Sources of data center energy estimates: A comprehensive review

Overview

This article is a systematic review by David Mytton and Masaō Ashtine, published in Joule in 2022 (volume 6, issue 9, pp. 2032-2056). The study examines 258 data center energy estimates from 46 original publications between 2007 and 2021 and traces the 676 sources they rely on to judge reliability.

The methods fall into three groups: bottom-up (equipment specifications times installed base), top-down (regional or corporate totals) and extrapolation (growth in traffic or intensity). Only one of the reviewed publications used a top-down approach. Extrapolation results share the bottom-up baseline but diverge quickly as the projection extends. The authors find that many estimates lack clear boundaries, methodology and data provenance.

On sources, 31% were peer-reviewed publications and 38% non-peer-reviewed reports; 43% relied on private IDC data and 30% on Cisco; 11% had broken web links and 10% were cited with insufficient detail to locate. Some estimates show global data center consumption plateauing from 193.7 TWh (2010) to 205.2 TWh (2018), whereas extrapolation scenarios claimed up to 51% of global electricity by 2030. Ireland's consumption rose 144% from 1.2 TWh (2015) to 3.0 TWh (2020) and was expected to reach 27% of Irish demand by 2029; Denmark projected 15% by 2030.

Estimates of future IT energy demand range by an order of magnitude, from about 200 TWh to about 8,000 TWh. The authors say extrapolation-based projections should be treated with great caution, and they give recommendations for end-users, researchers and policy-makers. As a limit, the study covers publications up to 2021 and does not include later developments driven by AI. It still provides a sound framework for questioning the method and data source when reading energy forecast figures.

Key points

1. 258 estimates from 46 original publications between 2007 and 2021 were analysed through 676 sources.
2. 31% of sources were peer-reviewed and 38% non-peer-reviewed reports; 43% relied on private IDC data and 30% on Cisco; 11% had broken links and 10% were cited with insufficient detail.
3. Some estimates show global consumption plateauing from 193.7 TWh (2010) to 205.2 TWh (2018), whereas extrapolation scenarios claimed up to 51% of global electricity by 2030.
4. Ireland's consumption rose 144% from 1.2 TWh (2015) to 3.0 TWh (2020) and was expected to reach 27% of Irish demand by 2029; Denmark projected 15% by 2030.
5. Only one reviewed publication used a top-down approach; extrapolation results share the bottom-up baseline but diverge quickly.
6. Estimates of future IT energy demand range by an order of magnitude, from about 200 TWh to about 8,000 TWh.
7. Methods: bottom-up, top-down and extrapolation.

Using it in design

• Where the tool shows national or global context figures or forecasts, label the method (bottom-up, top-down, extrapolation), the system boundary and the source, and prefer bottom-up or metered data.
• Build facility-level energy forecasts bottom-up from equipment power and utilisation, and avoid traffic-based extrapolation.
• Keep a provenance field for every default value in the catalog.

Who should read it

Designers, analysts and policy-makers who produce or use energy demand forecasts.

Publication details

Authors

David Mytton; Masaō Ashtine

Publisher

Joule (Elsevier BV)

Year

2022

Volume

6

Issue

9

Pages

2032-2056

Document language

English

Source type

Scientific

Access

Free to read

DOI

10.1016/j.joule.2022.07.011

Licence

Elsevier open access user licence — review the publisher’s conditions for reuse

Citation