LokVox

Science & TechTech policy

IndiaAI compute subsidy: model builders get ₹1,058 crore, researchers about ₹10 crore

Approvals on the IndiaAI Compute Portal show nine foundation model builders have taken most of the ₹1,275 crore committed so far, while students and academics have received less than 1%.

From the editorial desk · New DelhiPublished
The main building of the Indian Institute of Science, Bengaluru (file image)
The main building of the Indian Institute of Science, Bengaluru (file image)Photo: en:User:Mohammadulhaque. / Wikimedia Commons, CC BY-SA 3.0

Key points

  1. Nine foundation model builders have received ₹1,058 crore of the ₹1,275 crore committed.
  2. Researchers, students and fellows have received about ₹10.14 crore, or 0.8%.
  3. Academics say high costs, slow registration and grant rules keep them away from the subsidised GPUs.

India’s flagship scheme to give subsidised computing power to AI developers has so far sent most of its money to a handful of companies building large models, while university researchers and students have received a tiny share.

Of the ₹1,275 crore committed so far under the compute subsidy of the IndiaAI Mission, nine organisations building foundation models have received ₹1,058 crore. Researchers, students and fellows together have received about ₹10.14 crore, or 0.8% of the total. The figures come from the 304 compute approvals listed on the IndiaAI Compute Portal, as updated on 8 October.

Where the money has gone

Sarvam AI now has its subsidy listed at its full bill of ₹246.72 crore. BharatGen, led by IIT Bombay, has the largest allocation among the foundation model builders.

A separate category meant for researchers and academia carries ₹172.81 crore, but most of that goes to a single approval: 2,080 Google TPUs granted to Zenteiq Aitech Innovations, a company founded by a professor at the Indian Institute of Science (IISc). That one approval is larger than the 1,050 TPUs that a December 2025 white paper from the Office of the Principal Scientific Adviser counted for the whole pool.

The mission does not own any GPUs itself. Private cloud companies on its panel supply the computing, at a tender rate of about ₹140 an hour for an Nvidia H100 chip.

What researchers say

Academics say the scheme leans heavily towards language models, that costs remain too high for university labs, that registration is slow, and that the rules on their research grants make it hard to pay for cloud computing.

“Because of these infrastructure issues, good engineers tend to leave the country,” said Prathosh A.P., an associate professor at IISc Bengaluru. He still backs the scheme. “The IndiaAI Mission is definitely the right step towards making our country AI sovereign,” he said.

Manabendra Saharia, an associate professor of civil engineering at IIT Delhi, was blunter. “It sucks to think that we can be at the very frontier of our area, but can’t because we are poor,” he said.

Start-ups have their own complaints about the portal. Dhrumil Barot, founder of Vynara Technologies, described the upfront payment it demands. “It requires me to pay to reserve the compute, and then if I’m accepted, I’ll get 40 per cent back,” he said.

What comes next

Low uptake at home has led the government to consider letting the GPU suppliers on its panel sell capacity abroad. The IndiaAI Mission has not yet responded publicly to questions about how the subsidy has been shared out.