India's AI problem isn't ambition, it's absorption

This article is authored by Kushal Sachdeva, author and founder, Protect Paws.

The computing power is arriving but spending, grid, and entry-level job market are not keeping pace.

AI

Two numbers from this year's Union Budget say more about Artificial Intelligence (AI) and the Indian economy than any summit communiqué. The IndiaAI Mission was allocated ₹1,000 crore for Financial Year (FY) 2026–27, half of the previous year's ₹2,000 crore. The cut was not a loss of faith. It was arithmetic: of the earlier allocation, only about ₹800 crore was actually spent.

Private capital, meanwhile, is moving at an entirely different velocity. India's data-centre development pipeline has swelled to roughly $90 billion, nearly six times the $13–15 billion invested between 2020 and 2024.

The gap between what the State has promised, what it has managed to disburse, and what the market is building anyway is the most important fact about India's AI economy right now. It is also the fact least discussed.

The IndiaAI Mission, approved in March 2024 with a ₹10,372 crore outlay, has delivered on its most visible pillar. More than 38,000 GPUs have been onboarded into a common compute facility and offered to startups, researchers and academia at rates well below commercial cloud pricing. One hundred and ninety projects have been approved. AIKosh, the national dataset repository, now hosts thousands of datasets. Sarvam AI released open-weight Indian foundation models in February. That same month, India hosted the AI Impact Summit in New Delhi, the first global AI summit convened in the Global South, and announced 20,000 additional GPUs.

Set against the EU's risk-tiered AI Act, this looks like under-regulation. Set against the alternative, a comprehensive statute that stalls in drafting for three years while the technology moves, it looks like realism. The honest position is that we do not yet know, because a voluntary framework has not been tested by its first serious harm.

The transmission channel that matters most is the one policy addresses least.

NASSCOM's Annual Strategic Review put the technology sector at $315 billion in FY26, growing 6.1%. Headcount grew 2.3%. That divergence is the story. Revenue and employment, historically locked together in Indian IT, have come apart.

AI services themselves remain small, $10–12 billion, under 4% of sector revenue. But the productivity effect is not waiting for the revenue to arrive. Freshers now make up a small fraction of active tech openings. TCS shed more than 23,000 roles in FY26 while moving to an AI-first delivery model. India's graduate unemployment rate sits above 11%, more than three times the overall rate, in a country where 8–10 million people enter the workforce every year.

Indian IT's model was to hire graduates cheaply, train them on the job, and move them up. AI compresses precisely the rung that made the model work: routine, rules-based, supervised output. This is not mass unemployment. It is something harder to legislate for, the quiet disappearance of the first job.

The other thing policy underestimates is that AI is a power problem wearing a software costume. India's installed data-centre capacity is projected to rise from about 1.5 GW in 2025 to 6.5 GW by 2030, with sector electricity demand reaching an estimated 13.5 GW by 2031–32. Reliable, affordable power, not land, not capital, has become the binding constraint on where facilities get built. More than half of India's data centres already sit in water-stressed regions.

The chips, too, remain imported. The India Semiconductor Mission has approved roughly ₹1.65 lakh crore across projects; three facilities are operational, and Tata's Dholera fab targets first silicon at 28nm by the end of this year. That is genuine progress toward the chips India's electronics industry actually needs. It is not progress toward AI accelerators, which require leading-edge logic India will not possess this decade. Sovereign compute, for now, means sovereign access to someone else's silicon.

NITI Aayog estimates AI could add $500–600 billion to Gross Domestic Product (GDP) by 2030; a recent IBM–IndiaAI study lands in the same range. Both deserve to be read as a ceiling rather than a forecast. The same IBM study found that 72% of surveyed Indian organisations consider themselves behind global peers on adoption, and that only about 30% of employees have the AI literacy their employers require, against a need closer to 57% by 2030.

That is the real policy agenda, and it is unglamorous. Diffusion into Micro, Small and Medium Enterprises (MSMEs), which employ the bulk of the non-farm workforce and have no procurement budget for AI. Grid planning and transmission is treated as AI infrastructure rather than as a separate ministry's problem. Public procurement used deliberately to create demand for domestic AI, an approach the current strategy has begun to recognise. And a serious instrument for the entry-level labour market, which so far has produced committees.

India has done the demonstrative part well. The computing power is arriving, the guidelines are written, the summit was hosted. The harder work is the boring work: spending what has been budgeted, powering what has been announced, and giving the 22-year-old graduate a rung to stand on. AI's contribution to Indian GDP will be settled there, not at the frontier.

(The views expressed are personal)

This article is authored by Kushal Sachdeva, author and founder, Protect Paws.