Compared to what?
India's datacenter water debate, 358 billion litres, and the math nobody ran
India’s AI Impact Summit last week was a serious event. Twenty heads of state in New Delhi. Five hundred global AI leaders. Major investment announcements. A genuine conversation about how the Global South actually participates in the AI economy rather than just hosting its call centers.
Then there was the other conversation.
Outlook India covered it from the summit itself. Down to Earth and Earth Journalism Network published a series of field-reported pieces worth reading — on datacenters straining Bengaluru's water supply, on a booming corridor in Uttar Pradesh running on empty wells, on what datacenter companies aren't disclosing about their actual consumption. [Links in the reading list below] The numbers in these pieces are real. The field reporting is credible. The concern is legitimate.
Read them. Then come back. I’ll wait…
Every one of those pieces cites 358 billion liters — India's projected datacenter water consumption by 2030, up from roughly 150 billion litres today — and none of them ask the prior question. Is 358 billion liters a lot? Compared to what, exactly — last year's figure? A different country? A different industry? The articles don't say. They don't compare to agriculture, to steel, to the semiconductor fabs India is simultaneously trying to attract and celebrate. They don't ask what India gets back per liter — in jobs, GDP, digital infrastructure, or the small matter of not permanently licensing its AI future from Singapore.
Just the number. The alarm. And the implicit conclusion that the number is self-evidently too high.
That’s not analysis. That’s a fact wearing an attitude.
Here’s the comparison nobody made. A hyperscale datacenter runs at 1-5 million gallons per day, mostly grey water for evaporative cooling. India wants ten TSMC-class fabs and should build every single one of them. A single such facility runs at 8-10 million gallons of ultra-pure water per day — chemically processed, largely consumed, not recirculated. One is a thirsty marathon runner. The other is a small city. The coverage of India’s fab ambitions is celebratory, as it should be. The coverage of its datacenter buildout is alarmed. The fab uses more water. Can someone please explain the logic?
So as I was saying — 358 billion liters by 2030 sounds alarming until you ask: compared to what?
Act I — Punjab Is Right There
In the SAS Nagar district of Punjab, groundwater depth went from 3.6 metres in 1995 to 30.7 metres by 2021. Nearly 9x deeper in 26 years. Across central Punjab, extraction has reached 150-200 metres in most places and is projected to drop below 300 metres by 2039. This is being driven by paddy-wheat cultivation under guaranteed procurement prices — a policy architecture that was designed in the 1960s to solve a food security problem India genuinely had, and that has been running on institutional momentum ever since. The MSP system creates a specific incentive: grow the crops that get guaranteed prices, even in regions where the water table is collapsing under the weight of doing so. It is one of the largest groundwater crises in the world, and it is a direct consequence of a policy nobody in power currently has a strong incentive to reform.
Nobody covering the datacenter water story is spending much time on it.
That gap is the tell, because it points to something structural about how water gets scrutinized. Agriculture accounts for roughly 90% of India’s freshwater withdrawals. Steel, textiles, thermal power, and cement take most of what remains. Datacenters are currently a rounding error in that picture, growing toward something more significant. The concern about their trajectory is legitimate. But the distribution of journalistic and regulatory attention doesn’t map onto the distribution of actual consumption — it maps onto novelty.
There’s a framework that would help here, and it’s worth naming precisely: return on resource. What does society actually get back per unit of water consumed — in output, employment, GDP, strategic capability? It’s the question you’d want answered before deciding whether a steel mill in a water-stressed district makes sense, whether sugarcane cultivation in Marathwada deserves subsidized irrigation, whether a coal plant should be permitted next to a depleted aquifer. The framework is neutral — it doesn’t favor technology over agriculture or industry over services. It just asks the same question of everyone.
