Introduction

India's data-centre industry has grown more than fourfold in five years from 375 MW of installed capacity in 2020 to roughly 1,575 MW today, with the government's own projections putting demand at 13 - 17 GW by 2031-32. That growth is now colliding with three resources India cannot manufacture more of, which is land, electricity and water, in a country that already runs on a majority-coal grid and holds only about 4% of the world's freshwater despite 18% of its population. This article checks the industry's own numbers on power, water, jobs and investment against government data, city-level records and global benchmarks, to separate what's verified from what's simply claimed.

1. How Many Data Centres, and Where?

India had 271 data centres as of January 2026 (Rubix Data Sciences, cited by Business Standard), ranking the country 7th globally by facility count. Just five metros Mumbai, Hyderabad, Delhi NCR, Bengaluru and Chennai account for nearly 65% of all facilities, and Mumbai alone holds roughly 49 - 50% of national installed capacity by megawatts, thanks to its submarine-cable landing stations. The remaining sites are scattered across 20+ other locations, with Andhra Pradesh, Madhya Pradesh, Chhattisgarh and West Bengal named by the government as the newest expansion states.

Figure 1. Data-centre concentration by city (approximate share of 271 facilities).

2. Electricity: How Much, and From Where?

India’s data-centre capacity has grown from 375 MW in 2020 to about 1,575 MW today, still representing less than 1% of India’s total installed power capacity. Looking ahead, the Central Electricity Authority (CEA) of India (Ministry of Power) projects that data centres could require 13.6 - 17 GW of power capacity by 2031 - 32, underscoring the sector’s growing importance to India’s power infrastructure. While the government has not disclosed actual annual electricity consumption by data centres or its split between coal and renewables, S&P Global Commodity Insights estimates that data centres consumed approximately 13 TWh of electricity in 2024, or 0.8% of India’s total electricity demand, with consumption projected to rise indicatively to 57 TWh by 2030, equivalent to 2.6%.

Figure 2. India's data-centre power capacity, per government statements to Parliament.

What GW and Twh mean and how do you visualize it? 

A gigawatt is the power capacity required by data centres or how big the electricity requirement is. In comparative terms India's entire data-centre fleet today has a combined power capacity of about 1.575 GW which is roughly 27% of the record 5.87 GW peak load of a city like Bengaluru has drawn on a hot summer afternoon (BESCOM data). In other words, if you added up every data centre in the country right now, they would still draw meaningfully less power at once than Bengaluru's own homes, offices and ACs do on an April afternoon. That changes fast though, because the government's own 2031-32 projection (13.6 - 17 GW) is 2 - 3 times Bengaluru's entire recorded peak, spread nationally.

A Twh tells you how much electricity the data centre uses over time. If Indian data centres consumed 13 TWh in 2024, that is roughly equivalent to four-fifths of all the electricity consumed across Bengaluru in a year. And the projected 57 TWh in 2030 would be equivalent to more than 3.5 times Bengaluru's current annual electricity consumption (using 2024 - 25 as the benchmark).

What's Being Proposed for the Future, and Where Will the Extra Power Come From?

The future datacentre projects include Google's 1 GW hub near Visakhapatnam (construction underway since April 2026, first phase targeted for late 2028), a similarly sized Reliance-Brookfield joint venture in the same city, Reliance's own gigawatt-scale Jamnagar project, and Lodha Group's 400-acre, ~1 GW park at Palava near Mumbai. The total capacity is projected to rise from 1.5 - 1.6 GW today to 4.8 - 6.7 GW by 2030

Where will the power come from?

For the near term, the honest answer is mostly the existing grid, which is still roughly 71% coal-generated nationally. The CEA projects that coal will still account for around 55% of India’s electricity generation in 2030, down from 73% today. However, as overall electricity demand grows, coal-based generation is expected to increase in absolute terms even as its share of the generation mix declines. On top of that, the CEA has begun directing states like Maharashtra, Andhra Pradesh, Telangana, Tamil Nadu to factor data-centre demand into their resource-adequacy planning specifically because of concerns about local grid strain.

