AI Data Centres Are Rewriting India’s Energy Playbook: Why Governance Has Not Kept Pace

Infrastructure and Environmental Stress Points: Grid Pressure, Coal Lock-In, and Water Risk

AI services that appear intangible to consumers are anchored in highly local, resource-intensive infrastructure. Data centres must maintain stable voltage and frequency at all times, because even brief disruptions can cascade into system failures. Three recurring physical constraints define the challenge.

First, grid stress manifests locally before it becomes national. Data centres operate at continuous high load and disproportionately strain local distribution and sub-transmission infrastructure—feeders, substations, and city-edge transmission—long before aggregate national generation capacity becomes binding. This leads to familiar but costly outcomes: delayed augmentation, the need for dedicated lines, constrained site selection, and a rising premium on “power-ready” land.

Second, there is the firm-power dilemma. Wind and solar, while central to India’s energy transition, are intermittent. In the absence of storage, hydro balancing, nuclear power, or structured firm-power products, systems default to coal or gas to maintain reliability. This preserves uptime but shifts governance from a decarbonisation problem to a reliability trade-off, exposing AI infrastructure to future carbon-cost and compliance shocks. The policy challenge is therefore not simply capacity addition, but reliability planning.

Third, water intensity introduces a political constraint. AI-heavy compute significantly increases cooling demand. In water-stressed metropolitan and peri-urban regions, data-centre cooling competes directly with households and industry, turning infrastructure approvals into distributive politics. Without enforceable norms—treated-wastewater cooling, heat reuse, strict water accounting—data-centre growth risks colliding with municipal priorities and social legitimacy.

Energy Policy and Governance Structure: Fragmentation as a Structural Weakness

India’s electricity governance architecture is fragmented by design. Generation, transmission, and distribution operate under different institutional and regulatory frameworks. Planning authority is split between the Centre and the States, while regulatory decisions are divided between the Central Electricity Regulatory Commission and State Electricity Regulatory Commissions. This model has supported large-scale capacity addition, but it performs poorly when the policy objective shifts from “more electricity” to high-quality electricity—characterised by low downtime, stable voltage and frequency, and predictable long-term contracting for very large, continuous loads such as AI data centres.

A second fragmentation is sectoral. Data centres are framed as “digital infrastructure” under the Ministry of Electronics and Information Technology and state IT or industry departments. Their binding constraints—grid capacity, substation augmentation, transmission corridors, cooling water, and environmental approvals—sit elsewhere. Without an integrated planning lens, site selection can be misaligned with grid headroom and long-run water security, pushing the system toward the easiest firming option: coal.

This structural mismatch explains why the nuclear-AI linkage has re-entered policy debate. Whether or not India ultimately adopts nuclear power as a primary solution, the governance point remains unchanged. AI infrastructure requires firm, dispatchable power and contract structures that reward reliability and low-carbon intensity, not merely the lowest short-run energy cost.

Policy and Institutional Gaps

Despite the scale of expansion, India still lacks a fully operational, nationally enforced data-centre sustainability regime covering energy efficiency, water use, and emissions reporting. Mandatory standards that apply across the sector remain absent. The result is uneven outcomes: large operators voluntarily adopt international benchmarks, while inter-jurisdictional competition risks a race-to-the-bottom driven by incentives rather than sustainability.

Institutional capacity is also lacking. Many State utilities and regulators are not yet equipped to model large digital loads alongside future grid complexity involving storage, smart grids, demand response, and algorithmic load management. Treating data governance and energy governance as separate silos is increasingly untenable when AI infrastructure is, in practice, an electricity-and-water project that merely presents itself as software.

The Catch-Up Question: Global Competition and India’s Strategic Risk

Internationally, the AI infrastructure race is being reframed as a firm power plus industrial policy problem. Leading technology firms and jurisdictions are exploring nuclear energy, including SMRs, as stable low-carbon complements to renewables, and governments are adjusting grid and transmission planning accordingly.

India’s own policy discourse has begun to reflect this reality. Discussions around a Nuclear Energy Mission and long-horizon capacity expansion, including SMRs, indicate growing awareness that 24×7 clean power will be a strategic enabler for AI-era infrastructure. The competitiveness implication is straightforward: if India cannot guarantee reliable, reasonably priced, low-carbon electricity for large compute clusters, capital will gravitate toward jurisdictions that can.

India’s Build-Out: Who Is Building, Where, and at What Scale

India’s data-centre expansion is now large enough to be treated as an electricity-planning problem rather than a niche technology trend. A 2025 study by the Takshashila Institution estimates installed data-centre capacity at approximately 1.2 GW as of April 2025, with projections scaling materially by the end of the decade.

