On August 8th, the Prime Minister inaugurated Param Pragya, an AI-powered supercomputing facility at IIT Delhi’s Sonipat campus. Gold medals, 3,000 graduates, a shiny new machine — a glorious moment indeed.

But if you’re an executive trying to understand what this actually means for India’s AI ambitions?

So let’s do some comparison just to see how we are progressing in AI adoption.

The Same Week, A Different Scale.

While Param Pragya was being switched on in Sonipat, the US Department of Energy was several months into deploying a very different kind of infrastructure buildout.

At Argonne National Laboratory, two new systems — Solstice and Equinox — are coming online with a combined 110,000 NVIDIA Blackwell GPUs, delivering 2,200 exaflops of AI performance between them. At Oak Ridge, the Lux AI cluster is live, with a successor called Discovery already funded for 2028, built on next-generation chips designed specifically for what the US government is now calling “sovereign AI.”

And here’s the part worth sitting with: this isn’t a one-off. It’s the fourth or fifth machine in a lineage. Oak Ridge alone has fielded seven flagship supercomputers since 2004. Each one was the fastest system of its time when it launched.

Param Pragya’s technical specifications — the numbers that would let us make an honest comparison — haven’t even been published yet.

I want to be careful here, because this isn’t a “India is behind” column. That framing is lazy, and it isn’t useful to you.

What’s actually useful is understanding what different scales of compute are for, so you can calibrate your own expectations about what India’s AI infrastructure can realistically deliver in the near term — and where the real opportunity sits.

What These Machines Are Actually For.

The American systems are explicitly not being built to make chatbots faster. Look at what the DOE says out loud about Lux and Discovery: fusion research, materials discovery, quantum computing, grid modernisation, drug design.

This is compute aimed at national-scale scientific and industrial problems — the kind of work that takes a decade to pay off but changes what a country can manufacture, cure, or power.

That’s the real lesson for India, and it has nothing to do with GPU counts.

The question isn’t “can we build something as big as Argonne?” We can’t, not yet, and pretending otherwise wastes energy. The question is: what focused, well-chosen problems can India’s compute footprint actually solve well, right now?

A few candidates, based on where India already has genuine structural advantages:

Monsoon and climate modelling. India’s agricultural economy is more exposed to monsoon variability than almost any other major economy on earth. Higher-resolution climate simulation isn’t a nice-to-have research topic here — it’s closer to critical infrastructure.

Pharma and drug discovery. India already has a globally competitive generics and pharma manufacturing base. HPC-accelerated molecular simulation is one of the few AI use cases with a clean, provable ROI path — shorter discovery cycles, lower trial failure rates. This is where compute investment could show up on a P&L, not just in a research paper.

Materials science for manufacturing. With the PLI-driven push into semiconductors and electronics, computational materials design is directly useful — not academic, applied.

Population-scale genomics and public health. India has data scale that few countries can match. That’s a real asset if the compute and the governance around it are built together, not as an afterthought.

The Chief Economic Advisor, V.Anantha Nageswaran, has underscored the need for India to stay on top of critical technologies such as AI and hence has suggested skilling people to take its full advantage and stressed government-industry collaboration to harness the full AI potential. India’s IndiaAI Mission has set up 27 data and AI labs so far, spread across 21 states. Another 188 labs are planned for ITIs and Polytechnics, including 28 for Andhra Pradesh, but none of these has opened yet.

The Part Executives Should Actually Worry About.

Here’s the uncomfortable question buried under all the convocation photos: who gets to use this?

One detail in the Param Pragya coverage caught my attention more than the machine itself — a reader comment under one of the news pieces, essentially saying: this is encouraging, but we need this in state universities too.

That’s not a cynical take. It’s the right question. Compare it to what the NSF is doing with the Horizon supercomputer at UT Austin — a facility explicitly designed to be opened up broadly to researchers across the country, not just to one flagship institution.

Scale matters less than access, over a ten-year horizon. A powerful machine that serves five hundred researchers at one institute is a very different national asset than a smaller machine that serves five thousand researchers across fifty institutions.

If you’re building AI strategy for your own organisation, this is the same trap to watch for internally. The compute or the tooling you invest in matters less than who in your organisation actually gets access to it, and how fast that access spreads beyond the team that procured it.

What We’d Tell A Board About This.

If I were briefing a board on what Param Pragya signals, I’d keep it to three points:

One, this is a genuine and welcome step — national AI compute infrastructure is a legitimate strategic asset, and India building it at all matters more than the specs on day one.

Two, don’t benchmark India’s AI ambitions against US hyperscale numbers. It’s the wrong comparison, and it will lead you to either false confidence or false despair, neither of which is useful for actual decision-making. Benchmark against a narrower question: which specific, high-value problems can this infrastructure credibly solve in the next three to five years?

Three, watch the access model more closely than the machine. That’s the leading indicator of whether this becomes a national capability or a very expensive photo opportunity.

The Investment Question Nobody’s Asking Loudly Enough.

Here’s the piece that’s missing from most of the coverage of Param Pragya, and of India’s AI infrastructure push generally: none of this works on government money alone, and pretending otherwise is the more dangerous story than the compute gap itself.

