Cloud Carbon Coefficients

Compute and infrastructure coefficients

What each coefficient covers

CPU, watts per vCPU

SPECpower measurements of whole servers, at idle and at full load. Calculators show both ends because they do not know your load. These are scenarios, not confidence bounds: Cloud Carbon Footprint itself uses a 50% load point estimate with no confidence interval, and the actual figure can fall outside the range.

Memory, 0.392 W per GB

On AWS and Azure the server watts already include the memory those servers carried, so only memory beyond it is added: beyond 0.84 GB per vCPU on AWS and 0.84 on Azure (Cloud Carbon Footprint’s 80.69 and 80.47 GB per chip, at most 96 vCPUs a chip). On Google Cloud all allocated memory is added, as Cloud Carbon Footprint does with Google’s separate memory billing.

Network, 0.001 kWh per GB

Data moved between data centers or regions, over the providers’ own fiber. Internet egress to end users is outside this coefficient and outside these estimates, as in Cloud Carbon Footprint.

Storage, Wh per terabyte-hour

Allocated capacity, not the share in use, times the copies the service keeps. Copies counted on this site, from Cloud Carbon Footprint’s replication factors: AWS EBS 2, S3 6; Azure managed disks 3, LRS storage 3; Google zonal disks 1, Cloud Storage 2. Databases with high availability: RDS Multi-AZ 2, Azure SQL Database 3, Cloud SQL 2.

Not included

Manufacturing (embodied) emissions, except as an optional line in the Cloud Carbon Calculator, and anything beyond the data center. These are operational estimates, not a complete lifecycle footprint.

Sources and vintage

Retrieved is when our script fetched a figure. Evidence is how old the data behind it is, which is often much older.

Google Cloud grid intensity by region (46 regions)

Google is the only provider that publishes this per region. Grid intensity is what the local grid emits per kilowatt-hour, location-based, and it is the figure the calculators use. Carbon-free energy percentage is the hourly share of Google’s use in the region that is carbon-free, counting both the grid’s own carbon-free supply and the clean energy Google buys there, shown for reference. Across these regions grid intensity runs from 11 g (Montréal) to 641 g (Johannesburg), about 58 times.

AWS and Azure regions: the grid figure we apply (95 regions)

AWS and Azure publish no per-region grid figure. Each region takes Google’s 2025 figure for the same grid or, where Google has no region on it, Ember’s national average. These are regional estimates, not provider data, and the basis column says which kind each one is. Google’s figures, and every region that takes one, count operational emissions: what the grid emits generating the power. Ember’s country averages are lifecycle figures: they also count fuel supply and building the plants, so they run higher for the same grid. The tools therefore rank country averages only among themselves, never against Google’s figures. Cloud Carbon Footprint publishes its own AWS and Azure factors (EPA eGRID by US grid region, EEA, carbonfootprint.com); these more recent figures are used instead.

Use these instead, where they apply

• AWS Sustainability

  Free, native, and authoritative for AWS. Emissions and water, by Region, service and scope.

• Google Cloud Carbon Footprint

  Free, and exports to BigQuery. Google also publishes per-region CFE%, which this page uses.

• Azure Carbon Optimization

  Free and in-portal, and issues recommendations rather than only measurements.

• EcoLogits (GenAI Impact)

  Per-model LLM inference emissions, operational and embodied, with a published methodology.

• Cloud Carbon Footprint

  The open methodology the energy coefficients on this page come from. Grid figures come from Google and Ember.