DeepSeek, at 450 billion tokens a day

Sarvam AI launched Token Factory on Monday, an inference platform that runs open-weight models from DeepSeek, Moonshot AI, Z.ai and Alibaba on servers in India. The company says it already serves more than 450 billion tokens a day through DeepSeek V4.1 Flash. Sarvam is a beneficiary of the Rs 10,370 crore IndiaAI Mission, with Rs 246.72 crore of support.

· · 1 min read

New Delhi: One of India's most prominent homegrown AI startups, and a beneficiary of the government's sovereign AI programme, is now selling models built by Chinese companies. Sarvam AI announced Token Factory for developers and enterprises on Monday.

The platform offers open-weight models from DeepSeek, Moonshot AI, Z.ai and Alibaba, alongside Google's Gemma and Sarvam's own models. The list includes DeepSeek V4.1 Flash, Kimi K3.1 Pro, GLM 5.3, Qwen 3.8 27B and Sarvam 105B. Some are available on demand, while others need dedicated capacity.

Inference is the business of running a model that has already been trained. Instead of buying expensive hardware, a company calls an API and pays by the token. Together AI, Fireworks AI, DeepInfra and OpenRouter do this globally, and Amazon's AWS Bedrock offers the same access to models from several developers.

Sarvam says it is already serving more than 450 billion tokens daily through DeepSeek V4.1 Flash. Hosting the models in India means a business does not have to send its data to servers run by Chinese companies. Local hosting does not settle every security question, though, including those around model behaviour, software dependencies and access controls.

The decision lands awkwardly for a company that was selected under the Rs 10,370 crore IndiaAI Mission to build homegrown foundation models. Government disclosures put Sarvam's financial and computing support at Rs 246.72 crore.

The argument underneath is about what sovereignty means. Building frontier models at home, or keeping the pipes they run on under Indian control? Using open-weight foreign models does not by itself weaken domestic work. It lets Indian firms build applications, develop expertise and serve customers while indigenous capability matures.

Sarvam is competing on price, speed and efficiency with global inference providers, and it is now doing so with a catalogue that includes the very models New Delhi's programme was meant to replace.