Bengaluru-based Sarvam AI has introduced Token Factory, a commercial platform enabling developers and enterprises in India to access diverse AI models without hardware procurement, marking a significant advancement in local AI infrastructure.

  • Token Factory offers API-based multi-model access without GPU ownership.
  • Pricing based on input and output tokens, starting at ₹10 and ₹25 per million tokens.
  • Platform supports over 450 billion tokens daily, highlighting strong demand.

What happened

Sarvam AI launched Token Factory on November 16, 2026, a platform designed to provide scalable AI inference services to Indian developers and enterprises. It enables users to access various AI models through a single API, removing the need to purchase GPUs or manage the underlying infrastructure. The platform hosts Sarvam’s proprietary Sarvam 105B model and several open-weight models, billing customers per tokens consumed rather than through GPU time or server usage.

This development introduces a new layer to India’s AI infrastructure ecosystem, complementing existing investments into data centers and hardware. Token Factory operates using GPUs privately procured by Sarvam, separate from government-subsidized resources previously allocated under the IndiaAI Mission. The platform reportedly handles more than 450 billion tokens daily, indicating robust market uptake.

Why it matters

The Token Factory addresses critical operational challenges for Indian AI application developers by removing hardware burdens and providing flexible model access. This directly supports the growing demand for efficient AI inference services, which translate trained models’ capabilities into real-time responses for users. The platform’s pay-per-token pricing also introduces cost transparency and scalability, which are important for startups and enterprises managing AI workload costs.

Furthermore, Sarvam’s move underscores evolving debates in India's AI ecosystem regarding the balance between indigenous model development and utilizing open-source or foreign-developed models. While Sarvam was selected for government support under the IndiaAI Mission, the Token Factory currently does not use subsidized GPUs and hosts several Chinese-origin open-weight AI models, highlighting the complexity of fostering domestic AI sovereignty alongside commercial viability.

What to watch next

Market observers will be closely following how Sarvam leverages its Token Factory revenue to support its ongoing development of homegrown large-scale AI models, including its flagship Sarvam 105B. The company aims to demonstrate progress on proprietary AI capabilities by the end of 2027, a timeline that could influence investor confidence and competitive positioning.

Additionally, scrutiny of the wider token-based inference platform category in India is expected, as other players such as Larsen & Toubro, E2E Networks, and niche AI security firms develop similar offerings. The outcome of efforts to balance indigenous AI technology development with practical deployment models using open-weight architectures will be critical for India’s ambition to become a global AI hub.

Source assisted: This briefing began from a discovered source item from Inc42 India. Open the original source.
How SignalDesk reports: feeds and outside sources are used for discovery. Public briefings are edited to add context, buyer relevance and attribution before they are published. Read the standards

Related briefings