Why This Matters Now
The conversation around artificial intelligence is dominated by what it can do, write, predict, automate, and too little by what it needs. Every large model runs on data centres, and those centres are physical: they draw electricity, drink water for cooling, occupy land and produce hazardous electronic waste. As India races to build data-centre capacity, much of it in states already short of power and water, the environmental footprint of AI has moved from an abstraction to a planning problem that cannot be deferred.
The Crux in 60 Words
AI is not weightless. Data centres consume power, water and land, and generate e-waste, and India is building them fastest in its most stressed cities. A 100 MW facility can use about 20 lakh litres of water daily. Without mandatory disclosure, energy and water planning, recycled-water cooling and electronics recycling, India’s AI ambition will collide with the ground it occupies.
The Issue, Decoded
| Concept | What it means | Why it matters |
|---|---|---|
| Hyperscale data centre | A very large facility hosting AI and cloud workloads | High concentrated demand for power, water and land in one site |
| Water for cooling | Fresh water evaporated or circulated to cool servers | A 100 MW centre can use ~20 lakh litres a day, straining local supply |
| Grid emission factor | Carbon intensity of the electricity consumed | On a coal-heavy grid, AI’s power draw is carbon-intensive |
| GPU obsolescence | AI accelerators aging out in 2 to 3 years | Rapid, hazardous e-waste India cannot yet recycle safely |
| Environmental disclosure | Reporting of actual resource use | Absent it, regulators cannot plan or hold operators to account |
The Analysis
- Water is the sharpest constraint. A single 100 MW hyperscale centre can use around 20 lakh litres of water a day for cooling, and national data-centre water demand is set to rise steeply toward 2030, much of it drawn from municipal systems.
- Location multiplies the risk. Data centres cluster in Bengaluru, Hyderabad and Gurugram, cities that already face water rationing and grid strain, so new demand competes directly with residents.
- The power draw is carbon-heavy. Large, continuous electricity consumption on a coal-dominant grid means AI’s compute translates into significant emissions unless matched by renewable sourcing.
- Hardware becomes waste fast. High-performance GPUs turn obsolete within two to three years, generating hazardous e-waste laden with rare and toxic materials that India lacks the capacity to recycle safely.
- Opacity blocks governance. Many operators disclose little about their water and power use. Without transparency, the state cannot plan supply, price the externality or hold anyone accountable.
Data and Institutions Vault
Carry these into the exam hall.
- A 100 MW hyperscale data centre can use around 20 lakh litres of water per day for cooling (CEEW).
- India’s data-centre water demand projected to rise sharply toward 2030.
- AI GPUs turn obsolete in about 2 to 3 years, driving hazardous e-waste.
- Stressed hubs: Bengaluru, Hyderabad, Gurugram.
- Frameworks in play: E-Waste (Management) Rules, EIA process, state water and power planning.
- Concept anchors: grid emission factor, water stress, recycled-water cooling, digital sovereignty.
The Debate
For light-touch treatment: Data centres are strategic infrastructure underpinning India’s digital economy, AI capability and data sovereignty. They attract investment and jobs. Onerous disclosure or clearance requirements could divert this capacity to more permissive jurisdictions, ceding a critical industry.
Against (for accountability): Strategic value does not exempt an industry from its externalities. Concentrated water and power demand in already-stressed cities is a real cost borne by residents. Where operators disclose almost nothing, the public cannot even measure what it is subsidising. Governance is a precondition for sustainable growth, not an obstacle to it.
Balanced verdict: India should welcome data-centre investment while insisting it internalise its footprint. Mandatory resource disclosure, integration into state energy and water plans, recycled-water cooling and renewable sourcing let the industry grow without hollowing out local resources. The choice is not AI versus the environment; it is planned AI versus unplanned AI.
How to Think About This (Transferable Skill)
Technique: make the invisible externality visible before you regulate it. You cannot manage what you cannot measure. The first governance step for any resource-hungry industry is mandatory, standardised disclosure of what it actually consumes and emits. Only then can the state plan supply, price the externality and set limits. Apply this “measure, then manage” sequence to groundwater extraction, mining, thermal plants and any activity whose costs fall on the commons.
Diagram-in-Words
AI demand → data centres built in stressed cities → power + water + land drawn, GPUs obsolesce → local scarcity + carbon + e-waste (often undisclosed) → (IF disclosure + energy/water planning + recycled-water cooling + renewables + e-waste recycling) → AI growth without local depletion
The Way Forward
- Mandate resource disclosure. Require standardised reporting of water, power and land use and e-waste generation for every large data centre.
- Plan power and water together. Integrate data-centre demand into state energy and water plans so siting reflects genuine local carrying capacity.
- Incentivise recycled-water and renewable cooling. Reward operators that use treated or recycled water and renewable electricity over fresh water and grid coal.
- Scale hazardous-electronics recycling. Build GPU and server recycling capacity ahead of the two-to-three-year obsolescence wave.
- Strengthen clearance where stress is acute. Apply environmental-impact scrutiny to large centres proposed in water-stressed districts.
The Takeaway Box
Mains angle: Reframe AI as physical infrastructure with an environmental cost. Argue that disclosure and resource planning, not prohibition, reconcile digital ambition with sustainability, using the water and e-waste data as evidence.
Lift line: “The choice is not AI versus the environment; it is planned AI versus unplanned AI, and only disclosure makes planning possible.”
Prelims hooks: 100 MW data centre uses ~20 lakh litres water/day (CEEW); water demand rising toward 2030; GPUs obsolete in 2-3 years; stressed hubs Bengaluru, Hyderabad, Gurugram; E-Waste Rules and EIA process.
Ethics/Interview angle: Distributive justice, whether a strategic industry may draw scarce water from communities that already ration it, is an ethical question, not only a technical one.
PYQ linkage: Connects to GS3 questions on emerging technology, environmental impact assessment, e-waste management and sustainable resource use.
Connects to: E-waste rules, water security, grid decarbonisation, EIA reform, digital sovereignty and the India AI Mission.
Sources: Business Standard
Source: AI's Hidden Footprint — Ujiyari.com | Free UPSC & State PCS Editorial Analysis
