• India has approved four AI startups, Sarvam AI, Soket AI, Gnani.ai, and Gan.AI, to build foundational large language models.
  • The selected startups will receive GPU support, while funding details are still being finalised by the government

In a strategic leap toward digital autonomy, India has approved proposals from four AI startups, Sarvam AI, Soket AI, Gnani.ai, and Gan.AI, to build its own foundational large language models (LLMs), marking a significant step in the INR 10,037 Cr IndiaAI Mission.

While more than 500 proposals are still awaiting clearance, this move signals a shift in the government’s stance on sovereign AI infrastructure.

The Pivot Toward Indigenous AI Models

Until recently, many in India’s tech ecosystem questioned the need for desi LLMs, citing cost and scale concerns. Industry veterans like Nandan Nilekani advocated leveraging existing global models to build applications locally.

However, the release of DeepSeek-R1, a Chinese foundational model developed for under $6 Mn, has prompted a change in sentiment.

Now, the push for India-specific models is seen as essential for maintaining data sovereignty and catering to linguistic and cultural diversity.

What the Startups Are Building

Soket AI is developing a 120 billion-parameter open-source text model, with a 7 billion-parameter version expected in six months.

Gnani.ai is building a 16 billion-parameter Voice AI model, while Gan.AI is working on a 70 billion-parameter multilingual model targeting advanced text-to-speech capabilities.

Sarvam AI, backed by Peak XV, has already released Sarvam - 1 (2B) and Sarvam-M (24B), the latter being a hybrid model designed for Indian languages and advanced reasoning. It plans to launch three more variants tailored for different use cases.

Despite progress, the government remains undecided on full collaboration details. What’s clear is that the selected startups will receive computational support in the form of GPUs. Soket AI has also requested a grant of INR 14.5 Cr, although it hasn’t received confirmation yet.

Gnani.ai’s founder similarly confirmed the importance of GPU access, even as funding clarity remains elusive.

To complement these efforts, the Centre has also issued tenders for GPU provisioning.

Edited by Annette George