Operators are also striking dedicated deals. CtrlS signed an MoU with NTPC Green Energy (November 2025) for a 2 GW renewable project, Equinix signed India's first data-centre-specific renewable Power Purchase Agreement (PPA) (with CleanMax, November 2024, 33 MW solar-wind in Maharashtra) and Meta's newly announced built-to-suit Indian facility is designed to run on renewable electricity paired with desalinated seawater for cooling. These are not hypothetical but are signed, funded commitments, though they currently cover a small fraction of the sector's total load.

Are the proposed alternatives (renewables, air cooling) proven, or hypothetical?

• Renewable PPAs for data centres which are proven, but small scale. Equinix and CleanMax (33 MW, operational), CtrlS and NTPC Green (2 GW, MoU stage) are real transactions following a well-established global pattern of hyperscalers signing dedicated renewable contracts.

• Small modular nuclear reactors (SMRs) for data centres currently hypothetical in India. The SHANTI Act creates a legal pathway for SMRs to power emerging sectors such as AI and data centres, but no SMR is yet operating or under construction for an Indian data centre. Globally too, the IEA does not expect SMRs to meaningfully power data centres before 2030.

A question about the already established nuclear power plant Kaiga may arise for many but can data centres in Karnataka just draw power from the existing nuclear plant instead of waiting for SMRs?

Not really. Kaiga's four operating units generate 880 MW total, but that power is already fully allocated across a five-state southern grid pool across Karnataka, Andhra Pradesh, Kerala, Tamil Nadu and Puducherry and Karnataka itself currently receives only about 28% of Kaiga's own output. There is no idle capacity sitting there for a data centre to tap. Drawing more would mean taking power away from existing consumers in five states, not adding new supply. Two new 700 MW units (Kaiga 5 & 6) are under construction, which will more than double the site's capacity to 2,280 MW but construction only reached first concrete in March 2026, with NPCIL's own timeline putting completion roughly five years away. And even then, the plan is for that new power to flow into the state grid pool, not to any specific industrial customer. Karnataka is promised 50% of the new unit's output, the same way it already shares Kaiga's existing power, not a dedicated supply line to one buyer.

Net assessment: The specific alternatives cited by government and industry are technically real and working somewhere in the world (and in a few committed India-specific deals), but they cover a small share of India's current data-centre load. The bulk of the sector, today, still runs on ordinary grid electricity that is roughly 70% coal.

2a. Water and Land: How Much Is Being Used?

No national water-consumption figure for data centres has been published by any ministry . Industry and research-body has estimated water use at roughly 8 billion+ litres per gigawatt of capacity per year. Applied to India's current ~1.575 GW, that implies data centres nationally draw on the order of 12-13 billion litres of water a year (estimate number). The government confirmed that groundwater extraction by data centres, like any industrial user, is regulated under Ministry of Jal Shakti guidelines, and that operators are moving toward direct-to-chip liquid cooling, adiabatic cooling (a process which lowers air temperature by evaporating water into an air stream) and closed-loop systems to cut water use.

At ~12.6 billion litres a year, all of India's data centres combined, use roughly 15% of what a city like Mysuru city's water utility supplies its residents in a year (~84 billion litres, based on Mysuru City Corporation's 230 Million Liters per Day, Dec 2024).

If the water consumed is a small portion in comparison, then why are people protesting the datacentres citing water concerns?

Because the national number and the local reality are two different things. Nationally, data centres use a small slice of India's water but almost none of that water is spread out evenly. Half of it is drawn in and around one or two cities, and those happen to be places that are already short on water. So, the question people are protesting isn't “is this a lot of water for India” it's “why is a private company getting a guaranteed 20-year water supply in a city where residents already don't get enough.” That's a fair question, and the small national percentage doesn't answer it.

There are two other reasons the small national number understates the problem -

• First, most of the water figures used above are only rough industry estimates and no government body has published an official number for how much water India's data centres use, so nobody outside the companies really knows the true figure.

• Second, a lot of the real damage happens underground, through borewells, and that never shows up in any city-wide total. For example, one site near Ghaziabad was found pumping groundwater at nearly three times the recommended safe limit.

What about desalination? Using seawater instead, since many sites are near the coast?

This is already happening. Meta and Reliance's new data centre in Jamnagar, Gujarat will run entirely on renewable power and use desalinated seawater for cooling, so no drinking water is touched at all. It follows the same approach already used at large water-treatment plants in Chennai, which turn seawater into drinking water for the whole city.

The concern people raise is fair on its face which is, doesn't turning seawater into freshwater need its own extra electricity, which then needs its own extra coal, in an endless loop? In principle, yes. In practice, the extra amount is small. Turning seawater into freshwater takes a fairly modest, well-understood amount of power roughly 1 extra unit of electricity for every 100 units the data centre already uses. That is a real added cost, and it does add some pollution if that power comes from coal, but it is nowhere near a second crisis the size of the first. But this is unfortunately not the case for Bengaluru, Hyderabad and Delhi NCR which are not located in the coast and deal with real water crisis where there is not enough water for primary human consumption itself.

Can the datacentres share an existing desalination plant and save energy?

This is the most realistic near-term option for coastal cities (e.g. buying spare water from Chennai's city plants rather than building a private one). But those plants are usually already stretched thin trying to keep up with the city's own growing needs, which is exactly why a city like Chennai is building a fourth one. Handing some of that supply to a private data centre goes back to the  same argument - “who gets water first” argument as before.

How much land has the datacentres across India occupied?

On land, India's entire current data-centre footprint (~620 acres) is about 70% the size of Bengaluru's Electronic City Technology park (~900 acres across all phases). Today's national footprint is smaller than one large Bengaluru tech campus cluster, though the pipeline through 2030 (101 million sq ft, ~2,320 acres) would be nearly four times that.

3. The Upside of Datacentres: Employment. How Many, and Are They Real, Long-Term Jobs?

Government and industry sources cite 100,000 to 430,000 jobs by 2030 nationally, and up to 188,000 jobs for Google's single 1 GW Visakhapatnam project alone. These numbers deserve scrutiny. 188,000 is an extraordinary figure for one facility, larger than the entire workforce of many large Indian companies.

Cross-checking against how data centres staff up, globally studies from markets with a decade-plus of data-centre operating history converge on a consistent range. A large hyperscale campus needs roughly 1-2 permanent employees per megawatt once operational (automation does almost everything else), smaller colocation facilities need more, at 4-8 jobs per megawatt. A Brookings 2026 study of US counties that got their first large data centre found employment in the data-processing sector rose, but by an estimated 100 -200 jobs per county is ideal not tens of thousands.

Applying the same 1-2 jobs-per-MW hyperscale ratio to Google's 1,000 MW (1 GW) Vizag project implies roughly 1,000-2,000 permanent operational jobs once fully built not 188,000. The gap is not necessarily dishonest, the 188,000 figure almost certainly bundles peak construction employment (which is large but temporary, typically 2,000-5,000 workers for 2-3 years per campus, per US benchmarks) together with indirect and induced jobs across the wider ecosystem Andhra Pradesh's IT minister has described like AI, cybersecurity, cloud operations, manufacturing, and ancillary industries rather than jobs at the data centre itself.

Bottom line: treat large jobs created numbers for data centres as economy-wide, mostly-temporary job estimates, not payrolls. The pattern documented in the US, is a large, short construction boom followed by a small, permanent operations team is what global staffing data would predict for India's data centres too, and there is no technical reason India's hyperscale facilities would need dramatically more on-site staff than equivalent facilities elsewhere, since the same automated hardware and software run them.

4. Environmental Impact, Protests, and the Investment Question

Opposition has grown sharply through 2026, concentrated on water and pollution concerns at two flagship projects:

• Visakhapatnam, Andhra Pradesh (Google, $15 bn, 1 GW). Activists group such Green Visakha, have held marches and “beach protests,” citing that the city already receives 410 million litres of water a day against a requirement of 480 million (a 70-million-litre daily shortfall) in a city of 2.5 million. A public interest litigation from activist group Jal Biradari (Water Community) alleges the project will strain a nearby reservoir. The Andhra Pradesh High Court has so far declined to halt construction, and the state government disputes the activists claims. Construction is actively continuing, with first-phase completion still targeted for late 2028.

• Balkum, Thane, Maharashtra (Amazon, ₹47,000 crore / ~$5.6 bn). Residents are opposing a proposed 420 MW facility on a 53-acre plot, citing heat, noise, diesel-generator emissions and power-supply concerns.

Legal and civic opposition is real, growing, and now includes court cases, but projects have not been halted and construction continues and the state governments are actively contesting litigations. At the same time, national industry trackers such as Anarock Capital still project the sector-wide pipeline growing, and a separate climate-risk study flagged that 12% of planned data centers face high physical climate risk from extreme heat which is a different, longer-term risk to the sector than the current protests, but investors are increasingly pricing in alongside water and permitting concerns.

5. Zooming Out: Is This Just an India Story, or the World's?

Every pattern documented so far in India regarding the gap between cited and realistic jobs, the water fights that don't show up in national averages, the guarded optimism around desalination, the disputed investment slowdown also shows up, usually in a more advanced and better-documented form, in the rest of the world.

5.1 The Global Scale: Power and Water

There are about 12,200+ active data centres worldwide globally across 179 countries with more than one-third of that in the US. Data centres consumed an estimated 485 TWh of electricity in 2025, on track to reach 950 TWh by 2030 (IEA, 2026) a trajectory that dwarfs India's own 1.575 GW of capacity, and explains why every grid operator from Dublin to Virginia is now treating data centres as a distinct planning category.

Figure 3. Global data-centre electricity demand, IEA Electricity 2026 / April 2026 update.

Water follows a similar curve. Independent trackers put global data-centre water use at roughly 560 billion litres, rising toward 1.2 trillion litres by 2030 (S&P Global), with a separate 2025 estimate (Data Center Watch) of 264 billion US gallons (~1,000 billion litres) that year. As in India, no single global regulator publishes an audited, sector-wide water figure. Every number above is an independent estimate, not an official one.

Figure 4. Global data-centre water consumption, reconciling S&P Global and Data Center Watch estimates.

5.2 What's Already Built, and What's Coming

The existing global base is large but the pipeline dwarfs it. OpenAI alone has reportedly signed roughly $1 trillion in infrastructure commitments during 2025, anchored by the Stargate joint venture with SoftBank (announced at $500 billion), including a flagship Abilene, Texas site that has grown toward 7 GW combined with partner projects, plus newer sites in Michigan (1+ GW) and Georgia (3.2 GW).

Separately, OpenAI and SB Energy announced an Ohio project in August 2026 designed to reach 8 GW eventually, with the first 800 MW online by 2028. Microsoft, Google, Amazon and Meta are each running capital-expenditure programmes in the tens of billions of dollars a year on top of that.

The honest catch, confirmed by the companies' own suppliers, is that the power to run all of this is arriving slower than the buildings and chips are. Microsoft CEO Satya Nadella said in a November 2025 interview alongside OpenAI's Sam Altman that Microsoft has AI chips sitting in inventory because it lacks warm shells and they require powered, cooled data-centre space to plug them into.

In the US specifically, Goldman Sachs projects data-centre power demand more than doubling from 31 GW (2025) to 66 GW (2027) and 84 GW (2028), while Morgan Stanley (August 2026) estimates the grid can only reliably cover about 30 of the roughly 68 GW needed through 2028 leaving a 38 - 45 GW hole with no confirmed source of power.

This is the same dynamic playing out at smaller scale in India's own 1.5 GW-to-17 GW pipeline.

Figure 5. The gap between US data-centre power demand and confirmed supply is widening, not closing (Goldman Sachs; Morgan Stanley, 2026).

Figure 6. Data-centre electricity consumption by region, 2025 vs. 2030 (IEA / LBNL / Rystad). China and Southeast Asia are the two fastest-growing markets after the US.

China is the world's second-largest market, and the most self-contained. Its data-centre capacity is projected to nearly double from 32 GW (end-2025) to 60 GW by 2030, with electricity consumption rising from roughly 140 TWh to somewhere between 277 and 289 TWh (IEA; independent modelling) comparable in scale to the entire US shortfall discussed above, but arrived at very differently. China installed a record 357 GW of wind and solar in the first half of 2025 alone (more than India's entire installed power base, across all sources), taking renewables to roughly 60% of total installed capacity. In the very same period, it also commissioned 21 GW of new coal power, the highest addition since 2016, driven by concerns about grid reliability. Southeast Asia (excluding India) is smaller but growing faster in percentage terms.

The regional data-centre market is projected to more than double, from $13.7 billion in 2024 to $30.5 billion by 2030, led by Malaysia's Johor state, which has approved roughly $41.7 billion (164.45 billion ringgit) in data-centre investment since 2022 to become the region's fastest-growing hub largely by absorbing capacity that Singapore's own 2019 moratorium pushed across the border.

Europe is the most self-limiting market with grid caution, not resource scarcity being the ceiling. Data-centre electricity consumption is projected to rise from roughly 65 TWh in 2025 to 110 TWh by 2030, a comparatively modest trajectory next to China or the US, largely because several of Europe's biggest markets have chosen to restrict growth rather than chase it: the Netherlands and Frankfurt have paused new grid connections for data centres until at least 2030, and Dublin already draws 79% of its electricity for data centres, prompting an effective moratorium there too.

The result is that new European hyperscale sites are being pushed further from cities at 175 km from urban hubs on average for 2026-28 projects, versus 46 km for 2022-25 projects as developers chase available land and grid capacity rather than proximity. Ireland's national data-centre electricity share is projected to reach 32% by 2026, and the UK faces a potential sixfold increase in data-centre demand within a decade. France stands out as a partial exception. Its large nuclear fleet let it export a record 92.3 TWh of electricity in a recent year, positioning it as one of the more attractive sites in Europe precisely because it isn't fighting the same power-scarcity battle as its neighbours.

The Middle East is spending its way around a resource constraint rather than avoiding it. MENA is among the most water-stressed regions on Earth with summer temperatures regularly exceed 40°C, freshwater stress runs roughly four times the global average, and seven of the world's eight most water-stressed countries are there, yet it's one of the fastest-growing data-centre markets, not the slowest. The approach is capital-intensive rather than resource-light. Microsoft alone has committed $15.2 billion to UAE data centres, tied explicitly to the region's ability to guarantee reliable power. Saudi Arabia's Vision 2030 includes major AI infrastructure spending and Morocco has announced a 500 MW facility. The region is leaning hard on desalination to make this work. MENA produced roughly 12 billion cubic metres of desalinated water in 2024, a figure the IEA expects to triple by 2035 but desalination itself is electricity-hungry, so cooling and desalination together already account for an estimated 40% of MENA's projected electricity demand growth through 2035. In effect, the Middle East isn't solving its water problem; it's converting it into a bigger electricity problem it believes it can afford to solve with capital.

Japan is growing steadily but is constrained by geography as much as by power. Electricity consumption from data centres is projected to rise from roughly 18 TWh in 2025 to 32 TWh by 2030 a real growth, but modest next to its neighbours held back less by resource scarcity than by an ageing grid and genuine land constraints in a dense, mountainous country with limited flat, seismically stable sites near population centres.

