If you’re training or running AI models in India, the single most accessible piece of government support is subsidised GPU compute under the IndiaAI Mission. Here’s what it is, who can use it, and the caveats that matter before you build a plan around it. (dgm implements osFoundry, a separate company’s platform — dgm is an independent integration partner, not osFoundry.)
What it is
Under the Compute Capacity pillar of the IndiaAI Mission, GPUs from a shared national pool are made available to startups, researchers and developers at well-below-market rates. The mission has been onboarding capacity in stages, with the stated aim of building one of the largest accessible AI compute pools in the country. The intent is to remove the biggest practical barrier to building AI in India — the cost of GPU time.
The rates (indicative — confirm before planning)
Government and media reporting has placed access rates in the region of roughly ₹115–₹150 per GPU-hour, described as more than 40% below market, with tens of thousands of GPUs onboarded and a target of around 100,000 (DD News).
Two honest caveats:
- These numbers change frequently as capacity is added and the programme evolves.
- Some figures are reported by secondary outlets ahead of official confirmation.
So treat any specific rate or GPU count — including the ones above — as a snapshot, and verify the current terms on indiaai.gov.in before you commit a budget to them.
Who qualifies and how to access it
The pillar targets startups, researchers, students and developers building or training models. Access is administered through the IndiaAI portal and a set of empanelled compute providers; the eligibility criteria and application route are defined there. If you’re an established enterprise that simply wants to use AI in operations rather than train models, this pillar is less relevant to you — your realistic levers are described in AI funding & government support in India.
What it does and doesn’t pay for
| Subsidised compute covers | It does not cover |
|---|---|
| Model training and fine-tuning | Commercial AI-software subscriptions |
| Compute-heavy inference and experimentation | Integration/consulting fees |
| Research workloads | A general “adopt AI” budget for an enterprise |
The benefit is lower infrastructure cost for building AI, not a purchase subsidy. It pairs naturally with self-hosting and open-weight models — a model-neutral platform like osFoundry can run open models on your own (or subsidised) GPUs, which is also the route to keeping data in India (see AI data residency in India, since osFoundry’s managed regions are US, EU and Japan — there is no India region, so India residency means self-hosting).
How dgm helps
dgm can design an AI build that uses compute efficiently and, where you’re an eligible startup or researcher, plan the architecture around accessible IndiaAI compute — running open-weight models on osFoundry so you control both the data and the GPU bill. dgm does not allocate the compute itself; that’s done through the IndiaAI portal. Engagements are a transparent $399 assessment then $3,999/month (INR approximate; 18% GST for domestic clients). Explore the platform at osFoundry, or talk to dgm about an efficient build.
General information, not financial advice. IndiaAI compute rates and eligibility change — confirm on the official portal before acting.