Power usage effectiveness, a datacenter’s total power over the power drawn by the IT equipment inside. The closer it is to one, the less goes to cooling and conversion. Chapter 6.
From The AI Compute Race. Read the book
Where the book uses it
- Chapter 6 — Firm, Around-the-Clock
Where the electricity comes from: grid queues, transformers, turbines and the firm power a datacenter needs every hour of the year, on calendars the buildout does not control.
Sources that use it
- Meta Platforms, Inc., “Environmental Data Index — 2025 Sustainability Report, reporting on fiscal year 2024” (page H, section 2.4 Data Center Operations and Design, the row headed PUE, glossed “data center energy efficiency”: 1.10, 1.09, 1.08, 1.08, 1.08 for 2020 to 2024, so 1.08 for fiscal 2022, 2023 and 2024 — the report’s limited assurance by Ernst & Young LLP names “select environmental metrics” and does not itemise PUE; defined on page 18 as “dividing the energy consumed at the data center by IT electricity load”, on “internal meters, design estimates and utility bills where applicable”; an annual company-level ratio across “our data centers” — the index gives no owned-versus-colocation split for PUE and does not use the word “fleet”; held byte-exact in the Library, md5 02a85ec5524a273c8ff4f40fca17eb5b). In the Sources
- Comisión Nacional de los Mercados y la Competencia, Informe IPN/CNMC/038/26 on the draft royal decree regulating the energy-sustainability, environmental, resilience and digital-sovereignty requirements for data centres, 22 September 2026 (the national queue, pages 5 and 6: grid access “otorgado en torno a 12 GW (6 GW en transporte y 6 GW en distribución)”, about 12 GW granted since 2021, 6 on transmission and 6 on distribution; “las peticiones denegadas superarían los 32 GW, y habría en torno a 27 GW en tramitación”, refused requests above 32 GW and about 27 GW in process; data centres “habrían renunciado 1,6 GW de capacidad de acceso a la red”, 1.6 GW given up; the regulator lists among the measures of Real Decreto-ley 7/2026 “la prestación económica por reserva de capacidad”, and writes “Si bien algunas de estas medidas están pendientes de desarrollo o se encuentran en implantación”; its count of permits granted “se está actualizando a raíz de las recientes renuncias como consecuencia del RDL 7/2026”; the national AI strategy foreseeing 2.5 GW of computing in 2030, “equivalentes a entre 3,5 y 4 GW de demanda eléctrica”; the 2025 peninsular peak “40.070 MW”; the draft decree: renewable supply from plant in Spain “no anterior en más de dieciocho meses” before the data centre’s own start, a PUE “igual o inferior a 1,15” and a WUE “igual o inferior a 0,1”, on pain of surcharges on grid tolls and loss of grid access; the regulator offers a glide path as an option; a draft decree and an advisory opinion, nothing in force). In the Sources
- Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report, December 2024 (176 TWh, 4.4 percent in 2023; 6.7–12 percent by 2028; “Power Usage Effectiveness (PUE) is defined as the total electricity demand of the data center divided by the electricity demand of the IT equipment”; the average PUE across all U.S. data centers projected to fall to between 1.15 and 1.35 by 2028, in its figure of annual averages). In the Sources
- Space Exploration Technologies Corp., final prospectus (Rule 424(b)(4)), 12 June 2026, first disclosed in the Form S-1 of 20 May 2026 (Business, “Compute Services Agreements with Third Parties” — Cloud Services Agreements with Anthropic PBC, May 2026, entered 3 May 2026 and carried in Notes, “Cloud Services Agreement”: compute capacity across COLOSSUS and COLOSSUS II, “approximately 325,000 NVIDIA GPUs”; “$1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee”; “After the initial three-month period, the agreements may be terminated by either party upon 90 days’ notice”, the 424(b)(4)’s own sentence; Business — Market Opportunity, pp. 176–177: the AI compute infrastructure market opportunity of approximately $2.4 trillion, from RAND’s 235 GW of 2030 data center demand (70 percent AI), a PUE of 1.2, 1.3 kW per GPU, 104 million GPUs, 80 percent utilization and “a GPU rental rate of $3.33 per hour, according to Silicon Data, which is based on the median of neocloud GPU rental rates in 2025”, the filing noting that the rental rate “has historically varied subject to market conditions”; the $26.5 trillion total market opportunity, p. 176; the filing’s “quantifiable TAM is $28.5 trillion”, of which “$26.5 trillion in AI”; the lock-up, Underwriting: “Mr. Musk has agreed with the underwriters, that during a period of 366 days after the date of this prospectus, all of the shares owned by him are subject to the restrictions described in the paragraph below”, all his shares released on 12 June 2027, the 366th day, and “Shares held by Mr. Musk will not be subject to any early release provisions”; the restrictions do not apply to a transfer “as one or more bona fide gifts or charitable contributions” where “the donee, devisee, transferee or distributee, as the case may be, must sign and deliver a lock-up agreement for the balance of the applicable lock-up period”; Use of Proceeds, p. 67: “We will receive approximately $74.4 billion of net proceeds from this offering (or $85.7 billion if the underwriters exercise their option to purchase additional shares of Class A common stock in full)”; Business, Our Market Opportunity, p. 172: “$26.5 trillion in AI across $2.4 trillion in AI infrastructure, $760 billion in consumer subscriptions, $600 billion in digital advertising, and $22.7 trillion in enterprise applications”). In the Sources
- SemiAnalysis, “Datacenter Anatomy Part 2 — Cooling Systems,” 13 February 2025 (PUE definition — “The industry average PUE, calculated by the Uptime Institute, is around 1.6” on “a survey that largely excludes hyperscalers”, against “Google and Meta both operate at around 1.1 PUE, and Microsoft and AWS around 1.15 PUE”; the Uptime Institute survey average against hyperscaler-reported figures). In the Sources
And 2 more: search the Sources.
Related terms
From the glossary of The AI Compute Race.
How to cite this page
Dryburgh, Lee S. “PUE.” In The AI Compute Race. DRYBURGH. https://dryburgh.com/vol1/glossary/pue (accessed 2026-11-01).
BibTeX
@misc{dryburgh-pue-2026,
author = {Dryburgh, Lee S.},
title = {{PUE}},
howpublished = {In \emph{The AI Compute Race}, DRYBURGH},
year = {2026},
url = {https://dryburgh.com/vol1/glossary/pue},
note = {Accessed 2026-11-01}
}