Start with what’s actually been built versus what’s been announced.

Under the IndiaAI Mission’s FutureSkills pillar, the government has approved 58 AI Centres of Excellence and 543 Data & AI Labs across ITIs and polytechnics nationally — a serious number on paper. But the rollout is uneven and running well behind the announcements. Gujarat has 37 institutes identified. Jharkhand has 11 approved, with only six even identified so far. One policy review of the programme put the core problem plainly: the labs component has outpaced its numerical target, while the fellowship programme is lagging badly, and the conversion rate from course registration to actual certification is low. Approving a lab and opening a working one are two different achievements, and right now the gap between them is the real story, not the headline count.

Now put that against the scale of the money actually flowing into Indian AI infrastructure.

The entire IndiaAI Mission — every pillar combined, over five years — runs on roughly ₹10,372 crore, a little over a billion dollars.

Compare that to what private capital has already committed: Reliance is scaling its Gujarat facility toward 2,000MW of capacity and separately expanding Jamnagar to 3GW at an estimated $20-30 billion. Adani has committed $100 billion to AI-ready data centres by 2035, with a stated ambition to build a $250 billion AI infrastructure ecosystem in India over the decade. Microsoft has committed $17.5 billion. AWS has committed $12.7 billion through 2030. Industry-wide, India is targeting as much as $200 billion in data centre investment over the next few years.

India’s sovereign AI lab Sarvam AI has become the first Indian company to onboard global chipmaker Nvidia as a strategic investor in a $75 million extension of its Series B round to scale Sarvam’s compute needs, as confirmed by cofounder Pratyush Kumar to ET. AI-focused venture fund Activate has also announced a strategic investment in Sarvam.

Read those two numbers side by side and the strategic picture becomes obvious:

The government’s role was never going to be building India’s compute capacity directly. It’s closer to a billion dollars of seed capital and policy scaffolding sitting underneath a private capital wave worth fifty to a hundred times more. The 2026-27 Union Budget’s tax holiday for data centres through 2047 is the clearest signal that this is the model India has already chosen, whether or not it’s been said out loud.

The real strategic question for policymakers isn’t “can government fund enough compute” — it’s whether government coordination, subsidised GPU access, land and power approvals, and regulatory clarity can move fast enough to actually direct where all that private capital lands, and on what terms.

But even the best-funded compute buildout runs into the same wall if the next piece isn’t fixed: there simply aren’t enough people who can use it.

The numbers here are more alarming than the infrastructure story.

TeamLease Digital’s most recent skills report found that for every ten open GenAI roles in India today, only one qualified engineer is available — and projected the overall AI talent gap to reach 53% by 2026.

ManpowerGroup’s 2026 survey found 82% of Indian employers reporting difficulty filling roles, well above the 72% global average, with AI model development and AI literacy now the two hardest skills to find in the country — ahead of every traditional IT skill.

The Deloitte-NASSCOM numbers, the most widely cited, project AI talent demand growing from roughly 600,000-650,000 to more than 1.25 million professionals between 2022 and 2027, against a current workforce of only about 416,000 — covering barely half of what industry already needs, before the newer compute infrastructure even comes fully online.

One quote from that research is worth sitting with, because it’s a sharper diagnosis than “not enough people.” Anand Mahurkar, founder of Findability Sciences, put the actual problem this way: it isn’t a talent shortage so much as a gap between academic training and corporate application — graduates who understand machine learning theory and can code in Python, but haven’t built anything real, and don’t yet combine technical depth with business understanding. That’s precisely the gap those ITI and polytechnic Data & AI Labs are supposed to close. Whether they do depends entirely on whether the labs that get approved actually open, and whether the training inside them produces people who can do the job, not just people who’ve completed a course.

Put together, this is the fuller version of the argument:

India’s AI infrastructure story isn’t really a compute story. It’s an execution story, playing out on three fronts at once — private capital that’s already committed and dwarfs public spending by orders of magnitude, a skilling pipeline that’s been announced faster than it’s been built, and a talent gap that no amount of GPU capacity will fix on its own.

Get the coordination right across all three, and the compute gap with the US stops being the headline. Get it wrong, and it won’t matter how many exaflops either country has.

The compute gap between India and the US is real, and it isn’t closing this year. But the more interesting story — the one worth your attention as a business leader — isn’t the gap. It’s what gets built on top of the compute that already exists, who’s allowed to build it, and whether India can move fast enough on capital coordination and talent development to make its own investment count.

— Debiprasad.

Sources:

Param Pragya (IIT Delhi):

  • PMO/PIB coverage of the August 8, 2026 inauguration at IIT Delhi’s Sonipat campus, during the institute’s 57th Convocation Ceremony

Argonne National Laboratory — Solstice and Equinox:

Oak Ridge National Laboratory — Lux, Discovery, and “sovereign AI”:

IndiaAI Mission — Data & AI Labs rollout:

Private AI infrastructure investment:

India’s AI talent gap:

#AICompass #AIComputeGap #AIInvestGap #AITalentGap #AIExecutionGap #AIStrategy

Google search engine

RELATED ARTICLESMORE FROM AUTHOR