TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

Sarvam is developing sovereign AI while also reselling DeepSeek models.

2026.11.19 - 04:38
Exclusive
Share

According to TechFlow, on November 19, Bloomberg reported that as the Indian AI company Sarvam advances its "sovereign AI" goals, it is also selling API access to Chinese models through its newly launched inference platform, sparking debate over the autonomy of India's AI sector.

This week, Sarvam opened its Token Factory platform to developers and enterprises, offering a suite of open-weight models hosted on GPUs within India, including DeepSeek V4.1 Flash, Moonshot's Kimi K3.1 Pro, Z.ai's GLM 5.3, Alibaba's Qwen 3.8 27B, Google's Gemma, and its self-developed Sarvam 105B; available on-demand or via dedicated compute. The company stated that DeepSeek V4.1 Flash processed over 450 billion tokens in a single day, with input pricing starting at 14 INR per million tokens.

The controversy stems from overlapping identities: Sarvam is a beneficiary of the Indian government's IndiaAI Mission, having previously secured domestic GPUs priced at 40% of their standard value for training its proprietary models. Co-founder Pratyush Kumar clarified that Token Factory operates on GPUs purchased independently by the company, noting that subsidized compute privileges expired in March 2026 and that the open-source models trained on them have already been published under their respective licenses on Hugging Face and AIKosh. Analyst opinions are divided: while some view running inference services as a viable business model for Indian firms, others argue that hosting foreign models on domestic infrastructure amounts to only "partial sovereignty," suggesting that the associated revenue should be redirected to fund indigenous model research and development.

Add to Favorites
Share to Social Media