The investment pattern is overwhelmingly private-led, combining domestic infrastructure groups and hyperscaler-anchored campuses. State agencies primarily play enabling roles through land aggregation, utilities coordination, and permitting. A Reuters-reported illustration of scale underscores the shift: Adani, through AdaniConneX, is reported to be exploring investments of up to $5 billion linked to a 1 GW Google AI data-centre project in Andhra Pradesh, aligned with Google’s reported $15 billion multi-year investment plan for that campus.

In plain terms, India is no longer adding incremental server rooms. It is entering a phase where single projects approach power-station-scale loads. This demands governance mechanisms to allocate grid-upgrade costs fairly, secure firm low-carbon supply at scale, and prevent water stress from becoming the silent stopper of AI ambitions.

Global Comparison: Powering AI Data Centres — By the Numbers

Across major economies, AI data centres are now treated as power-station-scale loads, reshaping electricity planning. The International Energy Agency estimates that global data-centre electricity consumption is growing at ~12% annuallyand could reach ~945 terawatt-hours (TWh) by 2030, up from roughly 460–500 TWh in 2022, placing sustained pressure on grids and firm power supply.

In the United States, single AI campuses increasingly demand 300–1,000 megawatts (MW) each, equivalent to a mid-size thermal plant. Hyperscalers are signing 20–30-year power contracts, combining renewables, storage, and nuclear. Multiple US utilities are assessing SMRs in the 300 MW range specifically for data-centre anchoring, as grid congestion and interconnection queues now exceed 2,000 GW nationally.

The European Union has slowed expansion through regulation. Data centres above 500 kW face mandatory energy reporting, while several member states link approvals to renewable matching and water-use caps. This has reduced fossil lock-in but raised effective costs by 10–20% in some markets.

China integrates data centres with generation planning. Large clusters are co-located with hydro, coal, and nuclear capacity, enabling rapid scale. China added over 50 GW of nuclear capacity by 2024 and is piloting SMRs to support energy-intensive digital infrastructure, though coal still provides reliability backstop.

Japan and South Korea prioritise uptime, tolerating higher power costs to maintain redundancy. Nuclear power continues to supply ~20–30% of baseload electricity critical for high-reliability digital loads.

India’s position is stark by comparison. Installed data-centre capacity is estimated at ~1.2 GW (April 2025), but individual projects now approach 500 MW–1 GW scale. Coal still supplies the majority of firm power despite ~235.7 GW of non-fossil capacity. Without integrated planning, India risks higher carbon intensity, grid stress, and water conflict as AI infrastructure expands.

Short Run and Long Run Policy Options Snapshot

In the short term (1–2 years), classifying data centres as strategic energy consumers and mandating energy and water reporting for facilities above 10–20 MW can immediately improve grid visibility.

In the medium term (3–5 years), standardised firm-power products combining renewables with 4–8 hours of storage, hydro firming, or nuclear-linked supply can reduce coal reliance for the projected 2.5–3 GW data-centre load by 2027.

In the long term (5–10 years), integrating nuclear power—including 300 MW-class SMRs—into digital-infrastructure planning and creating a unified AI–energy coordination framework will determine whether India can competitively support multi-GW AI clusters without locking itself into high-carbon, high-cost reliability solutions.

Toward an Integrated AI–Energy Policy Framework

Preserving India’s existing institutional structure while upgrading its ambition requires a shift to integrated planning rather than ad-hoc coordination.

The practical shifts to that end must include:

1. Treat data centres as strategic energy consumers in long-term power planning (not just “commercial consumers”), with explicit reliability metrics and locational grid-impact assessments built into approvals.
2.  Create a firm low-carbon procurement pathway that data-centre operators can actually use at scale—standardised products for storage-backed renewables, hydro firming, and (where politically chosen) nuclear-linked supply, rather than ad-hoc bilateral arrangements.
3. Mandate efficiency and resource accounting: minimum energy-efficiency and cooling standards, water-use reporting, and phased requirements to shift cooling toward treated wastewater where feasible.
4. Upgrade regulatory capacity inside state utilities and commissions so they can evaluate large digital loads, contract structures, and grid augmentation needs without turning every project into a bespoke negotiation.

If India succeeds, the payoff extends beyond AI itself. It would deliver a durable, low-carbon, high-reliability electricity backbone capable of supporting digital growth without allowing energy scarcity, coal dependence, or water conflict to become the binding constraint on India’s technological future.