By – Ansh Goel

Abstract

Generative artificial intelligence is frequently seen as a weightless, digital phenomenon, yet its infrastructure relies heavily on electricity, land, and water. This article looks at how India’s environmental laws do not consider this fact, since hyperscale data centres are currently treated as regular commercial buildings under Item 8 of the 2006 Environmental Impact Assessment Notification (EIA), and the assessments carried out only cover the impact of construction and not the ongoing demand during operation. The article points out three specific regulatory gaps: firstly, there is no separate category for data centres in the EIA Schedule; secondly, there are no Water Usage Effectiveness standards under the Water (Prevention and Control of Pollution) Act, 1974; and lastly, data centres are excluded from Designated Consumer status under the Energy Conservation Act, 2001. It also demonstrates how the state-level policies in Maharashtra and Tamil Nadu widen this gap by providing fiscal incentives without attaching any corresponding environmental conditions. By referring to the comparative frameworks set out in the European Union’s Energy Efficiency Directive and AI Act, and in the forthcoming US Artificial Intelligence Environmental Impacts Act, the paper suggests three statutory reforms within India’s present legal framework: a separate EIA category for hyperscale facilities, mandatory water-efficiency thresholds, and the inclusion of data centres as energy-intensive consumers under the BEE.

Introduction: The Material Reality of the “Intangible”

The metaphor of the “cloud” creates misleading legal fiction. Generative AI is a physical, extractive industry that relies on extensive electricity grids, large land use, and billions of liters of water for cooling. India’s data center sector used about 150 billion liters of water in 2025, and this amount is expected to rise to 358 billion liters over the next five years. 

However, the  EIA Notification, 2006, India’s main environmental control system, still treats data centers as regular commercial buildings. They fall under Item 8(a) or 8(b) of the Schedule, with assessments mainly based on the built-up area and the impacts of construction. This approach overlooks the ongoing environmental impact of large-scale AI setups, such as constant grid demand, ongoing water use, and waste heat from facilities running at 20 to 50 kW per rack, which is five to ten times the density of typical commercial buildings. The Draft National Data Centre Policy 2025 (DNCP) compounds this issue by offering generous financial incentives without enforcing strict water or energy-efficiency rules. Deregulation is being presented as industrial policy.

Part I: India’s Regulatory Vacuum

The absence of regulation of the ecological impact of AI infrastructure stems from deficiencies in three areas of law. As stated in Clause 2 of the EIA Notification, 2006, only those projects which are included in the Schedule require prior environmental clearance. Since data centres are included in Item 8 (a) as ‘Building and Construction projects’, the assessment is based on the built-up area with a threshold of 20,000 sqm. As a consequence, environmental assessments focus on construction impacts, such as dust, noise, and waste, rather than on the ongoing consumption of electricity and water and the generation of heat. It therefore follows that a 100 MW hyperscale facility is treated in the same way as a commercial building.

The Maharashtra IT/ITeS Policy 2023 and the Tamil Nadu Data Centre Policy 2021 classify data center parks as essential infrastructure, granting electricity tax exemptions coupled with single-window clearances under respective state Industrial Facilitation Acts. Neither policy imposes statutory Water Usage Effectiveness (WUE) or Power Usage Effectiveness (PUE) standards as preconditions for clearance. This creates a regulatory race for investment in which states compete by relaxing, rather than strengthening, environmental requirements.

The Water (Prevention and Control of Pollution) Act, 1974 gives State Pollution Control Boards authority under Sections 25 & 26 to regulate industries that discharge sewage or trade effluent. While data center cooling systems fall under this regulation, SPCBs have not set WUE standards or caps on water withdrawal, and there is no central directive that mandates such actions. Similarly, Section 14 of the Energy Conservation Act, 2001 allows the Bureau of Energy Efficiency to label energy-intensive industries as “Designated Consumers.” However, data centers are excluded from this designation, leaving one of India’s fastest-growing electricity consumers without mandatory energy efficiency standards.

Part II: Comparative Frameworks

Two jurisdictions have taken steps to connect computation with ecological responsibility, and their frameworks provide useful models.

The European Union has included environmental responsibilities in AI governance through two binding laws. According to the EU AI Act (Regulation (EU) 2024/1689), specifically Article 53 and Annex XI, the providers of AI systems must keep technical documentation. The technical documentation should contain an estimate of the compute resources used, measured in FLOPS, among other details, and the energy consumed for all training. Providers must share this documentation with the AI Office when requested. This requirement is not about public disclosure, but it does create a necessary internal accounting rule that supports regulatory oversight.

