The two-gigawatt tenant: data centres are becoming India’s anchor load, and its most demanding clean-energy buyers

Every phase of Indian grid expansion has had an anchor load that pulled investment behind it — railway electrification, aluminium smelters, agricultural pumping, then urban cooling. The 2020s have produced a new one. India’s installed data-centre capacity stands at roughly 2 GW as of 2026, with around 500 MW of fresh supply expected in 2026 alone — growth of about 30% year on year, per CBRE. Against a system peak north of 250 GW, two gigawatts sounds marginal. It is not, for three reasons: the load is extraordinarily concentrated, it is growing faster than any other demand category, and it belongs to the most creditworthy electricity buyers ever to enter the Indian market.

The capital commitments behind the buildout removed any doubt about durability. In late 2025, Microsoft announced roughly $17.5 billion for Indian cloud and AI infrastructure, with Amazon and Google each committing in the region of $15 billion. Industry trackers suggest total investment commitments into the Indian data-centre market could cross $180 billion through 2026. Hyperscalers do not announce capacity they are unsure of filling — these figures are demand forecasts wearing the clothes of press releases.

The forecast range is wide, and the width is itself the finding. Base-case projections put India at 4–5 GW of data-centre capacity by 2030. AI-accelerated scenarios — in which training and inference workloads localise faster than expected — run to 8–9 GW. On the energy side, data-centre power demand is projected to reach around 57 TWh by 2030, roughly five times today’s consumption and comparable to the entire annual demand of a mid-sized Indian state.

Two properties of that demand deserve more attention than the headline. First, load factor: a data centre runs at 80–90% utilisation of its connected load, against perhaps 50–60% for typical heavy industry. A gigawatt of data-centre capacity therefore consumes energy like two gigawatts of conventional industrial connection. Second, the uncertainty band — the gap between 4–5 GW and 8–9 GW — is a planning problem in its own right. Whether the AI-accelerated case materialises will be decided in boardrooms in Seattle and Mountain View, not in the offices of Indian transmission planners, who must nonetheless build for a range that spans a factor of two.

What makes data-centre demand analytically interesting is not its size but its geometry. The load is flat — a near-constant band across all 96 fifteen-minute blocks of the day, every day of the year. It is uptime-critical: availability commitments of 99.99% and above mean the load cannot be curtailed, shifted, or interrupted in response to price. And it arrives in a market whose marginal supply increasingly has the opposite shape — solar, which delivers a midday hump and nothing after sundown.

India’s intra-day price curve now reflects that mismatch in published clearing prices: near-zero midday blocks under surplus solar, and evening blocks that run into the ₹10 per unit ceiling as the sun sets into peak demand. For a flat-load buyer, this shape cuts both ways. Across the midday belly, a data centre is the ideal customer — a guaranteed, price-insensitive sink for the cheapest utility-scale power India has ever produced. Across the evening neck, the same buyer is structurally short in precisely the blocks where the market is tightest — and, unlike a factory, it cannot respond by shifting a shift. A flat load buys the whole curve, every day, forever. That makes data centres simultaneously the best-positioned and the most exposed large buyers in the Indian market.

The standard corporate instrument — a flat-price PPA with a single solar plant — barely dents the problem. A solar asset generates for perhaps a quarter of the hours a data centre consumes. The remaining hours are grid or exchange exposure concentrated in the expensive end of the curve, which is exactly the exposure the PPA was signed to eliminate.

Hyperscalers are not annual-matching buyers anymore. The procurement standard the largest of them have adopted — 24×7 carbon-free energy — requires matching consumption with clean generation in every hour, not netting it out across a year. That standard is arriving in India with real weight behind it: in 2025, technology companies signed roughly 40% of all renewable PPAs globally. And Indian supply is, at the annual level, abundant — the country added a record 44.5 GW of renewable capacity in 2025. The megawatt-hours exist. The shape does not.

