What it all runs on.
Almost no chipmaker publishes an absolute electricity figure, so this page leads with the statistical offices that do, and then shows what each company publishes about itself, blanks included.
National data-centre electricity
Metered and official. Ireland publishes national data-centre electricity quarterly; the Netherlands publishes it annually, with the share of national use. Taiwan is shown as national context for its fab clusters, not as a semiconductor series.
Two reports that are not series
One modelled estimate for the United States and one reported subset for the European Union. Neither is metered national data, and they do not compare with each other or with the series above. Each says what it is, beside its figures.
What operators publish about themselves
Straight from each company's own sustainability or annual report. Nothing estimated, nothing interpolated, blanks left blank.
Industrial electricity prices
One panel per publisher, each on its own basis, never merged onto one scale: band, tax treatment, currency and period all differ between them. The line is the publisher's headline area; the table lists every place where this record holds a fab site. A large fab negotiates its own contract, so none of these is what a fab pays.
What this does and does not show
- The United States has no official data-centre electricity series. Lawrence Berkeley National Laboratory's figure is modelled from equipment shipments, and its 2028 figure is a range because the report publishes a low and a high scenario. This page shows both ends and no midpoint.
- The European Commission's first assessment of the EU reporting scheme covers the data centres that reported, about a third of those expected, and publishes no EU total. Multiplying its per-site mean by the count would produce a number the report does not state, so this page does not.
- The Netherlands figure is not in CBS's StatLine catalogue. It comes from a custom table CBS published on 15 December 2025, one file covering 2017 to 2024, so it moves only when CBS publishes a newer table. It counts electricity delivered over the public grid; power generated on site is not in it. The newest year is provisional, as CBS marks it.
- These are national totals. Fabs and data centres sit inside them and cannot be read out separately. Ireland does not split its data-centre figure by tenant, so Intel's Leixlip campus is in there and invisible.
- A data centre is not a fab. They are shown together because they compete for the same grid, not because the numbers are comparable.
- TSMC's own report appendices switch between "total energy consumption" and "indirect energy, non-renewable" across years. Those are different metrics, so they are not joined into one line here.
- Water is reported as withdrawal, not consumption, and definitions differ between operators. Treat cross-company comparison with care.
- The three price panels do not compare with one another. Eurostat's band is 150,000 MWh a year or more, excluding VAT; EIA's is every industrial customer, revenue divided by sales; Taipower's is its whole power-service class. Japan, South Korea and China publish tariff schedules rather than average prices, so they have no panel.
- No forecast of this record's own, no attribution of any total to any company, and no view on whether any of it is too much.