More applicable is Directive (EU) 2023/1791 (Energy Efficiency Directive). Article 12 mandates that all data centres with an installed IT power demand of at least 500 kW submit annual public reports on energy consumption, Power Usage Effectiveness, Water Usage Effectiveness, Energy Reuse Factor, and Renewable Energy Factor. These reports go to a centralised European database. This requirement is binding, can be audited, and has been operational since September 2024.

The situation in the US still continues to develop. The Artificial Intelligence Environmental Impacts Act of 2026 (S.4727), reintroduced on June 9, 2026, by Senator Markey, has not been passed and remains in committee. It suggests that the EPA carry out a life-cycle study of the environmental effects of AI infrastructure. It also calls for NIST to gather a group to create measurement standards for a voluntary reporting system. Its importance for India does not lie in its current lack of binding force, but in its clear acknowledgment that AI data centres are a unique industrial category that needs specific environmental regulations. The bill defines an “AI data centre” as any facility with a peak power load over 50 MW, which serves as a practical benchmark for India’s proposed Schedule amendment. Yet these Western frameworks are far from a solution, the US relies heavily on voluntary reporting, while the EU’s mandates focus on disclosure rather than enforceable resource caps. They should serve as structural starting points for India rather than a glorified standard

Part III: Proposed Statutory Reforms

Three specific changes within India’s current laws can close the regulatory gap without needing new primary legislation.

First, the Central Government should modify the Schedule to the EIA Notification, 2006 by adding a new category for “Hyperscale Data Centres and AI Computational Infrastructure.” This category should include facilities with installed IT power capacity over 50 MW or total rated power capacity above 100 MW. These facilities should be classified as Category ‘A’ projects, requiring prior environmental clearance from the MoEFCC instead of state authorities. The environmental review must go beyond construction impacts to consider operational electricity demand, annual water withdrawal, groundwater effects in water-stressed districts, and cumulative heat island impacts. The 50 MW limit is similar to the US proposal, but excludes smaller facilities.

Second, the Central Pollution Control Board should be mandated to formulate obligatory directives under Section 18, directing State Pollution Control Boards to consent to establish according to Section 25, dependent on the applicant proving a Water Usage Efficiency (WUE) of less than 1.5 liters per kWh for all newly installed data centers located in the areas categorized as water-stressed by the Central Ground Water Authority’s assessments. Consent to operate under Section 26 should not be granted before WUE reporting is done annually by an accredited third party.

Third, the Bureau of Energy Efficiency must exercise powers under Section 14 to include hyperscale data centres in the list of Designated Consumers, subjecting them to compulsory annual energy audits, Power Usage Effectiveness (PUE) reports, and energy use requirements. New data centres should also have to adhere to a maximum PUE of 1.5, in accordance with the European Union’s 2030 trajectory. This framework would mirror the EU’s Energy Efficiency Directive while requiring neither new primary legislation nor international treaty obligations.

Conclusion

India’s data center capacity will reach nearly 4.8 GW by 2030 based on current trends. This growth will demand power and water in amounts exceeding those of most industries currently regulated under the EIA Notification, 2006. The separation of digital infrastructure from environmental responsibility is not a simple mistake; it is a conscious policy choice. This choice is part of MeitY’s efforts to reduce regulations and the competition among states for hyperscaler investments. In a regulatory race to the bottom, governments are subsidizing ecological strain: the DNCP 2025 proposes electricity-duty exemptions coupled with extended tax holidays, while the Maharashtra IT/ITeS Policy 2023 provides 100% electricity and stamp-duty exemptions. Environmental deregulation thus becomes a state-sponsored incentive, effectively subsidizing corporate resource depletion. However, this comes with a constitutional cost. The right to live in a pollution-free environment under Article 21 does not differentiate between emissions from a steel factory and water depletion from a server farm. Changing the EIA Schedule, activating the Water Act’s consent framework, and classifying data centers as Designated Consumers under the Energy Conservation Act are not difficult actions. These steps only need political will to enforce existing laws in light of a new industrial reality before infrastructure decisions limit future options.

Author Bio

Ansh Goel is a II year Law student at O.P. Jindal Global University and a columnist at Nickeled & Dimed. His research interests span across environmental law, the regulation of artificial intelligence, and intellectual property rights.

Image source: https://www.janushenderson.com/en-se/investor/article/power-compute-and-capital-inside-the-ai-infrastructure-supercycle/