Are data centres bad for the environment? The global numbers say one thing, the local ones say another

The honest answer to whether data centres are bad for the environment is that it depends almost entirely on where you are standing, and that the two sides of this argument are usually citing figures at completely different scales.

Start with the global picture, because it is the one that surprises people. According to the International Energy Agency, data centres account for **just over 1 per cent of global electricity demand** and roughly **0.5 per cent of global CO2 emissions**. At planetary scale, that is not what is driving climate change. Anyone telling you AI is boiling the oceans is not working from these numbers.

The trajectory is steeper than the level, though. Global data centre electricity demand reached roughly 460 to 490 terawatt-hours in 2025 and grew 17 per cent that year. Consumption at AI-focused facilities specifically rose about 50 per cent. The IEA projects total demand roughly doubling by 2030, with associated emissions reaching around 350 million tonnes by 2035 — which would still be about 2 per cent of electricity-sector emissions.

Now change the scale, and the picture inverts.

In the United States, data centres used around **4 per cent of national electricity in 2023**, and Lawrence Berkeley National Laboratory analysis projects **7 to 12 per cent by 2028**. That is a national figure absorbing a very uneven distribution. Northern Virginia, parts of Texas and the Phoenix area face grid expansion costs that a globally-averaged percentage renders invisible.

Ireland is the clearest case. The IEA estimates data centres could reach **32 per cent of Irish electricity** — close to a third of a national grid, in a country whose entire consumption is a rounding error globally. Nothing about the 1 per cent global figure prepares you for that.

This is why the two camps talk past each other. The defenders cite global shares, which are genuinely small. The critics cite local grids, transmission queues and county-level water withdrawals, which are genuinely strained. Neither is lying. They are describing different objects.

The pattern repeats on water. Per-query water use is measured in millilitres under every current published estimate, including the one its own author revised downward — and a facility with a modest per-query footprint sited in a water-stressed county is still a local problem. Small unit costs multiplied by concentrated siting is the whole shape of this issue.

So the useful version of the question is not whether data centres are bad for the environment in the abstract. It is: what does this specific facility draw, from which grid, from which watershed, and who pays for the transmission built to serve it. Those have answers. They are just answers that vary by county rather than by industry.

One thing worth separating out: grid expansion costs and residential electricity bills are related but not the same claim. Utilities in high-concentration regions face material costs to serve new load. How much of that reaches household bills depends on rate cases, tariff design and who regulators decide should carry it — and that is a jurisdiction-by-jurisdiction determination, not a settled fact. Coverage that jumps straight from data centre growth to your bill going up is skipping the step where that gets decided.

Every figure above is a projection past its base year. The IEA's own numbers are scenario-dependent, and Berkeley's 7 to 12 per cent range is wide because the underlying build-out is uncertain. Treat the direction as well supported and the specific destination as not yet settled.