Tamil Nadu's colleges may not need to buy expensive graphics processing units (GPUs) to train students in artificial intelligence - but they still face a bureaucratic hurdle in getting access to the ones that already exist. Ministry of Electronics and Information Technology (MeitY) Secretary S Krishnan said academia and students are eligible to use subsidised GPU capacity through IndiaAI Compute, a union government scheme that has allocated 10,571 GPUs to date. The challenge, he told TNIE, is less about hardware shortages and more about institutions failing to tap resources earmarked for them.
The cost barrier for colleges
The Tamil Nadu AI Association has argued that typical AI coursework requires 15 to 20 research-grade GPUs for practical student training. A single GPU costs roughly Rs 20 lakh, which puts dedicated infrastructure out of reach for most colleges, according to Venkatesh Rajendran, chairman of the association.
The ministry's position is that universities shouldn't need to buy hardware outright. IndiaAI Compute operates on a shared, cloud-based model, letting eligible users access high-performance infrastructure remotely. Additional capacity exists through C-DAC's AIRAWAT facility and the PARAM Siddhi-AI system, which together offer close to 48 petaflops of processing power and 656 GPUs.
"No fixed percentage of compute capacity is earmarked exclusively for universities or undergraduate students," Krishnan said, even as he reiterated that academic users remain eligible to apply.
Eligibility vs. entitlement
The association has applied separately for around 15 GPUs through the union ministry, with plans to partner with a university to distribute the capacity to students if approved. Krishnan said the application would be assessed under the IndiaAI Compute End-User Policy, weighing eligibility, project justification, and technical need.
That approval process is a significant practical obstacle for institutions. Cloud allocations require project justification and case-by-case sign-off, but coursework demands routine, large-scale student access for training models and handling datasets. A NITI Aayog committee is now examining the broader question, Krishnan said.
An AI learning path for teachers could help address part of the gap, but computing power is not the only constraint. Dr N Bhalaji, a higher education expert in computing science, said colleges also lack curated datasets, sandboxed environments, and faculty with production-level AI experience. "I have watched thousands of students arrive with courage, but not always with the skills and tools their ambition deserves," he said.
State target: 5 lakh trained by 2031
Meanwhile, Tamil Nadu's 2026-27 budget includes an AI Economy Mission with a target of training 5,00,000 people in AI skills by 2031, delivered through existing engineering colleges, polytechnics, and ITIs rather than new dedicated infrastructure. Rajendran added that AI coursework remains heavily theory even as employers demand hands-on experience.
If the state is going to meet that target through existing institutions, faculty and students need practical pathways to shared compute resources - not just eligibility on paper. For educators already planning AI for education, the takeaway is that applying early and clearly for subsidized GPU access is a meaningful step, since funding is substantial but approval is competitive.
Why this matters for education professionals
The tension here isn't just administrative. For lecturers and trainers, the distinction between owning hardware and borrowing cloud time changes how courses are structured. A professor who must justify each request for GPU hours will handle student production workflows differently than one who has dedicated access. And students who learn on only theoretical AI models emerge underprepared for employers expecting hands-on model training and dataset handling. For education professionals, the practical implication is straightforward: build courses that make the most of any subsidized capacity you can get - and budget time for the approval process, not just the compute hours.
Your membership also unlocks:
