The Behemoths We Wave ThroughData centres, water, and the ocean gambit

Ask where the cloud lives and most people picture something weightless. The reality is a shed the size of a stadium, drawing the power of a small city and drinking millions of litres of water to keep its processors from cooking. As demand for AI compute climbs, these buildings are multiplying—and the striking thing is how easily they arrive.

Data centres are waved through planning with a speed that would astonish anyone who has tried to add a second storey to their house. They are pitched as clean, quiet, job-light infrastructure, and on paper they are. What the paperwork rarely foregrounds is appetite: a single hyperscale facility can consume as much electricity as tens of thousands of households and, in hot climates, evaporate enormous quantities of fresh water through its cooling. Increasingly these behemoths are sited within reach of dense residential populations—close to the grid connections and fibre they need, and therefore close to the people whose water and power they now compete for.

The economics explain the ease. A data centre is a cashbowl: enormous capital, dependable returns, and a resource bill that lands on the public purse—on the catchment, the grid, the community—rather than on the balance sheet. When the externalities sit outside the ledger, approval is cheap and the true cost is paid elsewhere.

China's answer is audacious. Off Hainan, the marine-technology firm Highlander has commissioned what is billed as the world's first commercial underwater data centre—roughly 1,300 tonnes of sealed compute submerged about 35 metres down, cooled not by scarce fresh water but by the sea itself. The efficiency case is real: seawater cooling is claimed to cut the energy spent on cooling by up to ninety percent, and the province's plan reportedly envisions a network of a hundred subsea cabins. A comparable trial is underway off Shanghai. Solve the water problem by moving into the water: on its face, elegant.

But the sea is not a heat sink without consequence. Every joule of cooling is heat moved into the surrounding water. A 2020 assessment of an early Zhuhai pilot reported nearby temperatures staying below thresholds—reassuring at pilot scale. Scale is the question. As Shaolei Ren of UC Riverside has cautioned, "for megawatt-scale data centers underwater, the thermal pollution problem needs to be studied more carefully." Cluster a hundred cabins in a warm, current-fed sea and the concern stops being a rounding error: localised warming around the arrays and, over time and volume, the possibility of nudging the temperature of the currents that move past them, with effects on marine life we are poorly placed to predict. Some have begun to call it thermal debt—efficiency borrowed now against an environmental bill that comes due later, in the water.

None of this is an argument against data centres; the compute has to live somewhere. It is an argument against waving them through. The same pattern recurs on land and now at sea: a genuinely useful facility, a genuinely large externality, and an approval process that treats the second as someone else's problem. Whether a data centre sits on a floodplain beside a suburb or on the seabed beside a reef, the governing question is the same—who accounts for what it consumes, and who is watching when it scales.

We are very good at approving the thing we can see, and very slow to price the thing we cannot. The cloud has a temperature. Someone should be measuring it.