Embodied emissions are the greenhouse gas emissions associated with the creation and disposal of devices (Green Software Foundation). This covers the manufacture, distribution, and end-of-life processes of products. In the case of cloud computing, embodied emissions are present in the physical hardware that hosts cloud services and other information and communications technology (ICT) equipment. They account for an average of 20-30% of the total emissions of ICT (Freitag et al., 2021). The processes that go into creating and hosting cloud services generate a range of greenhouse gases beyond CO2, such as SF6, NF3, CF4, and CHF3. These have differing effects on global warming, which are accounted for by a unified measure of global warming potential (GWP) called CO2e. This documentation explains how Tailpipe calculates the embodied emissions of cloud services through the virtual machines and bare metal instances used to run computing workloads.
The basic formula that Tailpipe uses to calculate embodied emissions is:
M = TE * TS * RS
Where:
M = Embodied Emissions
TE = Total Embodied Emissions
TS = Time Share (the share of the total life span of the hardware reserved for use by the software)
RS = Resource Share (the share of the total available resources of the hardware reserved for use by the software)
This formula was developed by the Green Software Foundation for its Software Carbon Intensity (SCI) Specification. The SCI Specification underpins the ISO 21031 specification as a globally recognized standard for calculating software’s carbon emissions.
Tailpipe further breaks down TE into two categories: manufacture and disposal.
M = (Man + Disp) * TS * RS
For example, let’s consider a customer who is using one cloud instance 50% of the time for one year. The instance type they are using has a lifespan of 6 years and generated 100 kg CO2e in its hardware components. Their cloud provider recycles a portion of its data center hardware, saving 2 kg CO2e in the disposal process.
In this case, M = (100 – 2) * 0.08 * 1
So, through their cloud computing, this customer has produced 7.84 kg CO2e of embodied emissions.
Tailpipe draws on two primary methodologies and datasets to calculate TE:
- Boavizta for manufacture. Boavizta is a French not-for-profit association dedicated to measuring the environmental impacts of cloud computing. They provide a methodology for calculating the embodied emissions of the hardware that hosts cloud services.
- Dell for disposal. Dell Corporation is a global manufacturer of PC and server hardware. They have put considerable effort into understanding the carbon footprint associated with the manufacture of their own-brand servers. It is their data that Tailpipe uses for emissions from computing hardware.
Tailpipe Overall Embodied Emissions Methodology
Tailpipe calculates the kilograms of CO2e embodied in an organization’s cloud computing using the following formula:
(((CPU + RAM + SSD + HDD + Motherboard + Power Supply Units + Assembly + Enclosure + Network Storage + Switches) * Disposal) * Time Share * Resource Share)
Time Share
Tailpipe designates embodied emissions based on customers’ usage. For AWS, this data is gathered from the Cost and Usage Report (CUR) files. For Azure, this is the Command Line Interface (CLI) and account invoices. These files are generated by the CSP to show their customers which cloud services they have used and what they will be billed for.
The CUR files and CLI provide the start and end date of each instance’s utilization. This time span is divided by the instance’s total lifetime. In 2022, Azure announced that their servers had a lifespan of 6 years. In AWS’ 2023 Sustainability Report, they also state that the average lifetime of AWS servers is six years. Because it is impossible to measure the actual lifetime of each server associated with each customer’s instance, Tailpipe assumes a six-year lifespan as a constant in its time share calculation.
Resource Share
In all cases, Resource Share is 1. This is because Tailpipe calculates emissions at the virtual machine level, multiplied by the unique time share allocated to each individual virtual machine. If a customer is utilizing multiple virtual machines in a cluster, the separate GWP values associated with each virtual machine and their individual time shares can be added together to provide a customer specific GWP.
Data Quality
Tailpipe draws its data from several recognized organizations. Data on instance and server characteristics (hardware type, specifications, quantity of components, memory and storage) is drawn from cloud provider specifications. Where this data is not available through cloud providers, Tailpipe uses data from Boavizta in collaboration with Teads Engineering. Constant data (GWP impact factors) is leveraged from Green Cloud Computing, the German Environment Agency’s detailed research into hardware emissions. Disposal data is drawn from Dell’s R740 Server Life Cycle Analysis.
Accounting for cloud computing emissions is not yet a well-established practice, and the data available is therefore often limited. Tailpipe is confident in the data that it utilizes, whilst being committed to regularly assessing its quality in search of better sources.
By creating Tailpipe, the available pool of data on cloud computing emissions has been expanded. Tailpipe hopes to motivate cloud platforms to take cloud computing carbon accounting seriously.
For a detailed breakdown of Tailpipe’s data sources, view the Data Quality Dashboard.
