India is offering AI computing at prices far below global rates. Here is how the subsidy works and why it matters.

The IndiaAI mission’s common compute pool has over 38,000 GPUs from 10 providers, including Yotta, Jio, Tata Communications, E2E and CtrlS. The standard price is Rs 115.85 per GPU-hour, with Nvidia H100 chips at about Rs 150.

A subsidy of up to 40 per cent brings the effective cost below Rs 100 an hour for projects of national importance. Comparable capacity costs $2 to 4 an hour on AWS or Azure, roughly two to four times more.

The goal is to let Indian startups, researchers and students train AI models without renting expensive foreign cloud capacity. The mission plans 100,000 GPUs by end-2026, backed by Rs 4,563.36 crore of the Rs 10,371 crore mission outlay.

Cheap compute alone does not build an AI ecosystem; data, talent and research funding matter too. But by removing the cost barrier, India is betting it can keep its AI builders at home.