·The Hindu·15 marks·250–350 words

"Compute is the new oil." Examine India's strategy to secure AI compute capacity through the IndiaAI Mission.

New · District CollectorCan you hold the district for 72 hours?Floods, a heatwave, an outbreak, a grievance queue. Take the decisions, see the consequences.Play →
In this answer
  1. Strategy: How India Is Securing Compute
  2. Gaps: Where the Oil Analogy Bites
  3. Way Forward

AI compute means GPUs and other chips that train and run AI models. Like oil in the 20th century, it is scarce, controlled by a few suppliers and strategic. The IMF argues that electricity and chip supply will decide which countries control AI [6]. India's IndiaAI Mission (outlay over ₹10,300 crore, 2024) [2] treats compute as public infrastructure, but so far it gives users access to compute more than it gives India ownership of it.

Strategy: How India Is Securing Compute

  • Shared national GPU pool: the original target of over 10,000 GPUs [2] has grown to 45,000+ GPUs (June 2026) [1].
  • Affordable access: approved private providers offer GPUs at about ₹65/hour [3], with a subsidy of up to 40% on the AI Compute Portal [2].
  • Creating demand: 237 projects had used 93.18 lakh GPU-hours by August 2026 [1]. 12 firms are building Indian foundation models [3].
  • Link to chipmaking: under the semiconductor push, 211 chips have been taped out by 75 institutions [4].

Gaps: Where the Oil Analogy Bites

  • Dependence on one supplier: in one count, 12,896 of 18,693 GPUs were Nvidia H100s [2]. Nvidia holds over 80% of AI GPUs, and TSMC makes most advanced chips [5]. This is a chokepoint, like oil supply.
  • Rented, not owned: the capacity stays with private providers once subsidies end. There is no state-owned "strategic reserve" of compute.
  • Capacity is not usage: GPU counts are published, but utilisation rates are not. So we cannot tell whether subsidised chips are busy or idle [1].
  • Energy and minerals: data centres need reliable power and inputs such as gallium [6].

Way Forward

  • MeitY should publish GPU-hours used against GPU-hours available every quarter, along with the number of models delivered.
  • Diversify tenders: give weight to non-Nvidia chips and chips designed in India [4][5].
  • Competition oversight: the CCI should study cloud and GPU contracts, using the enforcement, merger control and advocacy tools the OECD recommends [5].
  • Own core capacity: build a state-owned cluster for strategic uses, powered by renewable energy.

IndiaAI has moved India from scarce compute to cheap, widely available compute, much like securing oil supply contracts. Real security now needs owned capacity, more than one supplier and measured results, a "strategic reserve" for the AI age. That would advance Atmanirbhar Bharat and SDG 9 (resilient infrastructure).

Sources

  1. 1PIB Factsheet: IndiaAI Mission: 45,000+ GPUs (June 2026); 237 projects and 93.18 lakh GPU-hours (August 2026)
  2. 2PIB document on IndiaAI compute (August 2025): outlay of over ₹10,300 crore; original target of over 10,000 GPUs; subsidy of up to 40%; 12,896 H100s out of 18,693 GPUs
  3. 3PIB: India Accelerates AI Self-Reliance: 12 Companies Developing Foundation Models Using 38,000 GPUs at ₹65/Hour: ₹65/hour compute; 12 foundation-model firms
  4. 4PIB release on India's semiconductor design push (2026): 211 chips taped out by 75 institutions (April 2026)
  5. 5OECD, Competition in Artificial Intelligence Infrastructure (2025): Nvidia's GPU share and TSMC's dominance; enforcement, merger control and advocacy tools
  6. 6IMF Finance & Development, "Inside the AI-Led Resource Race" (December 2025): electricity and chips decide who controls AI; minerals such as gallium

More from this note