Why Data Centre Megawatt Announcements Can Mislead

Executive summary

A data centre announced at "500 MW" tells you far less than it appears to. Megawatts measure power, not electricity used; the figure quoted may describe computing equipment, the whole building or a grid connection; and it usually refers to an eventual build-out rather than anything operating now. This guide sets out the distinctions, works through a clearly labelled illustrative calculation, and ends with five questions worth asking about any capacity announcement.

Why this matters to the UK

British debate about AI and electricity runs on announced megawatts. Those figures feed into arguments about bills, grid capacity and whether the country can supply what is being planned. When power and energy are used interchangeably, or a connection capacity is reported as consumption, the resulting numbers can be out by a factor of several — usually upwards.

Power and energy are not the same measure

Megawatts (MW) measure power — a rate

A megawatt is one million watts: the rate at which electricity is being used at an instant. It is the size of the tap, not the volume of water. A facility rated at 100 MW has a maximum; it does not run at that maximum continuously.

Megawatt-hours (MWh), gigawatt-hours (GWh) and terawatt-hours (TWh) measure energy

Energy is power multiplied by time. One megawatt sustained for one hour is one megawatt-hour. A thousand MWh is a GWh; a million MWh is a TWh. Electricity bills, national statistics and carbon accounting are all in energy units. A megawatt figure cannot be converted into an energy figure without knowing how heavily, and for how long, the equipment runs.

Which megawatts are being quoted?

Announced, approved, connected, operational

These four are separate states and rarely coincide. An announced capacity is a developer's statement of intent. An approved capacity has planning consent, which says nothing about power. A connected capacity has a grid agreement — a right to import, not a load. Only an operational facility draws electricity, and it draws what its workload requires, not its rating. Our analysis of the UK data centre pipeline shows how few tracked projects disclose any defined figure at all.

What PUE does and does not establish

Power usage effectiveness, defined in ISO/IEC 30134-2, is the ratio of total data centre energy to the energy delivered to IT equipment. A PUE of 1.2 means 1.2 units of electricity enter the facility for every unit reaching the computing equipment; the balance is mostly cooling and electrical losses.

PUE is a ratio, so it says nothing about absolute consumption: a very large facility with an excellent PUE uses far more electricity than a small inefficient one. It also says nothing about carbon intensity, water use, or how useful the computing work was. An annualised, measured PUE is a meaningful operating metric; a design PUE quoted in a planning application is a target.

A worked example — illustrative assumptions only

The following calculation uses invented inputs to show the method. It is not a measurement, a forecast, or a description of any real project.

Assume a facility with 100 MW of maximum IT load, averaging 70% of that electrical load across a full year of 8,760 hours, with an annual PUE of 1.2.

Three cautions. The 70% figure is an electrical load factor — the share of maximum IT power drawn on average — not GPU utilisation, which is a different quantity. The PUE of 1.2 is assumed, not measured. And the method breaks if the headline 100 MW is in fact a grid connection capacity or a whole-facility rating: applying an IT-load calculation to those would inflate the answer. You can vary these inputs yourself in our data centre electricity calculator.

Why data centre electricity is not all AI

Data centres have hosted email, video streaming, banking, retail and government systems for decades. AI training and inference are additions to that workload, often inside the same buildings. No official UK dataset separates AI electricity consumption from other data centre consumption, because metering is by site, not by workload. A facility marketed as AI-ready may run substantial conventional load; a conventional facility may run AI inference. Attributing a site's whole consumption to AI overstates AI's share by an unknown margin. We keep that distinction in the electricity demand section of the UK AI Energy Index.

Five questions to ask about any capacity announcement

1. What is being measured? IT load, whole-facility power, or grid connection capacity?
2. Which phase does it cover? The first building, or an ultimate build-out across the whole site?
3. What is the delivery status? Announced, consented, connection agreed, under construction, or in service?
4. What load is actually expected? Is there a stated utilisation or load factor, and a measured or designed PUE?
5. Where does the figure come from? A planning document, a regulatory filing, a company statement — and on what date?

If the answer to any of these is unavailable, the honest treatment is to record the figure as unverified rather than convert it into a consumption estimate. That is how we handle it in the UK data centre tracker, and why many records show capacity as unknown.

Sources

• ISO/IEC 30134-2, Information technology — Data centres key performance indicators — Part 2: Power usage effectiveness (PUE) — the standard definition of PUE.
• DESNZ, Energy consumption in the UK — the official UK energy statistics series, reported in energy units.
• Ofgem, Proposed data centre connection reforms (July 2026) — for the distinction between contracted connection capacity and demand.