Using life cycle assessment to drive innovation for sustainable cool clouds

Overview

In this paper by Husam Alissa and co-authors, published in Nature in 2025 (vol. 641, issue 8062, pp. 331-338), Microsoft and WSP ran a cradle-to-grave life cycle assessment (LCA) comparing an air-cooled data centre with cold-plate, one-phase immersion and two-phase immersion designs. Impacts (GHG, non-renewable primary energy and blue water) were annualised and normalised per virtual core (Vcore), covering the building, servers, racks or tanks, support equipment, fluids, grid electricity and end-of-life.

Model assumptions: data centre lifetime 15 years; servers (compute, networking, storage) 6 years; racks, tanks and support equipment share the 15-year building lifetime. Data were collected per MW of IT capacity. Against the air-cooled baseline on 100% US-average grid electricity, savings were: GHG 15% (cold plate), 16% (one-phase immersion), 21% (two-phase immersion); energy demand 15%, 15%, 20%; blue water consumption 31%, 45%, 48%. Liquid cooling lowered impacts mainly by cutting use-phase energy and by allowing more servers, overclocking and longer server life; the building and the fluids mattered little.

The use phase is the largest contributor for every cooling technology; server manufacturing (embodied carbon) is second. Switching from 100% grid to 100% renewable electricity cut GHG emissions by 85-90%, energy demand by 6-7% and blue water consumption by 55-85%, so server embodied carbon became more significant. Use-phase energy was modelled as annual energy (MWh/yr) = IT capacity (MW) x load factor x PUE x 8,760 h/yr. As a limit, results depend on Microsoft and WSP's model assumptions and design deltas and cannot be generalised to every facility.

Key points

1. Against the air-cooled baseline on 100% US-average grid: GHG savings 15% (cold plate), 16% (one-phase), 21% (two-phase); energy demand 15%, 15%, 20%; blue water 31%, 45%, 48%.
2. Assumptions: data centre lifetime 15 years; servers 6 years; racks, tanks and support equipment share the 15-year building lifetime; data collected per MW of IT capacity.
3. Design deltas versus air cooling: MW IT capacity +3% (cold plate) / +6% (immersion); PUE 3% / 6% lower; overclocking +20%, +10%, +20%; Vcores +28%, +25%, +36%; compute server count +6% / +13%.
4. The use phase is the largest contributor for every technology; server manufacturing (embodied carbon) is second; building size reductions and cooling-fluid embodied impacts are small.
5. Switching to 100% renewable electricity cut GHG by 85-90%, energy demand by 6-7% and blue water by 55-85%; server embodied carbon became more significant.
6. Use-phase energy = IT capacity (MW) x load factor x PUE x 8,760 h/yr; for immersion fluid coolers adiabatic (wet) operation is assumed above 35 °C outdoor dry-bulb.
7. Compute, networking and storage servers make up 70%, 10% and 20% of IT capacity; a network server has 1.3x the embodied impact of a compute server; server end-of-life is 100% recycling after 6 years.

Using it in design

• Add a cooling-technology comparison calculator that computes annual energy as IT MW x load factor x PUE x 8760 and applies the paper's Table 3 deltas (PUE -3%/-6%, IT capacity +3%/+6%) and Table 4 savings ranges as tagged defaults.
• Show results for both grid and renewable electricity.
• Use a 15-year facility lifetime and a 6-year server lifetime as default amortisation periods, and amortise racks or tanks over the building lifetime.
• Warn that under low-carbon electricity server embodied carbon dominates, and ask for per-SKU embodied data.
• Flag two-phase immersion fluids as PFAS under regulatory review and include an adiabatic-cooler switchover temperature (default 35 °C dry-bulb) in water estimates.

Who should read it

Designers choosing cooling technology, sustainability analysts and decision makers reporting carbon and water footprints.

Related standards

• ISO 14040:2006
• ISO 14044:2006
• IPCC AR5 GWP100

Publication details

Authors

Husam Alissa; Teresa Nick; Ashish Raniwala; Alberto Arribas Herranz; Kali Frost; Ioannis Manousakis; Kari Lio; Brijesh Warrier; Vaidehi Oruganti; T. J. DiCaprio; Kathryn Oseen-Senda; Bharath Ramakrishnan; Naval Gupta; Ricardo Bianchini; Jim Kleewein; Christian Belady; Marcus Fontoura; Julie Sinistore; Mukunth Natarajan; Lauren Johnson; VeeAnder Mealing; Praneet Arshi; Madeline Frieze

Publisher

Nature (Springer Science and Business Media LLC)

Year

2025

Volume

641

Issue

8062

Pages

331-338

Document language

English

Source type

Scientific

Access

Free to read

DOI

10.1038/s41586-025-08832-3

Licence

CC BY-NC-ND 4.0 — share unaltered with attribution for noncommercial purposes

Citation