Google's biggest data center outside America needs a gigawatt and a city's water. The city already rations both.

The biggest single foreign investment Google has ever made in India is not, at bottom, a computing story. It is a water-and-power story wearing a computing headline. The plan, announced last October and now being fought in an Indian courtroom, is a $15 billion artificial-intelligence hub in Visakhapatnam, on the east coast of Andhra Pradesh, built with the Adani group and Airtel across three campuses and sized at one gigawatt of capacity. A gigawatt is the number that matters, and it decides two other numbers underneath it: how much electricity this thing draws continuously, and how much water it needs to keep from cooking itself. Both of those are quantities Visakhapatnam is already short of. That is the whole of the problem, and no amount of language about jobs and green energy changes the arithmetic.

Start with the water, because that is where the fight is. Visakhapatnam is a city of about 2.5 million people that receives roughly 410 million litres of water a day against a requirement of about 480 million. Do the subtraction: the city runs a standing daily deficit of around 70 million litres before anyone builds anything, and water rationing is a normal fact of life there, the way the generator was a normal fact of life where I grew up. Into that deficit you are proposing to place a one-gigawatt data center, one of the most water-hungry classes of industrial load there is. The question is not whether that is technically possible. It is who gives up their share so the servers get theirs.

Air cooling is a promise, not a number

Google's answer to this is that it will use, in its words, 'advanced air cooling to protect vital local water resources,' and the Andhra Pradesh government has added that 'no water for rural or residential purposes will be used' for the data centers. Take both statements seriously, because air cooling is a real engineering choice and not a slogan. A data center cooled by air rather than by evaporating water can cut its on-site water draw dramatically. That is genuinely the responsible design in a water-stressed place, and if Google builds it that way, the direct consumption falls.

But price the promise properly, because air cooling does not make the demand disappear — it moves it. Cooling a gigawatt of computers with air instead of water means moving much more air, and moving air costs electricity. Every joule you refuse to shed into evaporating water, you shed by running fans and chillers harder, and that electricity has to be generated somewhere, and generating it — unless every last megawatt is genuinely new renewable capacity firmed around the clock — itself consumes water at the power plant. The industry has a name for this trade because it is unavoidable: you can build a data center that sips water and drinks electricity, or one that sips electricity and drinks water, but a gigawatt of heat has to go somewhere. 'Advanced air cooling' is the choice to pay in power what you refuse to pay in water. It is defensible. It is not free, and the bill lands on the same grid the city's households are on.

You can build a data center that sips water and drinks electricity, or one that sips electricity and drinks water. A gigawatt of heat has to go somewhere.

This is the pattern anyone who covers energy in the global South learns to distrust: the headline commitment names the cost it has solved and stays quiet about the cost it has relocated. On-site water: addressed. The water embedded in a gigawatt of continuous generation, and the strain a gigawatt of new always-on demand puts on a grid that already rations, are the costs that move off the page. They do not vanish. They are just charged to a different account, usually the one with the least ability to contest the charge.

The sanctuary at 860 metres

Then there is the land, and here the specifics are unusually stark. One of the campuses sits about 860 metres from the Kambalakonda Wildlife Sanctuary, a 70-square-kilometre dry-evergreen forest that is home to leopards and pangolins inside a fast-growing industrial city. Eight hundred and sixty metres is not a buffer; it is a walk. The objection is not sentimental. A gigawatt-scale construction site and then a permanently operating facility bring earthworks, traffic, light and a constant low mechanical noise, and the petitioners argue the state cleared the project without properly weighing that against a protected habitat next door or against the rural drinking-water scheme the campuses may draw from.

That argument is now in court, which is the part that makes this more than a press release. A public-interest litigation filed by an activist group called Jal Biradari — the name means Water Community — is before the Andhra Pradesh High Court, with the next hearing set for August 24. Three further cases have been filed at India's environmental court by the Human Rights Forum. The state, governed by an ally of the national government that has called the project historic and transformational, denies fast-tracking it without assessing the risks. Whichever way the court rules, the litigation has already done one useful thing: it has forced the water and the habitat onto the public ledger, next to the investment figure, where they can be read together.

Now weigh it against the jobs

None of this means the project should not be built. A data center is not a vanity; India's own AI ambitions need domestic compute, and a coastal city with subsea-cable access and an industrial base is a rational place to put it. The case for the project is the economic development, so weigh that honestly too — and here the numbers deserve the same scrutiny as Google's cooling claim. The figure quoted in the coverage of the court fight is 'up to 188,000 jobs.' The figure in the original announcement is more modest and more specific: roughly 5,000 to 6,000 direct jobs, and 20,000 to 30,000 total once indirect employment is counted. Both can be true — the largest number is the broadest possible multiplier, the smallest is the payroll — but a reader deciding whether the trade is worth it should hold the payroll figure, not the multiplier, because the payroll is the part that is contracted and the multiplier is the part that is hoped.

• The load: one gigawatt of continuous capacity, $15 billion over 2026–2030, three campuses — built with Adani (via AdaniConneX) and Airtel, India's largest data-center campus and Google's biggest India investment.
• The deficit: Visakhapatnam already receives about 410 million litres of water a day against a ~480 million-litre need, and rations water routinely.
• The habitat: a campus sits about 860 metres from the Kambalakonda Wildlife Sanctuary (~70 sq km; leopards, pangolins); the case is before the Andhra Pradesh High Court, next hearing August 24.
• The jobs, priced honestly: 'up to 188,000' is the broad multiplier; the announced payroll is 5,000–6,000 direct, 20,000–30,000 total.

So measure this the way you would measure any large energy claim: not by the size of the investment but by the physical quantities it moves and who supplies them. A gigawatt of AI compute is going into a city with a 70-million-litre daily water deficit, next to a leopard sanctuary, on a grid that already rations. Google has chosen the cooling design that spends power to spare water, which is the right call and also a relocation of the cost, not a cancellation of it. The state has promised no residential or rural water will be diverted, a promise a court is about to test. The development is real and the compute is needed. But the honest accounting does not stop at $15 billion and 188,000 jobs. It runs down to the litres and the megawatts, and in Visakhapatnam both of those columns were already in deficit before the first server arrived. Watch the water filings and the power-purchase agreements, not the ribbon-cutting. That is where anyone actually finds out who paid.