Cloud Carbon Footprint Calculator
Estimate what your cloud workload emits, and how much of it region choice could remove
Where it runs
The highest-leverage choice on this page.
Monthly usage
One always-on 4-vCPU instance is about 2,920.
An idle server still draws roughly a fifth of its peak.
Leave at 0 if you run no accelerators.
Accounting
Used only for the market-based figure.
Leave at 0 to use the provider default.
Your workload
The same workload, every region
Cleanest first. Yours is highlighted.
Where the energy goes
Cooling and power distribution appear as their own line rather than being folded silently into the total.
Related Calculators
Corporate Carbon Emissions Calculator
Calculate Scope 1, 2, and 3 greenhouse gas emissions per GHG Protocol.
EnvironmentalCarbon Offset Cost Calculator
Estimate offset cost by project type, from cheap avoidance to durable removal.
EnergyEnergy Efficiency ROI Calculator
Calculate return on investment for energy efficiency upgrades.
Region beats optimisation
Engineering effort spent making a service twenty percent more efficient is worthwhile. Moving it from a coal-heavy region to a hydro-powered one can cut its emissions by ninety percent, and takes a config change and a migration window.
That is the uncomfortable finding this calculator keeps producing, and it is why the region chart is the centrepiece rather than a footnote. Latency, data residency and cost all constrain the choice in practice — but the carbon difference is large enough that it deserves to be on the table explicitly rather than defaulting to whichever region someone picked years ago.
What this deliberately does not include
Being clear about the boundary is what separates a credible estimate from a number that merely looks precise:
- Embodied emissions. Manufacturing the servers, network gear and buildings is excluded, and it can be a substantial share of a modern efficient data centre's lifecycle footprint.
- Over-provisioning. It counts the capacity you enter, not idle instances you are billed for and never use. In many estates that gap is the single largest reduction available.
- Hourly grid variation. Annual averages cannot capture the benefit of shifting batch work to hours when the grid is clean, which on some grids is a factor of two within a single day.
- Audited PUE. Provider figures are self-reported annual fleet averages, not independently verified measurements.
Where the numbers come from
- Compute, memory, storage and network coefficients: the open-source Cloud Carbon Footprint methodology, which models a server as an idle wattage plus a utilisation-scaled increment.
- US grid intensity: EPA eGRID2023 Revision 2 subregion factors.
- Non-US grid intensity: national annual averages. These are the least certain figures on the page, and national grids vary enormously by hour in ways an annual average cannot show.
- PUE: provider-reported fleet-wide annual averages; the on-premises figure is an industry-survey average.
How to Use This Calculator
- Pick your provider and region. Region is the highest-leverage choice on this page. The same workload can differ more than tenfold between the cleanest and dirtiest regions.
- Enter your monthly compute. vCPU-hours and average CPU utilisation. Utilisation matters because a server draws power when idle too - typically about a fifth of its peak. An over-provisioned instance is not a free instance.
- Add memory, storage and egress. Memory in GB-hours, storage in terabytes provisioned, and network egress in gigabytes per month. Your bill has all three.
- Set the renewable matching. This produces the market-based figure. Providers claim to match all their consumption with renewable purchases, which is why their own dashboards show near-zero. Both numbers are reported here because the GHG Protocol requires both.
- Look at the region chart. It prices your exact workload in every available region, sorted cleanest first, with yours highlighted. For most engineering teams, that chart is the whole decision.
Frequently Asked Questions
How do you calculate the carbon footprint of cloud computing?
Estimate the electricity each resource draws, multiply by the data centre's power usage effectiveness to account for cooling and power distribution, then multiply by the grid carbon intensity where the data centre sits. The Cloud Carbon Footprint methodology models a server as a minimum idle wattage plus a utilisation-scaled increment - for AWS, roughly 0.74 W per vCPU at idle rising to about 3.5 W at full load.
Which cloud region has the lowest carbon footprint?
Among the common ones, the Nordic and French regions are cleanest thanks to hydro, wind and nuclear, while India, Australia and Singapore are the highest. The same workload can differ by more than tenfold between them, which usually makes region choice the single highest-leverage decision available to an engineering team - larger than almost any code-level optimisation.
What is the difference between location-based and market-based emissions?
Location-based uses the actual carbon intensity of the physical grid serving the data centre. Market-based subtracts renewable energy the provider has contracted for. AWS, Google and Microsoft all claim to match or exceed 100% of their consumption with renewable purchases, which is why their own dashboards report far lower numbers than an independent estimate. The GHG Protocol requires dual reporting of both, and this calculator shows both.
Does PUE still matter?
Less than it used to at hyperscalers, more than ever on-premises. The major providers report fleet-wide PUE between roughly 1.09 and 1.2, meaning nine to twenty percent overhead on top of the IT load. The industry-wide average across all data centres is closer to 1.5, and it has barely moved in years. Moving a workload from an average enterprise data centre to a hyperscaler cuts roughly a quarter of the energy before you touch a line of code.
What does this calculator not include?
Three things worth knowing. It excludes embodied emissions from manufacturing the servers, networking equipment and buildings, which can be a substantial share of a modern efficient data centre's lifecycle footprint. It counts the capacity you tell it about, not idle over-provisioned instances you are paying for but not using. And it uses annual average grid intensity rather than hourly, so it cannot capture the benefit of shifting batch work to times when the grid is clean. Provider PUE figures are also self-reported and not independently audited.