Africa remains the starkest outlier globally, on the opposite end from the Middle East. The continent holds under 1% of global data-centre capacity despite housing 18% of the world's population, and roughly 600 million people in Sub-Saharan Africa still lack reliable electricity access. Per-capita data-centre electricity use there is below 1 kWh a year today, barely reaching 2 kWh by 2030 which is several orders of magnitude below the global average. Unlike the Middle East, Africa isn't spending its way around scarcity; it simply isn't yet part of the build-out at meaningful scale, which raises the equity question flagged earlier in this article from a different angle. Africa isn't just capturing a small share of AI's economic upside it's largely absent from the physical infrastructure story too.

5.3 The Green Promises: What's Real, and What Isn't

Every major builder like Google, Microsoft, Amazon, Meta has a public net-zero or carbon-neutral target, generally for 2030. The most rigorous independent check on these promises is NewClimate Institute's Corporate Climate Responsibility Monitor (July 2025), which assessed the climate strategies of 55 major global companies including Amazon, Apple, Google, Meta and Microsoft.

Its finding is that none of the 55 companies achieved a high or reasonable integrity rating, and the five tech companies specifically were found to fall short of demonstrating credible leadership, with outdated emissions accounting methods and surging energy demand undermining the meaning of Greenhouse Gas (GHG) targets. The report did credit Google and Microsoft with genuine leadership on one specific practice which is hourly (24/7) renewable-energy matching, a stricter standard than simply buying enough annual renewable credits to offset a year's usage but noted this remains “an isolated example rather than a sector-wide practice.”

The specific numbers explain why independent reviewers are sceptical. Big Tech accounted for 43% of all corporate clean-energy power-purchase agreements signed globally in 2024, which itself drove PPA prices up 35% in a single year meaning renewable procurement is scaling, but is simultaneously becoming unaffordable for smaller buyers.

Microsoft has purchased 3.5 million carbon credits to offset AI-related emissions which seems like an accounting mechanism rather than a reduction. And the technology most frequently cited as the long-term fix, small modular nuclear reactors is not close. The global nuclear fleet actually shrank by a net 1.1 GW in 2025 (two reactors added, seven retired), and the IEA does not expect SMRs to meaningfully power data centres before 2030.

Sam Altman has personally invested in nuclear fission (Oklo), fusion (Helion) and concentrated solar-thermal storage (Exowatt) as potential power sources for AI and has said plainly that none of those are ready for widespread deployment today. Separately, Nadella has noted that the cost of producing a unit of AI intelligence has fallen roughly 40x per year, historically the kind of efficiency gain that should reduce total energy use except that cheaper intelligence simply gets used more, a pattern economists call the rebound effect (or Jevons' Paradox), and one Nadella himself flagged as “a very scary exponent from an infrastructure buildout standpoint.” In other words, the people building this infrastructure agree that efficiency gains are being outrun by demand growth, not the other way around.

China's approach to the same problem is structurally different when compared to the Western PPA-and-pledges model. Rather than relying on individual companies to voluntarily buy renewable power, the Chinese government runs a binding national programme, Eastern Data & Western Computing initiative launched in 2022, to enable a more balanced and efficient digital economy by processing everyday digital activities such as mobile payments and ride-hailing data in real-time in eastern hubs, while heavier computing tasks are transferred westward. Independent energy analysts such as Oxford Institute for Energy Studies in report published in Feb 2026 find it has had limited impact in practice, because only a small share of total data-centre load can actually be shifted to the west and the upfront cost of the necessary transmission links falls mainly on state-owned companies.