The reason it’s rarely applied isn’t unique to India. In the American West, “first in time, first in right” water law hands 86% of allocation to agriculture while municipalities issue lawn-watering advisories during droughts. Reforming that regime would require navigating a century of property rights, rural political constituencies, and commodity markets built on cheap water — so it largely doesn’t happen. India’s version involves guaranteed procurement prices, fragmented state-level water rights, and agricultural subsidies whose beneficiaries have significant political organization. The mechanism is different; the underlying dynamic is the same. Scrutiny lands on whoever is newest and least politically organized, while structural inefficiencies in legacy sectors accumulate quietly under the heading of “complicated.”
The newest, least politically organized entrant right now is datacenters. Hence the coverage.
Act II — Let’s Do The Math
The 358 billion litre figure — a 2030 projection, up from ~150 billion litres today. Here is what the same scrutiny, applied consistently, looks like.
India’s thermal power sector — the coal plants generating the electricity that runs everything, including those datacenters — consumed approximately 2.1 billion cubic metres of freshwater per year as of 2016, and has grown since. [Source: WRI] That is 2,100 billion litres. Nearly fourteen times the datacenter sector’s current annual consumption, drawn from the same stressed river basins and aquifers. Per the Centre for Science and Environment, coal thermal plants account for roughly 70% of all industrial water withdrawal in India. [Source: Down to Earth] The NITI Aayog’s own figures put consumption at 5-7 cubic metres per megawatt-hour for plants with cooling towers. More than 40% of these plants are already located in areas of high or extreme water stress.
Between 2013 and 2017, sixty-one coal plants shut down due to water shortages. Not because of policy. Not because of economics. Because there was no water left to cool them. India lost 17,000 gigawatt-hours of electricity in the process — power that the economy it was supposed to serve simply didn’t get. [Source: Business Standard/IEEFA] The sector consuming fourteen times the water of every datacenter in India combined had already started failing because of it. This did not become an alarm series.
The BBC has not investigated this. Outlook India has not run an alarm series on it.
India’s textile industry uses approximately 1.6 billion litres of water per day — that’s the India figure, not global — translating to roughly 580 billion litres per year. The average water consumed per tonne of cotton cloth produced in India runs 200-250 cubic metres, against a global best practice of 100 cubic metres. Indian mills use two to two-and-a-half times more water per unit of output than they need to. Regulators have noted this. They have not made it an emergency. [Source: CEO Water Mandate/Gap Inc. Handbook] The textile sector contributes 2% of GDP and is the second-largest employer after agriculture. Nobody is proposing to stop building textile mills.
India’s integrated steel plants, per the Ministry of Steel’s own parliamentary disclosures, consume 1.5-4.1 cubic metres of water per tonne of crude steel on a net basis after recirculation. [Source: Ministry of Steel Lok Sabha filing] At India’s current production of roughly 150 million tonnes per year, that’s 225-615 billion litres of net consumption annually. Gross withdrawal — the actual draw on rivers and aquifers before any water comes back — runs 25-60 cubic metres per tonne by older estimates. Nobody is writing alarm journalism about Jharkhand’s aquifers to the same degree.
And then there are fabs. One TSMC-class facility: 11-14 billion litres per year, ultra-pure rather than grey water, largely consumed rather than recirculated. Each facility uses more water than most hyperscale datacenters. India wants ten of them and is in active competition to attract them, which is the correct policy. The coverage of that ambition is, as noted, largely celebratory.
The pattern is not subtle. India’s datacenter sector is being treated as a water emergency. India’s thermal power sector, consuming fourteen times as much from more stressed sources and directly responsible for power outages when those sources run dry, is treated as infrastructure. India’s textile sector, consuming comparable volumes at half the efficiency of global best practice, is treated as a heritage employer. The difference isn’t the volume or the stress. It’s the novelty.
Act III — The Return Math Nobody Ran
The volume comparison is actually the easier half of the argument.
The stronger part is this: the alarm is justified. The aperture is just too narrow. Water scarcity is real — the question is whether the solution is to scrutinize one sector or to finally build a framework that asks the same question of everyone drawing from the same wells.