Getting from a single solar PPA — perhaps 20-something percent hourly matching — toward genuine round-the-clock coverage forces a portfolio: solar for the day, wind for its evening and monsoon-season complementarity, and battery storage to move midday surplus into the neck of the curve. India’s round-the-clock RE tenders have already demonstrated the template, with blended wind–solar–storage portfolios bidding high-availability supply at tariffs that would have seemed implausible five years ago. For a data-centre operator, the procurement question stops being “which plant?” and becomes “which combination of assets, in which states, behind which contractual structure?”

Structure and geography then do as much work as generation. Open-access supply and captive or group-captive ownership carry materially different charge stacks — captive structures can escape cross-subsidy and additional surcharges entirely, at the price of equity participation in the generating asset. And state selection is driven by regulation as much as by fibre: banking provisions, open-access charge trajectories, and connectivity timelines vary so widely between states that the same portfolio can be viable in one and uneconomic across the border. Submarine cable landings decided where India’s first data-centre clusters sit; the electricity rulebook will have a growing say in where the next ones do.

The global data-centre story of the past three years has been that interconnection, not capital, gates growth — grid-connection queues in Northern Virginia, connection moratoria around Dublin, Singapore’s pause on new builds. India’s version of the constraint is arriving on schedule, and it binds at both ends of the wire: grid connectivity for a 100–300 MW campus in a metro whose substations were never sized for it, and evacuation capacity for the wind–solar–storage portfolio in Rajasthan or Karnataka that is meant to serve it. A data-centre shell can be built in eighteen to twenty-four months; transmission bays, substation augmentation, and interstate corridor capacity move on multi-year timelines that no purchase order accelerates. The buyers who treat grid access as a long-lead procurement item — secured before land, alongside power — will set the pace of the buildout. The rest will own generation they cannot deliver to load they cannot energise.

For a buyer whose load runs flat through a market this shaped, the difference between a good portfolio and an expensive one is invisible at annual resolution. Before committing to twenty-year contracts, a data-centre operator should be able to answer, quantitatively:

• Hourly versus annual matching: what does the portfolio’s carbon-free match look like block by block, not netted across a year? A 100% annually-matched portfolio can leave a third of actual consumption running on the grid’s evening mix — 15-minute matching is the only resolution at which a 24×7 claim can be verified or priced
• Banking rules as they will be, not as they are: banking is the implicit storage that makes many open-access economics work, and states have been steadily tightening it — from annual to monthly settlement, with haircuts. A portfolio underwritten on today’s banking provisions carries silent regulatory duration risk
• DSM exposure: deviation settlement charges accrue when scheduled and actual generation diverge — and wind and solar diverge daily. Someone in the contract chain bears forecast error at settlement prices linked to an increasingly volatile real-time market; the buyer should know who, and what it costs in a bad month
• Tariff and charge trajectories: open-access surcharges, transmission charges, and the retail tariffs that define the alternative are all moving. The economics of a captive structure versus open access versus grid supply should be stress-tested against charge trajectories, not point estimates
• Storage sizing against the shape: the battery is the portfolio component that converts belly-priced energy into neck-hour coverage. Undersize it and evening exposure persists; oversize it and the portfolio carries idle capital. The right size falls out of block-level simulation, not rules of thumb

Each of these is a 15-minute-resolution question. None of them can be answered with the annual spreadsheets that have historically underwritten Indian corporate PPAs — and getting them wrong is not a rounding error but a two-decade contractual commitment signed against the wrong curve.

This is the standard EarthSync’s simulation engine was built for: modelling wind–solar–storage portfolios against a data centre’s flat load at 15-minute resolution — hourly carbon-free matching, banking and DSM exposure, state-by-state charge stacks, and tariff trajectories — so that a 24×7 portfolio is designed and stress-tested before the first PPA is signed, not discovered in the settlement statements after. India’s most demanding electricity buyers are about to become its largest. The market that learns to serve a flat, carbon-free load profitably will have solved, along the way, most of what the duck curve broke.