Amazon Web Services (AWS) has added annual water-withdrawal estimates to AWS Sustainability, allowing customers to view their estimated share by AWS account, service and Region.
The July 16 launch expands AWS Sustainability beyond its previous focus on carbon-emissions reporting. The water data is available at no additional charge through the console and API in every Region where AWS Sustainability is available.
Defining the data boundaries
Customers receive figures in cubic meters for their estimated share of water withdrawn by AWS-owned, leased or controlled data centers. The boundary covers water used for cooling and domestic purposes inside operational facilities, as well as water used to fill fire-flow tanks.
AWS warehouses, offices, manufacturing facilities and customer-hosted infrastructure such as Outposts are excluded under the company’s customer water-withdrawal methodology.
How AWS allocates facility water to customers
AWS begins with site-level data drawn primarily from utility bills and reports supplied by third-party data center operators. Facility meters fill some gaps, while standardized estimates are used where those sources are unavailable. AWS then aggregates the figures into geographic clusters before allocating them to customers.
The allocation model assigns cluster withdrawals to server racks using planned power draw. It then allocates those volumes to cloud services based on infrastructure use. Customer accounts receive a usage-based share for foundational services with dedicated server racks, while non-foundational services use economic allocation based on equivalent revenue when normalized usage data is unavailable or incomplete.
Service visibility and temporal limitations
The estimates cover all AWS services, although the AWS Sustainability user guide says only EC2, S3 and CloudFront are currently listed separately. Remaining services appear under “Other,” limiting service-level comparison across much of the AWS portfolio.
The estimates also have a temporal limit. Most operators of AWS-controlled facilities report withdrawal data annually, so the model does not capture seasonal variability in facility water use or identify when withdrawals occurred. “The allocation remains strictly proportional to each customer’s total resource usage, regardless of when within the year that usage occurs,” the methodology states.
AWS says it will recast historical water-withdrawal data when significant methodology updates are introduced.
Independent validation and corporate disclosure limits
Apex Companies conducted a limited independent validation of the allocation tool under ISAE 3000 (Revised). The validation used a materiality threshold of plus or minus 5% for aggregate errors in sampled data and reviewed the methodology, data-collection and consolidation processes and AWS information systems. The assurance did not cover the underlying AWS total water-withdrawal figure from which customer totals were allocated.
The withdrawal metric covers all water drawn from sources including municipal supplies, groundwater and reclaimed water, regardless of how much is later returned. Under GRI 303, water consumption can be calculated as total withdrawal minus total discharge when consumption is not measured directly. AWS provides customer allocations for total water withdrawals, not consumption.
Those limits narrow how companies can use the figures in corporate water disclosures. GRI 303 requires organizations using the standard to report withdrawal, discharge and consumption separately.
The standard’s withdrawal disclosure requires totals by source and separate totals for areas with water stress, while facility-level source breakdowns in water-stressed areas are recommended. AWS’s customer dataset provides total-withdrawal allocations by account, Region and service, but no fields for water source, discharge, consumption or individual facilities.
Global water metrics and sustainability goals
AWS reported that its global leased, owned and shared data centers withdrew 9.4 billion liters of water in 2025. For its leased and owned data centers, 22% of withdrawals occurred in regions classified as experiencing high or extremely high water stress.
The company also reported a water use effectiveness of 0.12 liters withdrawn per kilowatt-hour of IT load and said it was 75% of the way toward becoming water positive across its global direct data center operations by 2030.
The service applies AWS’s allocation model to site-level withdrawal data to estimate each customer’s share. The resulting figures are annual allocations governed by AWS’s facility boundary and methodology, not measurements of water drawn by individual workloads.