AWS alone has committed $12.7 billion to Indian cloud infrastructure through 2030. Their own conservative projections put the GDP contribution of that single company’s footprint at $15.3 billion, supporting more than 81,300 jobs annually. That’s one company. Microsoft, Google, Reliance, and Adani are building at comparable or larger scale. Total capital committed to Indian datacenter infrastructure — from Amazon, Microsoft, and Google alone — crossed $67.5 billion by late 2025, against a global hyperscaler spend of $400–450 billion in 2025 alone.
Now run the same math on textiles. India’s textile sector employs roughly 45 million people and contributes approximately 2% of GDP — around $70 billion — against 580 billion litres of annual water consumption. That works out to roughly $120 million of GDP per billion litres. India’s datacenter sector, at its current early-stage market size of $5-10 billion against a projected 358 billion litres by 2030, runs at $14-28 million of direct GDP per billion litres. On raw GDP-per-litre, textiles look better.
But raw GDP-per-litre is the wrong unit. It measures what an industry produces today, not what it enables or where it’s going. Two things matter more.
First, multipliers. Datacenter infrastructure isn’t just an industry — it’s the substrate on which India’s entire digital economy runs. Every UPI transaction, every e-commerce order, every telemedicine consultation, every IT export is running on compute that lives in those facilities. Textile mills have supply chain multipliers too, but they run in a line — cotton to yarn to cloth to retail. Datacenters run underneath everything simultaneously. That’s a different kind of leverage. The Ministry of Electronics and Information Technology projects India’s digital economy reaching 20% of GDP by 2030 — an $800 billion sector sitting on top of infrastructure that uses water at roughly 60% of the rate of the textile industry.
Second, and more important for a country trying to solve its core structural challenge: productivity per worker. Agriculture employs 46% of India’s workforce and generates 16-18% of GDP. That gap exists because India’s Green Revolution created a floor — guaranteed procurement prices, subsidized inputs, rural credit — that made staying in farming rational for hundreds of millions of people even as productivity stagnated. The government’s own Economic Survey says India needs to create 7.85 million non-farm jobs annually through 2030 just to absorb its growing working population productively. Manufacturing and services employment shares have been declining. Agricultural employment is absorbing people who can’t find work elsewhere — not because farming is efficient, but because the alternatives aren’t growing fast enough.
The way out of that trap is not more low-productivity capacity. India’s textile sector: 45 million workers, roughly $1,500 of GDP per worker per year. India’s IT sector: 5 million workers, roughly $50,000 of GDP per worker per year. That’s a 33x productivity gap — and it maps almost perfectly onto the water debate. India’s productivity gap doesn’t close by building more of what it already has. It closes by building the infrastructure that makes higher-productivity sectors viable at scale. Datacenters are part of that infrastructure. The water they use is an input to that transition, not an obstacle to it.
The return math was never close. It was just never run.
Act IV — What Good Actually Looks Like
If you’re genuinely worried about water — and you should be, because peninsular groundwater depletion is real, documented, and accelerating — here is what useful policy looks like. None of it is “stop building datacenters.”
Mandate closed-loop cooling at the permitting stage. India already has a regulatory norm requiring new coal plants to hit 3 cubic metres per MWh — imperfectly enforced, but the logic is right. The same standard applied to datacenters would structurally reduce consumption without stopping construction. Several Indian facilities are already operating at near-zero water dependency using dry or immersion cooling. The technology exists. The mandate doesn’t. This is not a hard lift.
Require grey water for cooling, full stop. Datacenter cooling doesn’t need potable water. Evaporative cooling works fine with treated municipal wastewater — the same solution already standard in water-stressed parts of the US and Israel. India’s Draft National Data Center Policy gestures at water efficiency without mandating standards. Gesturing is not a policy. Make it a floor.
