Upstream Emissions

Upstream emissions (Category U) refer to the embodied carbon emissions of hardware purchased by an organisation and the carbon emissions of the development and distribution of installed software used by the organisation.

Understanding upstream emissions allows organisations to comprehensively assess the environmental impact of their assets and supply chains. It enables the development of strategies to reduce the carbon footprint of materials and processes, promoting sustainable procurement practices and circular economy principles.

Upstream emissions are related to GHG Protocol Scope 3.

Software

Emissions associated with developing and delivering off-the-shelf and open-source software installed on the organisation’s systems. Includes emissions related to:

• Energy consumed and hardware used during programming, testing and releasing new software versions.
• Packaging and digital distribution of software over the internet.
• Product support services for troubleshooting and guiding customers.

Hardware

Embodied carbon emissions associated with hardware devices owned by an organisation, including emissions from the manufacture, transportation, installation, maintenance, and end-of-life of a device.

Read more about embodied carbon emissions.

Employee Hardware

Laptops, desktops, mobiles, printers, and peripherals used by employees.

Click here to see a worked example of estimating embodied emissions for a laptop.

Networking Hardware

When considering the upstream emissions of a network, consider the embodied carbon of any networking devices that are owned by the organisation. These include, but are not limited to:

• routers
• switches
• bridges
• Wi-Fi access points
• firewalls
• modems
• hubs
• repeaters
• cables

Ideally, data sheets from the manufacturer should be used to gather this data.

Data Centre and Server Hardware

Servers, storage systems, and data centre infrastructure installed on-premise.

Click here to see a worked example of estimating embodied emissions for a server.

Content

Embodied carbon emissions associated with producing, distributing and storing digital content.

Foundation Models

AI workloads rely on specialised hardware and AI accelerators to perform complex computations at high speed. When assessing the upstream emissions of foundation models, it’s essential to account for the embodied carbon of the IT infrastructure involved in their training, deployment and with inference. This includes the carbon footprint associated with manufacturing specialised hardware and constructing and operating AI data centres. Key components include, but are not limited to:

• High-Performance Computing systems (HPC)
• Graphics Processing Units (GPUs)
• Central Processing Units (CPUs)
• Tensor Processing Units (TPUs)
• Memory (RAM)
• Storage
• Networking equipment
• Advanced cooling and power supply
• Distributed computing

Content and Data

When considering the upstream emissions of content and data, whether text, image, audio or video used as a commodity by the organisation, evaluate emissions that relate to:

• Content production
  • Energy used to produce content, including pre-production (such as research and planning), filming, recording, editing and any travel involved.
• Processing
  • Data processing and AI/ML workflows, if content is algorithmically recommended, auto-captioned or translated for example.
• Storage
  • Energy consumption and hardware required for storing content and data.
• Distribution
  • Digital distribution of content over the internet.
  • Network infrastructure such as content delivery networks (CDNs).
• Archiving
  • Energy used to migrate, compress, and maintain historical content and datasets.