Live on XaptiqUGC NET Economics (Dec 2026) and IIT JAM Economics (EN) test series are now live on Xaptiq.View test series
BusinessEnergy & Digital InfrastructureIndian EconomyData CentresPower Sector

AI data centres to add 26.3 GW to India's power load by 2031-32, doubling March estimate

The Ministry of Power told Parliament that AI data centres will add 26.3 GW of load by 2031-32 — nearly double its March estimate of 13.56 GW — mostly served by renewables; the grid operator has flagged the spiky, inverter-based nature of these loads as a stability risk.

What happened

The Ministry of Power told Parliament that AI data centres will put an additional 26.3 GW of load on the Indian grid by 2031-32, almost double the government's earlier March projection of 13.56 GW. The Ministry said the extra load will be integrated into the grid and served primarily by renewable energy capacity. It is holding consultations with state transmission and distribution utilities and data centre developers on reliable power supply, transmission adequacy and grid balancing, because the load from data centres can be spiky and prone to sharp ramps.

Context

India's data centre stock has scaled rapidly — from just 375 MW in 2020 to around 1.5 GW by 2025. According to Wood Mackenzie, operational capacity is projected to reach 12 GW by 2030, at a compound annual growth rate of roughly 40%. AI-dedicated capacity alone is expected to expand nearly 24-fold, from 275 MW to 6,546 MW. The Ministry is also discussing whether new data centres should locate away from urban centres — which are already IT-cluster heavy — and closer to green power generation, so that transmission congestion is avoided and their load can be matched to renewables at source.

Why it matters

AI runs on algorithms but consumes electricity in industrial volumes, so where it plugs in matters as much as how much it draws. Data centre loads are not smoothly rising baseload: they are inverter-based, variable and prone to sudden ramps, and Grid India has warned that inverter-based loads can "quietly exit" the grid when they detect a disturbance. A sudden withdrawal of 1-1.5 GW — small next to the projected 26.3 GW — can create a frequency and voltage disturbance because the grid must match generation and demand second by second.

Impact for India

  • Ministry of Power and Grid India: Must plan generation, transmission and balancing reserves for a load type that is highly variable and can withdraw quickly.
  • Renewable energy developers: A large captive-style buyer emerges, but geography must line up — the Ministry is pushing for data centres to be sited near green energy.
  • State DISCOMs and transmission utilities: Face concentrated demand near IT hubs unless the siting policy shifts, and will need investment in balancing infrastructure.
  • Data centre developers and hyperscalers: Face a nudge to relocate away from urban centres and closer to renewable-rich zones, changing the economics of new sites.
  • Consumers: Grid instability risk from spiky loads is a system-wide externality that ultimately shows up in reliability and tariff planning.

Key data

AI data centre load projected by 2031-32 (Ministry of Power): 26.3 GWUp from 13.56 GW in the March estimate
India's data centre capacity: Around 1.5 GW by 2025Up from 375 MW in 2020
Wood Mackenzie: India's data centre capacity 2030: 12 GWCAGR of approximately 40%
AI-dedicated data centre capacity: Rising from 275 MW to 6,546 MWNearly 24-fold expansion
Data centre electricity demand growth by 2040: 20-fold from 10 TWh in 2025Would account for 7% of total electricity demand
Sudden withdrawal that can destabilise grid: 1-1.5 GWDescribed by Grid India's chairman

Concepts

Grid stabilityThe ability of the electricity grid to keep frequency and voltage within safe bands despite unplanned changes in generation or load, on time scales as short as a few seconds.Grid India warns that the variable, ramp-heavy character of AI data centre loads directly stresses this stability.
Inverter-based loadAn electrical load that connects to the grid through power electronics rather than through synchronous rotating machinery, and so can be switched in or out of the grid without the mechanical inertia that traditional loads provide.Modern data centres, along with much of the solar and wind fleet, are inverter-based, which is why their exit is faster and "quieter" than a conventional heavy load.
Load rampThe rate of change of electricity demand over time — high ramps demand fast dispatchable generation or storage to keep supply matched to load.AI compute loads shift with training and inference schedules, producing much steeper ramps than industrial or residential loads.
Transmission congestionA situation where the physical capacity of transmission lines is insufficient to move all the electricity that generators can produce to where consumers want it, often forcing curtailment of cheaper generation.The Ministry is worried about clustering of data centres near IT hubs precisely because it would concentrate load and create such congestion.

Stakeholders

Gains: Renewable energy developersNew long-duration industrial demand tied to renewable-heavy zones
Gains: Regions with green energy resourcesChance to attract data centre investment away from IT hubs
Pressure point: Grid stabilitySharp ramps and "silent exits" of 1-1.5 GW loads can destabilise frequency and voltage
Pressure point: Urban DISCOMsFace concentrated load if siting policy does not shift and if new capacity clusters near IT hubs

Practice Question — from the story

With reference to the recent Ministry of Power estimate on data centre electricity load in India, consider the following statements:

  1. The Ministry of Power's latest projection of the AI data centre load by 2031-32 is 13.56 GW, unchanged from its March estimate.
  2. India's operational data centre capacity as of 2025 is around 12 GW, roughly the same as the 2030 target reported by Wood Mackenzie.

Which of the statements given above is/are correct?

  1. 1 only
  2. 2 only
  3. Both 1 and 2
  4. Neither 1 nor 2
Reveal answer & explanation

Answer: Neither 1 nor 2

  • Statement 1: Incorrect. The latest projection is 26.3 GW by 2031-32, almost double the earlier March estimate of 13.56 GW.
  • Statement 2: Incorrect. India's current data centre capacity is around 1.5 GW by 2025, up from 375 MW in 2020; 12 GW is Wood Mackenzie's projection for 2030, not the 2025 stock.

Therefore, the correct answer is Neither 1 nor 2.

Concept check — test your understanding

Grid India's chairman warned that AI data centre loads are "variable, spiky, with sharp ramps" and can make "silent exits" from the grid, and that a sudden withdrawal of 1-1.5 GW of such inverter-based load can destabilise the grid. Which of the following best explains why?

  1. Inverter-based loads do not add rotating grid inertia, so an equivalent-size withdrawal by a conventional rotating load could have been absorbed while the data-centre exit cannot — the issue lies purely in the missing inertia, not the load size.
  2. Grid frequency and voltage are maintained by second-by-second matching of demand and supply, so a sharp unplanned drop of 1-1.5 GW of load leaves generation instantaneously in excess, causing frequency and voltage transients that can destabilise the grid.
  3. Data centres in India draw power directly from dedicated renewable plants under captive contracts, so their sudden withdrawal only briefly overloads the transmission lines feeding those specific plants without affecting the broader grid.
  4. Modern data centres run on independent islanded microgrids that are only nominally connected to the main grid, so their exit merely disconnects them from a shared frequency reference and does not affect other consumers in real time.
Reveal answer & explanation

Answer: B

  • Grid stability rests on continuous, instantaneous matching of generation with load. When 1-1.5 GW of demand exits without warning, the grid still has that much generation running: frequency and voltage jump, and the system operator must find room for the excess or force generators to back down quickly. Inverter-based loads do this fast and "silently", which is what Grid India was flagging.
  • Option A over-narrows the problem to inertia; missing inertia matters, but the primary destabilising event is the sudden generation-load imbalance itself.
  • Option C invents a captive-plant-only supply model — most Indian data centres draw from the wider grid, and even captive renewable plants are grid-tied.
  • Option D describes a microgrid architecture that Indian hyperscale data centres do not currently use; they are grid-connected loads whose sudden exit is felt system-wide.