Sarvam AI’s new inference platform, Token Factory, which enables access to multiple global AI models including several Chinese ones, has triggered backlash in India over concerns it undermines the country’s sovereign AI goals by hosting foreign models on government-subsidized GPU infrastructure.

  • Sarvam’s Token Factory hosts multiple Chinese AI models on Indian GPUs.
  • Critics say government subsidies should foster indigenous AI development.
  • Sarvam clarifies Token Factory uses privately procured GPUs, not subsidized ones.

What happened

Bengaluru-based startup Sarvam recently launched Token Factory, an AI inference platform that allows developers and businesses to run a catalog of global AI models on infrastructure located in India. Despite its ambition to foster local AI innovation, Token Factory includes six Chinese open-weight models alongside Google’s Gemma and Sarvam’s own models. This mix raised significant criticism online, particularly concerning the use of government-subsidized GPU resources.

The backlash intensified after Sarvam secured a contract with India’s Ministry of Defence to build AI applications for the armed forces. Critics pointed out that Sarvam’s predominant hosting of foreign AI models contrasts with the government’s push for sovereign AI development and questioned the effective use of subsidies and capital raised by Sarvam.

Why it matters

India has prioritized developing sovereign AI capabilities to reduce reliance on foreign technology and protect data sovereignty. Sarvam’s Token Factory is intended to be a domestic alternative to foreign AI inference platforms by running models locally and pricing services in Indian rupees. However, the presence of multiple Chinese models on the platform raises concerns about whether the initiative truly aligns with India’s strategic objectives.

Additional concerns stem from fears of centralizing critical AI infrastructure within a single entity, which some argue could create vulnerabilities and uneven growth in India's AI ecosystem. The debate highlights the complexity of balancing open innovation with nationalism in fast-evolving technology sectors.

What to watch next

Sarvam has clarified that Token Factory operates solely on privately procured GPUs and no longer relies on government-subsidized GPUs, which were used only during an earlier phase ending in March 2026 for training models. How Sarvam addresses ongoing criticisms regarding the composition of its model catalog and infrastructure transparency will be crucial to its reputation and government relations.

The broader AI ecosystem in India will also be closely observed for how public-private partnerships balance opening AI innovation while safeguarding sovereign interests. Future government policies or subsidy guidelines may evolve to ensure domestic AI progress aligns with national security and economic goals.

Source assisted: This briefing began from a discovered source item from MediaNama. 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