This dashboard displays the data sources that Tailpipe uses to generate its carbon emissions calculations. In the interest of transparency, Tailpipe has scored each source on its accuracy, completeness, credibility, conciseness, latency, and plausibility. Tailpipe uses the most up to date and valid data sources available, and has highlighted any improvements to the current data set that it is working to implement in the To Do column.

Key

Quality Dimensions

Accuracy

● Valid, replicable, and reliable.
● Some gaps, but margins for error identified.
● Gaps, and margins for error cannot be identified.

Completeness

● Fulfills all parameters.
● Does not fulfill all parameters, but margins identified.
● Does not fulfill all parameters, margins unidentified.

Credibility

● Recognized source, published to scientific standards or from CSP.
● Recognized but not published to scientific standards.
● Not recognized and cannot be traced.

Latency

● Methodology: <2 yrs. Variable: <1 month.
● Methodology: 2–5 yrs. Variable: <2 months.
● Methodology: >5 yrs. Variable: >2 months.

Plausibility

● Aligns with Tailpipe's reasonable assumptions.
● Sits outside Tailpipe's reasonable assumptions.
● Exceeds Tailpipe's assumptions significantly.

Amazon Web Services

Updated: 22/04/26

AWS Operational Emissions Data Sources

Category	Source	Acc	Comp	Cred	Lat	Plaus	Improvement	Notes
CPU Utilisation	Cloudwatch	●	●	●	●	●		Customer must pay for 5 minute updates when making more than 1 million API requests.
Billing & Usage	Cost and Usage Reports	●	●	●	●	●		AWS retroactively changes the contents of the CUR in line with new data.
CPU Power	Tailpipe's CPU lab-testing (See Processors Methodology)	●	●	●	●	●	Test an even wider pool of CPUs.	Stress tests run on 12 CPUs across 37 instance configurations.
GPU Power	Tailpipe's GPU lab-testing (See Appendix A)	●	●	●	●	●	Two methodologies to distinguish between users with GPU-tracking enabled, and those without.	Stress tests run on AMD and NVIDIA GPUs.
RAM Power	Tailpipe's RAM stress testing (See Memory Methodology)	●	●	●	●	●	Test an even wider pool of RAM configurations.	Stress tests run on RAM across 37 instance configurations.
SSD Power	13 drives with capacities from 0.5TB to 8TB (See Appendix B)	●	●	●	●	●	Assess a wider range of SSD datasheets.
HDD Power	71 drives with capacities from 1TB to 24TB (See Appendix B)	●	●	●	●	●
Motherboard Power	Institute of Electrical and Electronics Engineers, 2021	●	●	●	●	●	Source a more recent datacenter power consumption study.	Find a published study on maximum motherboard power consumption in datacenter servers.
Network Storage Power	AFCOM, 2025 and Seagate, 2024	●	●	●	●	●		Based on extrapolations from existing data.
Intra- and Inter-region Networking	Ficher et al., 2021	●	●	●	●	●	Find an inter-region specific study.	Research paper only analyzed networking within one country.
External Networking	Collected research papers, 2014-2021 (See Appendix E)	●	●	●	●	●	Find a more recent study to validate these results.
Power Supply Efficiency Factor	2017 GeekWire Article and PSU manufacturers	●	●	●	●	●	Replace 2017 GeekWire reference with current manufacturer documentation.	2017 GeekWire article is outdated but manufacturer data provides credibility.
PUE	2024 Amazon Sustainability Report — AWS Summary	●	●	●	●	●
Grid Mix - Non-UK	Ember	●	●	●	●	●	Waiting for Ember to update their dataset to daily or hourly measurements.	1 month latency for every country. Unknown which datacenters generate their own power
Grid Mix - UK	National Energy System Operator (UK National Grid)	●	●	●	●	●

AWS Embodied Emissions Data Sources

Category	Source	Acc	Comp	Cred	Lat	Plaus	Improvement	Notes
Constant hardware GWP values	Green Cloud Computing report, German Environment Agency	●	●	●	●	●	Waiting for German Environment Agency to confirm whether the report will be updated in future.	Published in 2021, hardware data potentially becoming outdated.
Variable hardware figures (units, capacity, weight, case type)	Boavizta database and AWS specifications	●	●	●	●	●	Add instance data through testing or available data sources.	Crowdsourced data backed up by CSP specifications.
CPU die size	Boavizta database	●	●	●	●	●	Build an independent dataset based on manufacturer data.	Crowdsourced data.
RAM density	TechInsights and Semiconductor Digest	●	●	●	●	●	Find more data.	Independent analysis of RAM hardware.
SSD density	Researcher analysis of hardware	●	●	●	●	●	Find more data.	Independent analysis of SSD hardware.
Variable virtual machine figures (instance memory, storage, vCPU)	AWS instance specifications	●	●	●	●	●
Virtual machine characteristics	Boavizta API (originally based on Teads Engineering's 2021 carbon footprint estimator)	●	●	●	●	●		Originally based on Teads Engineering's 2021 EC2 carbon footprint estimator. Data now maintained in the Boavizta database.
Network Storage Equipment	Seagate Exos X22 Life Cycle Analysis	●	●	●	●	●	Find an LCA for the X24 hard drive.

Other AWS Data Sources

Category	Source	Acc	Comp	Cred	Lat	Plaus	Improvement
Kubernetes hardware	ScaleFactory and Tailpipe's CPU stress testing	●	●	●	●	●	Get confirmation from CSP about which CPUs are being used; find information on other hardware components.
Serverless hardware	ScaleFactory and Tailpipe's CPU stress testing	●	●	●	●	●	Get confirmation from CSP about which CPUs are being used; find information on other hardware components.
Emissions Equivalent Miles Driven	DEFRA Conversion Factors, 2025	●	●	●	●	●
Emissions Equivalent Cups of Tea Brewed	Clever Carbon	●	●	●	●	●

Microsoft Azure