The numbers
The IndiaAI Mission was approved in March 2024 with an outlay of ₹10,372 crore. Its common compute facility now has more than 38,000 GPUs onboarded through empanelled providers, offered to startups and academic institutions at subsidised rates; the government reported an average of about ₹65 per GPU-hour in mid-2025.
At the India AI Impact Summit in New Delhi in February 2026, the IT Minister announced plans to add 20,000 more GPUs.
Who benefits
Subsidised compute matters most for fine-tuning and evaluating models on Indian-language or domain data: agriculture advisories, legal documents, vernacular customer support, health triage.
Most business applications still do not need their own model. Retrieval over your documents with a hosted or open model is usually faster and cheaper to ship.
Outlook to 2030
Cheaper compute, Indian-language models and growing data-centre capacity point to more AI products built in and for India this decade. Tier-2 cities like Raipur gain because talent no longer has to move to where the hardware is.
Our advice: pick a narrow problem with data you own, prototype on hosted models, and move to dedicated compute only when usage justifies it.
Sources
Have a project like this in mind?
Talk To Our Team →
