Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsBefore choosing a cloud provider, compare its energy, water and emissions disclosures only where the definitions, reporting periods, facility scopes and regions line up. Then check how renewable-electricity claims are calculated, whether you can see estimates for your own cloud usage, which claims have independent assurance, and what local grid and water conditions apply to the regions you plan to use. Provider-wide averages are useful for screening, but they do not establish the impact of a particular workload or its effects on a local community.
Which sustainability measures should you compare?
Start with the measures the provider actually defines and reports. A number without its calculation method, reporting period and facility scope is difficult to interpret—and may not be comparable with another provider’s figure.
Power usage effectiveness (PUE)
PUE is total data-centre facility energy divided by energy used for computing. A lower value indicates less facility-energy overhead relative to IT energy, but it does not, by itself, show that a workload has lower emissions: electricity sources and workload energy use matter too.
For context, Microsoft reports a global PUE of 1.17 for FY25, covering data centres it fully owns and controls that had operated for 12 months at calculation time. FY25 ran from 1 July 2024 to 30 June 2025. AWS reports an average global PUE of 1.14 for 2025, while Google reports a fleet-wide PUE of 1.09 for 2025. These are provider-reported averages, not measurements for the specific facilities serving a customer’s workload. Check each provider’s PUE definition and scope, AWS cloud sustainability figures and Google data-centre disclosures before comparing them.
#1 Best Overall
Water usage effectiveness (WUE)
WUE is commonly expressed as water per unit of IT energy, but the exact water quantity matters: a provider may report water withdrawn rather than water consumed. Microsoft defines its measure as annual water used for humidification and cooling divided by annual IT-equipment energy, in litres per kilowatt-hour. It reports global WUE of 0.27 L/kWh for FY25. AWS reports 0.12 litres withdrawn per kWh of IT load globally for 2025. Do not rank those values as if they use the same definition; verify the water measure, scope and reporting period in the providers’ Microsoft and AWS disclosures.
How the published figures compare—and why they do not form a league table
| Provider | Published figure | Scope and qualification |
|---|---|---|
| Microsoft | PUE 1.17; WUE 0.27 L/kWh | Global, FY25 (1 July 2024–30 June 2025). PUE covers facilities Microsoft fully owns and controls that had operated for 12 months at calculation time. WUE is humidification and cooling water relative to IT-equipment energy. Source and definitions. |
| AWS | PUE 1.14; WUE 0.12 litres withdrawn per kWh of IT load | Global, 2025. WUE is stated as water withdrawn; the cited disclosure does not make it directly comparable to Microsoft’s differently defined WUE. Source. |
| PUE 1.09 | Fleet-wide, 2025. The figure is not a region- or workload-specific result. Source. |
These provider-reported numbers use different definitions, periods and facility scopes. Use them to identify follow-up questions, not to declare a winner without confirming that the underlying measurements are aligned.
What should you ask about renewable electricity and emissions?
A corporate renewable-energy or carbon-free-electricity claim does not automatically show what electricity served a particular workload at every hour. Ask for the calculation method and whether the claim is matched to the location and time of your usage. Also establish which emissions scopes and emissions factors are included, and whether figures describe operational electricity or a broader boundary.
- How are renewable or carbon-free electricity claims accounted for, and are they location- and time-matched?
- Which emissions factors and scopes are included in the provider’s calculations?
- Can the provider provide information relevant to the regions and services you will actually use?
- Are the stated figures measured, modelled or based on an allocation method?
If the provider cannot connect a broad claim to your expected regions and usage, treat it as company-level context rather than evidence of your workload’s electricity mix or emissions.
Rank #3
How do cooling and local water conditions affect the choice?
Cooling design can involve a trade-off: Google says water cooling can reduce energy use and related carbon emissions compared with air-based cooling, while recognising that it can increase water footprint. That makes local water availability and watershed conditions relevant alongside PUE and WUE. A global water average can conceal differences between locations.
Ask which water sources and locations are included in the reported metric, whether it measures withdrawal or consumption, and what evidence is available for the specific regions you intend to use. Microsoft estimates that its direct-to-chip cooling saves over 125 million litres of water per facility each year; this is a Microsoft-reported estimate, not an independently validated comparison across providers. See its 2025 Environmental Sustainability Report page.
Can you estimate the impact of your own cloud usage?
Provider-level averages cannot tell you the environmental impact of your actual mix of services. Ask whether the provider offers customer-level estimates for the services you use, and how its method allocates energy, water and emissions to your account. AWS describes environmental-impact estimates associated with AWS usage and notes that its methodologies can evolve. Its methodology documentation is a useful example of why the calculation method and version matter.
- Which services and lifecycle stages are included, and which are excluded?
- What assumptions allocate shared infrastructure energy, water and emissions to a customer?
- When was the method last updated, and can you retain the method version and assumptions used for a report?
- Can the estimate be broken down by the regions and services relevant to your usage?
Customer estimates are methodology-dependent. Keep the reporting version and assumptions with any figures you use for internal comparisons or public reporting.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
What evidence should you request for your intended regions?
Ask for evidence that connects provider-wide claims to the places and services you expect to use. Relevant context includes regional electricity conditions, local water availability and watershed conditions, and any published information about community impacts. The provider-wide figures cited above do not establish those local conditions or provide a consistent independent comparison of AWS, Google Cloud and Azure for a buyer’s exact regions.
Also check which figures have independent assurance. Assurance applies to a particular claim and scope; it does not necessarily cover every environmental metric, site or reporting period. AWS’s sustainability resources identify an independent assurance letter for its water-withdrawal calculation approach, but that does not establish equivalent assurance for every reported metric or for other providers. Ask what sites, period, methods and exclusions the assurance covers, and consult the provider’s sustainability resources for the scope of the AWS letter.
How does your workload design affect cloud impact?
The provider’s infrastructure is only part of the picture. Google describes sustainability as a shared responsibility between provider and customer. Microsoft recommends reducing unnecessary resource use through right-sizing, effective autoscaling, and removing idle or redundant resources. Guidance is available in the Google Cloud Well-Architected Framework and Microsoft Azure Well-Architected guidance.
Quick Recap
- Can you choose resource sizes that match actual demand rather than leaving excess capacity running?
- Can workloads scale down when demand falls, and can you remove idle or redundant resources?
- Can flexible workloads be scheduled or moved in a way that suits the available regional options and your operational needs?
A practical way to make the comparison
- Define your requirements. List the services, expected workload patterns and cloud regions you need. Without these, provider averages cannot answer the specific selection question.
- Collect each provider’s definitions and reporting boundaries. Record PUE, WUE, reporting period, facility scope and geography; distinguish water withdrawn from water consumed.
- Request electricity, emissions and customer-impact methods. Ask how claims are accounted for, which services are covered and how shared infrastructure is allocated to your usage.
- Check local evidence. Ask about relevant grid and water conditions and published local-impact information for your intended regions.
- Review assurance claim by claim. Record the metric, period, facilities and calculation method covered, along with any exclusions.
- Compare only aligned figures. Where definitions or scopes differ, mark the numbers as not directly comparable and use the gap to guide follow-up questions rather than treating one as a better result.
- Include workload controls in the decision. Assess whether your architecture can right-size, autoscale and eliminate idle capacity, and retain the assumptions behind any impact estimates you rely on.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
