Tailpipe calculates embodied manufacture emissions based on a methodology developed by Boavizta, a French not-for-profit association dedicated to measuring the environmental impacts of cloud computing. Tailpipe is an active member of Boavizta and contributes to their organizational needs and their datasets. One of the greatest strengths of Boavizta’s methodology is its bottom-up approach to calculating the embodied GWP impacts of cloud services. This methodology calculates virtual machine impacts from the component level. Boavizta draws its data from a variety of independent sources and is therefore a more reliable source itself than the datasets produced by hardware manufacturers or cloud providers.
The bottom-up approach works as follows:
Tailpipe gets the customer’s virtual machine type from their usage and billing data. It matches this to the bare metal server hosting the virtual machine.
- The embodied CO2e of the individual hardware components associated with the bare metal server are calculated.
- This generates an embodied CO2e figure for the bare metal server.
- The embodied CO2e figure for the specific virtual machine is calculated based on the ratio of the virtual machine to the bare metal server.
Boavizta’s own justification of its methodology can be found in their API Documentation. Tailpipe supports Boavizta’s methodology as the most comprehensive approach to calculating the embodied carbon emissions of cloud computing. The methodology provides a detailed assessment of each component’s contribution to emissions, based on a wide range of independent data sources. It also integrates a sophisticated calculation that accounts for the physical to virtual ratio of the hardware to instance relationship. Tailpipe then adds its own methodology for calculating networking-related equipment to deliver a comprehensive calculation of the embodied emissions of a virtual machine’s hardware.
Tailpipe’s implementation of the Boavizta methodology is detailed here:
1. Calculating the embodied CO2e of hardware components
First, the components of the bare metal server associated with the virtual machine are identified. A bare metal server is a physical server that can only be utilized by one customer at a time. Each virtual cloud machine runs on bare metal hardware. This means each virtual machine utilizes a portion of an underlying host hardware with the same characteristics as the associated bare metal server (such as the a1.medium instance and the a1.metal bare metal server, as AWS examples).
Then, the embodied CO2e of the following server hardware components are calculated: CPU, RAM, storage equipment (SSD, HDD, or network storage), motherboard, power supply unit, network equipment, GPU, and enclosure (case). Values for the assembly of the server are also integrated.
Where CPU:
CPU Hardware = (CPU Die Size) * CPU Die Impact + CPU Base Impact
Where RAM:
RAM Hardware = (RAM Capacity/RAM Density) * RAM Die Impact + RAM Base Impact
Where SSD:
SSD Hardware = (SSD Capacity/SSD Density) * SSD Die Impact + SSD Base Impact
Where power supply:
Power Supply Unit Hardware = Power Supply Unit Weight / Power Supply Unit Weight Impact
Where enclosure (if blade):
Enclosure Hardware = Blade Impact + (Blade Enclosure Impact/16)
Values for HDD, motherboard, assembly and enclosure (if rack) are always constant.
Using the inputted virtual machine type, variable data for the physical characteristics of hardware components is drawn from Tailpipe’s central lookup table. Tailpipe has collated a central database that details every virtual machine hosted by in-scope cloud providers, along with their associated variable components as listed above. Tailpipe ensures a high-quality level of data by drawing on official sources, such as CSP specifications, manufacturer blueprints, and TechPowerUp. See Tailpipe’s Data Quality Dashboard for more information on our data sources.
Constant data for the CO2e impact of each hardware component comes from the German Environment Agency’s comprehensive 2021 study, Green Cloud Computing (GCC). The component impact figures (constants) are listed in the table below:
| CPU Die Impact | 1.97 | kgCO2e /cm2 |
| CPU Base Impact | 9.14 | kgCO2e |
| RAM Die Impact | 2.20 | kgCO2e /cm2 |
| RAM Base Impact | 5.22 | kgCO2e |
| SSD Die Impact | 2.20 | kgCO2e /cm2 |
| SSD Base Impact | 6.34 | kgCO2e |
| HDD Impact | 31.11 | kgCO2e |
| Motherboard Impact | 66.10 | kgCO2e |
| Power Supply Unit Weight Impact | 24.30 | kgCO2e/kg |
| Assembly Impact | 6.68 | kgCO2e |
| Rack Impact | 150 | kgCO2e |
| Blade Impact | 30.90 | kgCO2e |
| Blade Enclosure Impact | 880 | kgCO2e |
| Network Storage Impact | 28.7 | kgCO2e |
| Switch Impact | 243 | kgCO2e |
| GPU Impact | 164 | kgCO2e |
Tailpipe then multiplies each of the component embodied figures by the quantity of component in the server.