Route capacity toward lower-stress locations through incentives, not mandates. Most of India’s datacenter buildout is clustering in Bengaluru, Mumbai, and Noida — which are also the most water-stressed metro areas. State-level policies already compete on land subsidies and power tariff concessions. Adding water infrastructure access to the siting incentive framework — proximity to wastewater treatment plants, coastal sites with seawater cooling options — would route capacity naturally without command-and-control. The same mechanism that created the concentration problem can solve it.
Apply the efficiency framework to everyone, not just the newest entrant. This is the one that actually matters. India doesn’t have a water allocation framework with consistent return-on-resource logic across sectors. If it did — water pricing that reflects actual scarcity, mandatory sector-by-sector consumption reporting, efficiency standards for the largest industrial users — the datacenter conversation would take care of itself. So would the textile mills running at twice global best-practice water intensity. So would the coal plants consuming 70% of industrial freshwater withdrawal with essentially no public scrutiny.
What is not a policy is “don’t build datacenters.” That move doesn’t reduce water stress — thermal power, textiles, and irrigation keep consuming what they consume regardless. What it does is redirect $67.5 billion in committed investment, and the compute sovereignty that comes with it, to Singapore, Malaysia, or the UAE, which have no such objection. India retains the water stress from every legacy industry nobody is protesting. It forfeits the infrastructure that the next century of its economy runs on.
That’s not an environmentalist position. That’s an own goal wearing environmentalism as a costume.
The Part Where I Stop Writing
The journalists who documented Bengaluru’s reservoir stress and the UP corridor’s depleted wells did real work. The facts they found are real. The concern about where India’s water goes as its economy industrializes is legitimate and worth sustained attention.
What those pieces lack is the one thing that would make the concern actionable: a consistent framework applied across sectors, not a spotlight trained on whoever built something new this year.
A number without a comparison is not analysis. An alarm without a baseline is not reporting. And a critique of one industry’s resource use that goes silent on every comparable industry isn’t environmentalism — it’s technology anxiety that happened to find some water statistics.
India is making a generational infrastructure bet. The countries building the compute own the AI supply chains; the countries that don’t are buying access on terms set by others. That bet has real costs. Those costs deserve honest, rigorous, cross-sector scrutiny.
The question was never whether to scrutinize. It was whether to do it honestly.
So far: no. This was an argument for changing that.
Reading List
The pieces that prompted this: Outlook India — AI Impact Summit 2026: Water Stress From Data Centres A Cause For Concern · Down to Earth — Bengaluru's tech dreams collide with a worsening water crisis · Down to Earth — A booming data centre corridor in UP's GB Nagar is running on empty wells · Down to Earth — What data centre giants aren't saying about their water use
Datacenter water consumption: 150B litres (2025) → 358B litres projected (2030): Mordor Intelligence — Data Center Water Consumption in India Market Report
Thermal power water consumption (2,100 billion litres/year): World Resources Institute — Parched Power: Water Demands, Risks, and Opportunities for India’s Power Sector
Thermal power: 70% of India’s industrial water withdrawal: Centre for Science and Environment / Down to Earth — Water footprint of thermal power plants in India
61 coal plants shut down 2013–2017, 17,000 GWh lost: IEEFA / Applied Economics Clinic — Risks Growing for India’s Coal Sector
Textile water consumption (580 billion litres/year, 2x best practice): CEO Water Mandate / Gap Inc. / Institute for Sustainable Communities — Water Stewardship in the Indian Textile Industry: A Handbook of Recommended Good Practices
Steel water consumption (1.5–4.1 m³/tonne crude steel): Ministry of Steel — Lok Sabha Unstarred Question No. 1096
AWS India: $12.7B investment, $15.3B GDP contribution, 81,300 jobs: AWS — Press release, May 2023 and Maharashtra MoU announcement, January 2025
$67.5B combined datacenter investment (Amazon, Microsoft, Google): Washington Post — The AI spending frenzy reaches India, December 2025