India's data-centre build-out collides with groundwater stress
India's data-centre capacity is set to rise from 1,500 MW today to 13.56 GW by 2031-32, but the projects are being sited in water-stressed districts — Gautam Buddha Nagar draws 104.79% of its natural groundwater recharge, and Visakhapatnam has the lowest groundwater reserves in Andhra Pradesh.
What happened
India's data-centre capacity has grown from about 375 MW in 2020 to around 1,500 MW by 2025 and is projected to reach 13.56 GW by 2031-32. State policies are aggressively courting the industry: Andhra Pradesh's Data Centre Policy 4.0 offers 100% state GST reimbursement on capital goods, 100% stamp duty exemption on the first sale and deemed distribution licences for direct energy purchase for projects of at least 300 MW. Uttar Pradesh's Data Centre Policy, 2026 targets over 2 GW of additional capacity, and Gujarat's 2026-29 policy targets 7.5 GW and ₹6 lakh crore of investment. UC Riverside research estimates a typical ChatGPT conversation of 20 to 50 exchanges can consume up to half a litre of water, and a UN report puts the global electricity-linked water footprint of data centres at 4.5 trillion litres in 2025, projected to reach 9.3 trillion by 2030.
Context
The build-out is landing on already stressed groundwater. The Central Ground Water Board's 2024 Dynamic Ground Water Resources assessment puts India's overall stage of groundwater extraction at 60.47%, with 11.1% of assessment units classified as over-exploited. Gautam Buddha Nagar in Uttar Pradesh, home to at least 17 operational or upcoming data centres, has a stage of extraction of 104.79% — it draws more water each year than nature replaces. Hyderabad is officially over-exploited. Visakhapatnam, where Google and AdaniConneX plan a 1 GW AI hub, had just 2.12 thousand million cubic feet of groundwater in reserve as of April 2026 — the lowest of any Andhra district. Gujarat's Saurashtra-Kutch belt near Jamnagar and Dholera is grappling with seawater ingress and salinity. Protests have already erupted in Visakhapatnam and in Thane against Amazon's proposed hyperscale project.
Why it matters
Data centres are a textbook case of a negative environmental externality colliding with an industrial-policy incentive stack. State GST reimbursement, stamp-duty exemptions and deemed distribution licences all lower the private cost of a data centre; groundwater is priced at close to zero and cooling water is not disclosed at plant level, so the social cost is not internalised. Siting decisions are therefore driven by fiscal and land incentives rather than by hydrological carrying capacity, which is why 17 data centres now cluster in a district (Gautam Buddha Nagar) that already extracts more groundwater than it recharges. Without mandatory disclosure of monthly and peak water use, water source and whether potable, reclaimed or groundwater is being used — the transparency the source article's expert asks for — the state effectively subsidises an environmentally destructive input.
Impact for India
- Data-centre operators (Google-AdaniConneX, Meta-Reliance, Bharti Nxtra, Amazon): Get generous state fiscal and power incentives but face rising social-licence risk from protests in Visakhapatnam and Thane.
- Water-stressed districts: Gautam Buddha Nagar (extraction 104.79%), Hyderabad (over-exploited), Visakhapatnam (2.12 thousand million cubic feet of groundwater reserves in April 2026) and the Saurashtra-Kutch belt face compounded pressure as data centres draw large volumes of freshwater or tanker water.
- State exchequers: Trade near-term GST and stamp-duty revenue foregone (100% reimbursement on capital goods and 100% first-sale stamp duty exemption in Andhra) for long-run investment and jobs — a fiscal-for-environmental trade with no plant-level disclosure to price it.
- AI users: A typical 20-to-50-exchange ChatGPT conversation is estimated to consume up to half a litre of water; the cost is invisible in the query price.
- Marine ecosystems near coastal data centres: Seawater cooling — pioneered by Google in Finland for more than 10 years and now planned along India's coast — cuts freshwater use but discharges warm brine, a separate externality.
- Regulators (CGWB, MoEFCC, State PCBs): Face pressure to make water disclosure mandatory, enforce stricter permitting in over-exploited zones and set brine-discharge standards for coastal sites.
Key data
Concepts
Theory lens
Common-pool resource governance — groundwater is a classic rival, non-excludable resource whose extraction is hard for any single user to internalise. Fiscal incentives that ignore hydrological carrying capacity — GST reimbursement, stamp-duty exemption, deemed distribution licences — reduce private cost while the social cost falls on the aquifer. The article's proposed fix (mandatory disclosure of monthly and peak water use, water source, and stricter permitting in water-stressed regions) maps directly on to the Ostrom-style prescription: information, monitoring and enforceable use rules turn a de facto open-access resource into a governed commons.
Stakeholders
Practice Question — from the story
With reference to India's data-centre expansion and groundwater stress, consider the following statements:
- India's overall stage of groundwater extraction, as per the Central Ground Water Board's 2024 Dynamic Ground Water Resources assessment, is above 100%.
- Gautam Buddha Nagar in Uttar Pradesh, home to at least 17 operational or upcoming data centres, has a stage of groundwater extraction of 104.79%.
Which of the statements given above is/are correct?
Reveal answer & explanation
Answer: 2 only
- Statement 1: Incorrect — India's overall stage of groundwater extraction, per the CGWB's 2024 Dynamic Ground Water Resources assessment, is 60.47%; 11.1% of assessment units are over-exploited, but the national number is well below 100%.
- Statement 2: Correct — Gautam Buddha Nagar's stage of extraction is 104.79% and it houses at least 17 operational or upcoming data centres.
Therefore, the correct answer is 2 only.
Concept check — test your understanding
Groundwater aquifers drawn on by clustered data centres are classic "common-pool resources". Which of the following best captures what makes a resource a common-pool resource, and why such resources tend to be over-exploited?
Reveal answer & explanation
Answer: A
- Common-pool resources are rival (one user's withdrawal directly reduces what is available to others) but non-excludable (excluding additional users is costly). This asymmetry produces the tragedy of the commons: each user's private cost falls below the social cost, and the shared stock — here, the aquifer — is drawn down.
- B confuses common-pool resources with state monopolies; the defining feature is exclusion difficulty, not legal ownership.
- C conflates zero market price with zero economic value — unpriced natural capital still has real value, which is precisely what environmental economics tries to price.
- D describes a public good (non-rival, non-excludable), which is a distinct category from a common-pool resource (rival, non-excludable).