Data centre statistics 2026

Electricity use

How much electricity data centres use, globally, by region and by country, with each source's own scope. Estimates for the same place differ because sources count different facilities and years.

National figures on our market pages

These come from the country records and update with them. 20 regional and national electricity figures, including GB, Norway, Sweden and Australia, are on the hyperscaler and cloud market hub.

Not shown

• A 2024 IEA total for Europe in TWh is not stated in the text we checked, only the growth to 2030.
• The EU Joint Research Centre estimate is not shown because its publication page could not be checked.

Download electricity use figures (CSV)

National share of electricity

The share of a country's electricity used by data centres, from national statistics offices and governments. Each country measures this differently.

Not shown

• A national share for the United Kingdom, France and the Netherlands is not shown because we have not verified an official figure.

Download national share of electricity figures (CSV)

Power usage effectiveness (PUE)

PUE is total facility energy divided by IT energy. A lower figure means less overhead for cooling and power conversion. Survey averages hide wide variation by size, age and climate.

Not shown

• Operator-reported fleet PUE is on each operator's page where we hold it, and is not averaged here.

Download power usage effectiveness (pue) figures (CSV)

Rack density

How much power a typical rack draws. Survey averages are still in single digits even as AI racks reach well above 100 kW, so both views are shown.

Not shown

• Vendor rack figures for AI systems are on the chip power page, not repeated here.

Download rack density figures (CSV)

Liquid cooling adoption

How many operators use direct liquid cooling (DLC), from the largest public survey we could check.

Not shown

• Nearly half of DLC users said less than 10% of their racks use DLC. Uptime does not give an exact share in the public summary.
• The 2025 Cooling Systems Survey results are behind a paywall, so they are not shown.
• The share of installed capacity that is liquid cooled is not publicly reported by a neutral source we could check.

Download liquid cooling adoption figures (CSV)

Colocation vacancy and pricing

Vacancy and asking rents from commercial property advisers. Each adviser defines its markets differently, so figures for the same region do not match.

Asking rents, Q1 2026

Rent changes, North America primary markets, H1 2026

Not shown

• Cushman & Wakefield figures are not shown because we have not checked them against a public primary page.
• JLL asking rents are not shown because we could not find them on a public JLL page.
• CBRE gives North American H1 2026 rents only as changes, the price levels are in charts we cannot read as text.

Download colocation vacancy and pricing figures (CSV)

Hyperscaler capital expenditure

Cash spent on property and equipment each quarter, from the companies' own SEC filings. This is company-wide spending, not only data centres. Microsoft and Oracle use different fiscal years, so quarters are labelled with each company's own calendar and are never added together.

Values are the cash flow line "purchases of property and equipment" (XBRL PaymentsToAcquirePropertyPlantAndEquipment, or PaymentsToAcquireProductiveAssets for Amazon). Equipment obtained under finance leases is not a cash payment and is not included, so these figures can be lower than the capex a company quotes on its earnings call. Where a filing gives only a year-to-date figure, the quarter is the difference between two year-to-date figures, marked as derived.

Set against operating cash flow, quarter by quarter: hyperscaler free cash flow.

Source: US Securities and Exchange Commission, EDGAR XBRL company facts (10-Q and 10-K filings) (opens in a new tab). Accessed 19 Oct 2026. Full-year 2026 capex guidance (5 figures) is on the hyperscaler and cloud market hub, so it is not repeated here. Guidance and filed spending are different measures. Market share and cloud region data are on the operator pages and in cloud regions.

Not shown

• Capex spent on data centres alone is not publicly reported by these companies in their filings.
• Google Cloud, AWS and Microsoft cloud capex is not reported separately.

Download hyperscaler capital expenditure figures (CSV)

Grid connection queues

How much demand is waiting to connect to the grid, from grid operators and regulators. Queues count applications, not demand that will be built, and each operator defines them differently.

14 more national grid connection and request figures, from Norway, Finland, Austria, Spain, France, Malaysia, Brazil and others, are on the hyperscaler and cloud market hub. They are not repeated here.

Not shown

• Queue figures for EirGrid (Ireland), PJM, TenneT and Amprion are not shown yet because we have not verified them on the operator's own pages.

Download grid connection queues figures (CSV)

Water

Water used on site, mostly for cooling, and water used to generate the electricity data centres consume. Direct and indirect water are different measures.

Not shown

• Water usage effectiveness (WUE) across the industry is not publicly reported by a neutral source we could check.
• Other operators' water figures are not compiled here yet.

Download water figures (CSV)

Construction costs

Construction cost benchmarks per MW, with their sources and scope, are on their own page so they are kept in one place: construction cost benchmarks.

Related data

• Country indicators and open data
• Markets
• Operators
• Cloud regions
• Construction cost benchmarks
• Chip power and cooling
• Energy pathways

Status labels: Official statistic or regulator figure Published estimate Projection or forecast Survey result Commercial market report Company filing Derived from company filings (difference of year-to-date figures) Company report