Urgent.News

What's breaking now, across hundreds of outlets.

AI

Sarvam’s Token Factory Raises India’s AI Inference Stakes

Sarvam AI’s latest announcement of its Token Factory has brought a new layer of India’s artificial intelligence (AI) infrastructure ecosystem…

Sarvam’s Token Factory Raises India’s AI Inference Stakes

Sarvam AI unveiled its Token Factory on November 16, a commercial platform that enables developers and businesses to access various AI models through a unified API, without the need to manage GPUs or infrastructure. This move highlights India's expanding AI infrastructure ecosystem, as companies focus on AI inference, the process of generating responses from trained models.

While data centers and GPUs have been primary targets for AI infrastructure investments, the increasing demand for AI applications is driving the need for efficient inference infrastructure. A token factory streamlines the process by handling model hosting, computing resources, and inference optimisation, charging customers based on input and output tokens rather than GPU capacity or server time.

Sarvam's Token Factory offers a diverse range of models, including its proprietary Sarvam 105B model and open-weight models like DeepSeek V4.1 Flash, Kimi K3.1 Pro, GLM 5.3, and Google's Gemma 4 31B, all priced in Indian rupees. The platform currently handles over 450 billion tokens daily, with the Sarvam 105B model priced at ₹10 per million input tokens and ₹25 per million output tokens.

Sarvam's launch of the Token Factory has raised questions about the use of Chinese-developed models and the utilization of subsidized GPUs under the IndiaAI Mission. However, Sarvam clarified that its Token Factory does not use subsidized GPU capacity and that the access to government-supported GPUs ended in March 2026. The company argues that the Token Factory helps maintain its B200 cluster between training runs and generates income while developing its trillion-parameter model.

Despite the revenue and data retention concerns, Sarvam's Token Factory provides a unique solution for prompt processing on Indian GPUs, keeping data within the country, which global providers may not easily offer.

Rewritten by urgent.news; may contain errors — check the original.

Read the original at inc42.com →

Report a problem with this story

More in AI

America, China, and the Costs of the Bipolar AI Race

The competition between Washington and Beijing over AI is intensifying. The rest of the world has little say in how this technology is built, tested, and governed.

  • U.S. and China dominate global AI, creating geographic divides.
  • 72% of AI computing capacity in U.S., 10% in China.
  • AI systems' safety and control largely decided by U.S. and China.

More from Saturday 21 November →