CPU Units * CPU Hardware
RAM Units * RAM Hardware
SSD Units * SSD Hardware
HDD Units * HDD Hardware
1 * Motherboard Hardware
Power Supply Unit Units * Power Supply Unit Hardware
1 * Assembly Hardware
1 * Enclosure Hardware
The figures for component units are again drawn from Tailpipe’s central lookup table.
The methodology containing all component impact figures (constants) is as follows:
CPU Server = CPU Units * ((CPU Die Size) * 0.0197 + 9.14)
RAM Server = RAM Units * ((RAM Capacity/RAM Density) * 2.2 + 5.22)
SSD Server = SSD Units * ((SSD Capacity/SSD Density) * 2.2 + 6.34)
HDD Server = HDD Units * 31.11
Motherboard Server = 66.10
Power Supply Server = Power Supply Unit Units * (Power Supply Unit Weight * 24.3)
Assembly Server = 6.68
Enclosure Server = 150 (if rack) OR 1,699.5 (if blade)
3. Calculating the embodied CO2e figure for the specific instance
Tailpipe then calculates the embodied CO2e figure for the specific virtual machine (VM) using the following formulae:
CPU = CPU Server * (vCPU VM/Server CPU Threads)
RAM = RAM Server * (Memory VM/Memory Server)
SSD = SSD Server * (Storage VM/Storage Server)
HDD = HDD Server * (Storage VM/Storage Server)
For all other components:
Component = Component Server * (vCPU VM/Server CPU Threads)
vCPU stands for virtual central processing unit. It represents the share of the underlying physical processing unit assigned to the virtual server or machine. Variable data (virtual machine memory, storage, vCPU) for these calculations is again drawn from Tailpipe’s central lookup table.
Tailpipe adds to the Boavizta methodology by also calculating for the networking-related equipment that comes with a virtual machine. This covers network storage and network switches. Network storage provides an organization with a storage volume in a dedicated storage server. This is different to onboard storage, which is processed by SSDs or HDDs associated with a virtual machine. This means Tailpipe must calculate the embodied impacts of network storage differently to onboard storage. Organizations will either have onboard (SSD or HDD) or network storage. Network storage is calculated using the following formula:
Network Storage = Network Storage Units (GB) * 0.0013
This calculation is based on the embodied emissions in a Seagate Exos X22 Hard Drive. The most recent insight into the storage systems used by Cloud Service Providers comes from AWS, which stated in 2024 that it uses the Seagate Exos X System, with a maximum disk capacity of 24 TB (24,000 GB). Seagate publishes Life Cycle Analyses, including embodied carbon figures, of all of its Exos drives, apart from the maximum capacity X24 (24TB) drive. Therefore, Tailpipe uses the embodied emissions figure of the next largest drive, the X22 (22TB). The X22 contains 28.7 kgCO2e of embodied emissions, which Tailpipe divides by 20,000 to get a per GB figure of 0.0013 kgCO2e/GB.
A switch facilitates communication between devices. In a data center, switches are needed to transfer data between servers and the internet. GCC calculates that a switch contains 243 kg CO2e of embodied emissions, based on a 48-port model (Edgecore’s AS7326-56X). This works out at 5.06 kg CO2e per port. The number of ports servicing each organization’s networking needs is unknown, so Tailpipe assumes two connections. Tailpipe therefore adds a figure of 10.12 kg CO2e (2*5.06) to each instance to account for the embodied emissions of network switch hardware.
Finally, Tailpipe factors in the embodied impact of GPUs. For virtual machines that utilize NVIDIA GPUs, Tailpipe adds a factor of 164 kgCO2e, based on NVIDIA’s life cycle assessment of their H100 GPUs. NVIDIA supplies the hardware for the majority of the GPU-enabled virtual machines across cloud platforms, but because no other manufacturer releases data on the embodied emissions of their GPUs, Tailpipe omits this factor for virtual machines that utilize non-NVIDIA GPUs.
The embedded virtual machine figures for each component are then added together to get the embedded GWP of the virtual machine:
Total Virtual Machine Embodied Emissions = CPU + RAM + SSD + HDD + Motherboard + Power Supply Units + Assembly + Enclosure + Network Storage + Switch
