The “cloud” is a seductive misnomer. It evokes something weightless, borderless and almost free. Yet artificial intelligence reaches us through intensely physical systems: chips, server halls, cooling equipment, substations, transmission lines, backup power and water. India’s next digital-policy question is therefore not simply how much compute to build, but what each scarce megawatt should buy the country.
Brookfield Asset Management expects about 6.5 gigawatts of AI data-centre capacity to come online in India over the next five years. India currently has about 1.5 GW of installed capacity. Separately, the Services chapter of the Economic Survey expects total capacity to reach about 8 GW by 2030, from 1.4 GW in the second quarter of 2025. It also notes that India generates nearly 20 per cent of the world’s data but hosts only about 3 per cent of its data centres. The gap is a commercial opportunity. It is not, by itself, an argument for support at any cost.
The International Energy Agency expects global data-centre electricity use to rise from 485 terawatt-hours in 2025 to 950 TWh by 2030, with consumption by AI-focused facilities tripling. Policy has moved quickly. Union Budget 2026-27 proposed a tax holiday until 2047 for qualifying foreign companies providing global cloud services through Indian data centres. Services to Indian users must pass through an Indian reseller. The Budget also proposed a 15 per cent safe harbour for eligible related Indian providers.
The bargain appears attractive: tax relief for investment, exports and an AI foothold. But capital expenditure is not a welfare statistic. A data centre can anchor research, start-ups and domestic cloud capability. It can also become a digital enclave that processes global workloads without creating commensurate Indian knowledge, procurement or innovation linkages. Public policy must distinguish between the two.
There is an incidence problem as well. The Union offers the income-tax concession, while state utilities and local communities confront demands for land, grid connections, water and new substations. Benefits may travel; congestion and scarcity remain local.
The Economic Survey itself identifies power, finance and water as binding constraints and warns that indiscriminate compute expansion competes with households and other industries.
For an economist, the relevant number is not the average electricity tariff but the shadow price of reliable power at the project’s location. That includes incremental generation, transmission, balancing, backup and congestion costs. A renewable power contract does not erase these costs. An annual solar purchase may balance energy on paper while a server hall demands electricity through the night. The proper test is whether clean supply is additional, deliverable and firm, and who finances the necessary storage and network reinforcement.
India therefore needs an AI Compute Compact. Every large facility should disclose standardised power usage effectiveness and water usage effectiveness, peak demand, cooling technology, operational carbon intensity and the temporal source of its electricity. ISO already recognises PUE and WUE as data-centre performance indicators. Disclosure should be facility-specific because water and grid stress are inherently local.
Connection charges and time-varying tariffs should recover incremental system costs. Operators that provide battery support, verifiable demand response or other grid services should receive credit. Fiscal incentives should meanwhile be conditional and periodically reviewed against domestic value created per megawatt: Indian R&D, skilled employment, local procurement, taxable income, cloud exports and affordable compute supplied to Indian firms and universities.
Public support should also earn a compute dividend. A defined volume of compute credits could be made available competitively to universities, start-ups and public-interest projects in health, agriculture, education and disaster management. Society should receive verifiable capability, not merely promises of spillovers.
India need not choose between AI ambition and resource prudence. Prudence is what makes ambition credible. Countries that merely host machines rent out electricity, water and land. Countries that convert compute into firms, skills, research and public capability capture the larger surplus. India’s success should be measured not by gigawatts switched on, but by Indian value created per gigawatt.
Soumyatanu Mukherjee, Associate Professor of Economics, XLRI Delhi-NCR