Outgrace, home
Technology & AIData brief

Data centre power demand: what the sources can and cannot tell you

The IEA analysis this brief was meant to rest on is absent from every source we checked. Here is what leaders can still measure, and what to ask for.

Close-up view of a row of industrial electricity meters for power monitoring and technology.
Photo: Connor Scott McManus / Pexels

The brief behind this article rests on analysis from the International Energy Agency (IEA) of electricity use by data centres and artificial intelligence. We could not find that analysis. The three sources our researcher checked, the U.S. Energy Information Administration's electricity hub, the IFRS Foundation's climate disclosure standard and Our World in Data's energy pages, contain no IEA report, no IEA figure and no number on data centre or AI power demand. This piece therefore cannot tell you how much electricity data centres will draw. It can tell you what those sources do establish, and how to build a defensible estimate from the records your organisation already holds.

What the data shows

The U.S. Energy Information Administration publishes an electricity hub that covers sales, prices, generation, demand and emissions. Our check of that page found no data centre figure and no IEA figure. A reader who opens it expecting a data centre number will leave empty-handed. What the hub does supply is a baseline: the regional and national series against which any new load would eventually appear. For a company with sites in the United States, those series are the place to start, because they show what the grid looked like before any forecast changed the picture.

Our World in Data says electricity accounts for roughly one-fifth of global primary energy use. It defines primary energy as energy entering the system before transformation. That definition is easy to skim past and expensive to ignore. A board member who sees a primary-energy figure beside a metered electricity bill is looking at two different measures, and a forecast expressed in one cannot be set against the other without a conversion that the forecast itself should explain.

The same source states that its electricity chart measures generation, not consumption. It adds that the difference can be substantial for countries with heavy imports or exports. This matters for data centres more than for most loads. A site in a grid that imports a large share of its power will appear in consumption figures for one place and in generation figures for another. If your emissions or cost estimate uses a generation series for a grid like that, the number describes the grid, not your building.

Our World in Data also states that power from new renewables is now cheaper than new fossil fuels in most places. The statement is general, and the page does not say how it applies to any particular contract. It does tell a leader that the cost of new supply depends on what kind of generation is built. That is a question for the contract and the grid, not for a global average, and it should be answered by the supplier's own figures.

Our World in Data notes that there is no international open database for final or useful energy consumption, and that primary energy is the metric it uses. Anyone who wants a consumption figure for data centres across several countries should expect to assemble it from national sources, each with its own definitions. That is slow work. It is also why an outside number on data centre demand, when it arrives, should come with its method, its boundary and its base year. A board should be wary of minuting a figure whose method nobody in the room can describe.

There is a difference between a forecast and a measurement, and boards often blur them. A measured series, such as the metered consumption of your own sites, is evidence. A forecast of a sector's demand is an assumption about the future, and its usefulness depends on the assumptions behind it. The sources we checked provide measured series of the kind the EIA publishes, and a definition of primary energy. They do not provide a forecast for data centres. Keep the two apart in every paper that comes before the board.

What it means for a leader

Start with your own load, not the market's. Pull at least twenty-four months of invoices and, where you have it, interval meter data for every site with significant computing equipment. Separate the steady base from the peaks, and find out which part of your bill follows which. Then ask your supplier or broker to model your cost under two cases: your load held flat, and your load grown by a percentage your team chooses and can defend. The purpose is not to predict the wholesale price. It is to learn how sensitive your costs are to your own growth, which is a question you can answer this month.

Read the pricing terms for any clause that ties your cost to capacity, to a wholesale index or to a named demand forecast. Ask which forecast the supplier uses, who prepared it and when it was last updated. In our experience, the clause that causes trouble is rarely the headline rate. It is the adjustment mechanism, the formula that moves your price when a number outside your control changes. Have a lawyer or a specialist energy adviser check the wording before you sign or renew, because the cost of a badly drafted adjustment falls on the buyer.

  • Which demand forecast does this price depend on, and who produced it?
  • What is its base year, and what boundary does it cover?
  • Is the figure a measure of generation or of consumption?
  • Which clause changes my price if that forecast moves, and by how much?
  • Can you run the same calculation on my own load rather than a sector average?
  • What would reverse the adjustment if the forecast proved wrong?

Decide, in writing, which basis your organisation uses for electricity emissions, and keep it consistent from one reporting period to the next. Where a site draws from a grid whose generation and consumption differ, as the Our World in Data chart warns they can, say so in the note. A reader who later finds that a figure described one thing when it measured another will not assume good faith. A footnote costs little. A restatement costs a great deal more.

If your organisation reports under IFRS S2, the question is narrower than whether data centres matter in general. According to the IFRS website, the test is whether climate-related risks and opportunities could reasonably be expected to affect cash flows, access to finance or cost of capital, over short, medium or long term horizons. Rising power costs at a data-heavy business are a plausible candidate for that test. Whether they meet the threshold is a judgement for your finance team and your auditors. This piece cannot make it for you.

The same page says the standard is effective for annual reporting periods beginning on or after 1 January 2024, and that earlier application is permitted if IFRS S1 is also applied. It also says the ISSB issued IFRS S2 in June 2023, that the standard builds on the TCFD recommendations, and that it incorporates industry-based requirements derived from SASB Standards. A reporting team that has not reread the industry requirements for its sector recently should do so before its next cycle.

The ISSB issued amendments to greenhouse gas emissions disclosures in December 2025. According to the IFRS website, their aim is to reduce complexity, duplicative reporting and the cost of greenhouse gas disclosure under IFRS S2. For a company whose emissions note has grown by accretion, that is a reason to check the note against the amendments before the next reporting cycle and to remove whatever no longer serves a reader.

This week, do three things. Ask finance for two years of electricity cost by site and the pricing clauses for each supply contract. Ask the reporting team which electricity emissions basis is used, and whether the IFRS S2 test has been run for power costs. If an outside forecast of data centre demand reaches your desk before the IEA report itself does, treat it as a claim to be checked, with its method and date written beside it, and not as an input to next year's budget.

Limits of the data

The sources have three limits that matter for a board. None of them contains a data centre demand figure, so nothing in this article is an estimate of that demand. The EIA hub and Our World in Data measure different things, national sales and generation in one case and global primary energy in the other, so their numbers should not be placed side by side without a stated conversion. And the IFRS standard tells a company what to disclose, not what the underlying number will be. A standard can make a risk visible. It cannot tell you its size.

The IEA analysis this brief was meant to rest on should be obtained directly from the agency, with its publication date, its scenario assumptions and its definitions checked before anyone quotes it. Until that is done, this article is a guide to what can be verified and what cannot. It should not be read as a view on how much electricity data centres will need.

Newsletter

The Briefing

One email a week: the leaders we profiled, the ideas worth stealing, and nothing else.

Weekly, free, one click to leave. Read our privacy policy.