5.4 Do Bills Actually Go Up? The Verified Evidence

This is the one question with a hard, audited, real-world answer because the US grid serving data centres has existed long enough to measure it. PJM Interconnection, which serves 67 million people across 13 eastern US states, ran a capacity auction in which prices rose 833% (roughly 10x) between the 2024/25 and 2025/26 delivery years. PJM's own independent market monitor (a body required by US federal regulators) modelled what the auction would have priced at, with no data-centre load at all, and found data centres responsible for 63 - 75% of the entire price increase translating to $9.3 - 9.4 billion in added costs for the year, with a further $1.4 billion increase already locked in for the following year. Cumulative costs are projected to reach $100 - 163 billion through 2033, with the average PJM household facing roughly $70 extra per month by 2028 ($840 a year), according to US Senate testimony and independent utility analysts.

The effect is sharpest closest to the data centres themselves. In Virginia, which now sends 40% of its electricity to data centres (up from under 5% in 2010), one county with 37 existing data centres saw its own government and school electricity rate jump 25% in a single year, and typical residential customers are facing state-regulator-approved increases of $11.24/month in 2026 and a further $2.36/month in 2027 specifically to fund grid upgrades for data-centre demand. Wholesale electricity prices in some zones near data centres have risen as much as 267% over five years. This is, in short, exactly the scenario India's own residents worry about when a government guarantees a data centre 20 years of water or power access. The US case shows what happens several years further down that same road, in hard billing data rather than a citizen's fear.

This particular mechanism, however, is not universal and China is the clearest counter-example. Household electricity prices there are centrally set (around 7.8 US cents/kWh as of April 2026) rather than determined by a competitive capacity auction like PJM's, so a surge in data-centre demand does not automatically flow through to residential bills the way it does in the US. The lesson for India is that how a grid prices electricity, not just how much of it a data centre uses, determines whether the bill lands on residents or is absorbed elsewhere. A design choice India can still make deliberately, rather than discovering by accident the way PJM's ratepayers did.

5.5 Weighing the Good Against the Harm

On the benefit side independent economic modelling (CSIS, 2025) still expects AI to generate enormous global economic value, even though 70-75% of it is projected to concentrate in just ten countries. India's NITI Aayog separately estimates AI could add $500-600 billion to India's GDP by 2030 which is a large number, if the infrastructure, skills and chip-access gaps are closed in time.

On the harm side, the additions to the planet's burden are concrete rather than hypothetical. 56% of the electricity supplied to data centres globally still comes from fossil fuels (30% coal, 26% gas), a global nuclear fleet that is shrinking, not growing, and a water and land footprint that as this briefing has shown for India specifically concentrates real local stress even when national averages look modest. The core tension is not good versus bad but concentrated benefit versus distributed cost. The economic upside is projected to land overwhelmingly in a handful of wealthy countries and corporate balance sheets, while the water, electricity-price and land costs are borne locally, in specific cities and specific countries like Visakhapatnam and Henrico County alike.

5.6 So Why Build More? The Industry's Own Answer

Far from downplaying the issue, the AI industry has been remarkably candid about the scale of the power problem it faces. The most immediate concern is capacity: data centers worldwide are reportedly already operating near their limits, leaving little room to expand without new electricity supply. OpenAI's Sam Altman in 2024, had underscored just how large that gap is, reportedly telling White House officials directly that meeting future demand would require building next-generation data center campuses of up to 5 Gw each — roughly the output of five nuclear power plants, and enough electricity to power nearly three million homes. Notably, this pitch wasn't framed as a warning about strain on the grid, but as an economic opportunity, with the investment positioned as a boost to the local economies that would host such facilities.

The deeper rationale offered by the industry is that AI capability keeps improving with more compute, and unlocking that capability is itself the goal. Altman has argued that the industry is currently under-investing in data centres capable of building more data centres — a vision in which AI-directed robots and automated supply chains eventually construct physical infrastructure with minimal human labour, something he has described as "a very wonderful thing to do relative to other uses of that compute if we're right about what that additional compute will eventually unlock." Nadella makes a parallel argument from the cost side: as the cost of producing a unit of AI capability keeps falling by roughly 40x a year, those falling prices are expected to unlock so much new demand that total infrastructure needs keep climbing rather than levelling off.

