India's AI infrastructure policy rests on capacity numbers that, according to the ministry responsible, nobody in government officially keeps. A parliamentary committee report on the Ministry of External Affairs' technology diplomacy, reported by MediaNama on October 14, 2026, records MeitY telling the committee: "Officially, no data is maintained in MeitY regarding the capacity of Data Centres".
What the committee found
- MeitY acknowledged that a figure often used in policy debate, that India produces about 20% of the world's data but holds only about 3% of its data-centre capacity, is not verified.
- The figures on the record are industry estimates: about 520 MW in 2020, about 1.5 GW in 2025, about 2 GW in 2026, and more than 8 GW by 2030.
- The committee also notes estimates that capacity may need to grow to "between 5 and 10 GW, by 2030". That range is itself wide enough to change investment decisions.
- It asked MEA to press MeitY to collect and publish verified data on capacity, utilisation and growth.
Meanwhile, the compute tender
The finding lands as the IndiaAI Mission buys more capacity. MediaNama reported on October 13 that C-DAC, the mission's implementing agency for compute, has issued a Request for Empanelment for providers of up to 10,000 AI Compute Units, covering GPUs, TPUs and other accelerators. The document allows growth beyond that: "C-DAC may, as per the requirement, scale up the total AI Compute capacity beyond 10,000 AICU". Units are offered in clusters of 4,000, 1,000 or 512. At least 512 must sit in a single Indian data centre, with capacity spread across several locations. All infrastructure, data, logs, backups and disaster recovery must stay in India, physically and logically isolated from other customers. The document does not say how much the first round will buy.
Those localisation clauses decide who can bid. A provider needs isolated capacity on Indian soil, which favours operators with data centres in the country over resellers. States including Andhra Pradesh, Telangana, Karnataka, Uttar Pradesh and Maharashtra are competing for exactly that kind of investment, without an official national baseline to measure it against.
What remains uncertain. How much of the 10,000 units the first round will actually procure, how prices will compare with earlier IndiaAI rounds, and whether MeitY will commit to publishing capacity data on a schedule. Reports that bidders may be allowed to sell surplus capacity overseas could not be verified for this entry.
Analysis: For anyone building AI products in India, the data gap matters more than the tender headline. Subsidised compute under IndiaAI is only useful if it is available where and when teams need it, and planners cannot judge scarcity without utilisation data. Startups relying on the mission's GPUs should plan for queues and keep a commercial fallback. Policymakers have an easy first step available: publish what is already measurable, starting with the capacity the government itself contracts and how much of it is used.
A parliamentary committee report reported on October 14 records MeitY admitting it keeps no official data on India's data-centre capacity, as C-DAC seeks providers for up to 10,000 AI compute units under IndiaAI with strict in-country isolation rules; the panel asked for verified capacity and utilisation data.