समाचार Samāchār · Tech & AI · 3 November 2026

India drew $300 billion of data centre pledges. IndiaAI’s GPUs reportedly sit 80% idle

A Rhodium Group report argues India is building AI capacity faster than it can use it, mostly for global clouds. Its bright spot is services, where Indian engineers close the gap between a model and a working business.

The question on India’s AI build-out has moved from how much to who for. Rhodium Group, a US research firm, makes that case in a report by Ankit Tiwari and Reva Goujon, dated 27 October and reported in India on Monday. Its argument: India is building data centres faster than domestic demand can absorb them, and the capacity will mostly serve global cloud providers. This is one firm’s reading. The numbers deserve testing either way.

Commitments passed $300 billion over the past year. Operational capacity is expected to at least triple within three years. Hyperscalers lease roughly 60% of it today. Enterprise and edge centres used by Indian entities are under 13%. Axis Capital estimates hyperscalers will lease over 90% of additions by 2030. The $1 billion IndiaAI Mission reportedly leaves nearly 80% of its GPUs unused despite below-market prices. Yotta’s CEO estimates nearly 80% of its revenue comes from foreign clients.

Power and jobs are the harder tests. India curtailed over 8,100 GWh of solar in Q2 2026, or 11%, because the grid could not carry it. On Google’s 1 GW project, Andhra Pradesh’s IT minister cites 2 lakh direct and indirect jobs. The state’s own industrial agency estimates 200 direct jobs and about 5,000 construction workers.

Credit where due. Sarvam has built a 105-billion-parameter model across 22 Indian languages, on FY26 net revenue under $5 million. The report’s point is that the home market for Indic products alone will not fund the frontier. Services look stronger. AI brings the sector an estimated $12 billion a year, 7.5% of revenue at TCS and 8% at Infosys. An entry-level forward-deployed engineer in India earns under a tenth of the US base salary.

What a founder takes from it: Compute without users is a building. India needs sovereign models and compute treated like UPI: public infrastructure that people actually run on, with the grid to carry it. Founders can take the near-term money where the report finds it: in making AI work inside a real business.