What's notably missing from the public record — across every source reviewed for this briefing — is a specific, independently verified answer to the question of how much power is actually enough. By Nadella's own admission, in a November 2025 interview, "nobody knows" how much power AI will ultimately require. That uncertainty is exactly why the resource, water, land, and community questions raised throughout this article, in India and elsewhere, can't be waved away with the assumption that the next data centre will be the last one needed.

5.7 Full Circle: What This Means for India

India's data-centre story, examined closely in-here is not a uniquely Indian dilemma. It is the same global dilemma at an earlier, smaller stage. The US shows what a mature version of the water-and-power fight looks like in audited bills and courtroom-tested moratoriums. The corporate climate record shows what happens to green pledges once independent auditors check them and the industry's own executives admit, on the record, that they do not know how much is ultimately enough. India still has the chance to build its large-load pricing strategy, its water-disclosure rules and its community-consultation processes before its pipeline reaches PJM's scale, rather than after which is, ultimately, the most useful thing a fact-check like this article can offer. Not a verdict on whether to build, but a clearer view of what the rest of the world already learned the hard way.

Conclusion: Four Very Different Reactions to the Same Two Words

Say "AI" and "data centre" in the same sentence today, and you get four completely different reactions, depending on who's listening.

• Governments hear opportunity. For them, a data centre means capital inflows, jobs and GDP growth but also something less economic and more geopolitical: a seat at the table in a race where being early, and being a maker of this technology rather than just a user of it, counts for a lot.

• The companies building this like Google, Meta, OpenAI and others are chasing something bigger than the next data centre. Their own executives have said, in public, that they are building toward a future where far less human labour is needed to run the economy at all. For them, the data centre isn't the end goal, it's the physical shell around a much larger ambition.

• Ordinary people feel none of that optimism. What they see is rivers running low, bills going up, and a planet already struggling with heat, being asked to carry an enormous new load driven, as far as they can tell, by profit and power rather than public need.

• Investors are asking a colder, narrower question: will this actually pay off? A data centre earns nothing by itself. Its returns depend on a whole stack of things no single investor controls. Chip prices, semiconductor supply, the cost of energy, and how much computing power customers are actually willing to pay for. That uncertainty already shows up in how AI-linked stocks move. Sharp rallies followed by sharp corrections whenever doubts surface about whether the spending will ever be recovered. Nobody has a clean answer for how this cycle ends. The honest position is that the sector's valuations currently mix real revenue with real speculation, and the two haven't been cleanly separated yet.

But one thing cuts through all four reactions is that AI is not going anywhere. Whether people invited it in or not, it has already worked its way into daily life in your search, writing, translation, coding, customer service, decision-making. That alone settles the question of whether the data centres behind it will keep being built. They will.

What's actually being negotiated, then, isn't whether. It's how. The realistic future is not a sudden stop brought on by protest or policy. It is a long tug-of-war: governments and AI companies pulling to build faster, people pulling back to be heard and protected, with real friction at almost every site along the way but no red light bringing the whole thing to a halt.

If there's a way to make that tug-of-war less damaging, it isn't a technical fix, it's basic honesty. Give people real, public numbers on how much water a project will use and how much power it will draw, and where that power comes from. Not once at launch, but on an ongoing basis.

Meet fear with empathy rather than dismissal. A community worried about its water supply is not being irrational dismissing these fears is part of why trust breaks down. Create an actual, reachable point of contact. Someone positioned between the company and the community, empowered to answer real questions and flag real problems, rather than a statement issued after the protest has already happened. That middle layer will not be perfect. There is no guarantee it will be unbiased, or that it will side with the community when it should. But the alternative which is no middle layer, no real transparency, and a public left to guess at the truth from press releases and protest placards has already produced exactly the standoffs documented throughout this article, from Visakhapatnam to Virginia.

AI and the people living next to its infrastructure are both here to stay. The only real choice left is whether they learn to coexist with some accountability built in, or without it.