IT operator infrastructure strategies depend on cloud services for some or all of their IT operations. Meanwhile, public cloud use will grow as AI adoption expands to augment and drive enterprise and government IT workloads. And so, operators should account for and establish reduction goals for the greenhouse gas emissions associated with these operations.
Operators have this responsibility because they select the location(s) where they will deploy workloads and the type, size, and work capacity of CPU cores, GPUs, memory modules, and storage devices on which they are hosted.
They also control infrastructure utilisation through the management strategies and operating systems under which they deploy their workloads. The efficiency of a public cloud datacentre, as measured by a metric based on work delivered per unit of energy consumed, depends on how efficiently users deploy and execute their workloads.
To optimise this metric, users must operate workloads as efficiently as possible, and minimise energy use and GHG emissions. While emissions are accounted for in Scope 3, operators can calculate reasonable GHG emissions estimates, and in some cases energy use estimates, by using calculators offered by cloud providers or open source tools such as Cloud Carbon Footprint. The ability to develop a reasonable estimate of these values enables operators to take responsibility for and set energy use and carbon emissions reduction goals for their operations.
Cloud providers should deliver highly efficient facilities and IT infrastructure where customers can deploy and run their IT workloads. They should report the emissions associated with these operations as Scope 2 and pass those emissions to their customers as Scope 3. They are also responsible for Scope 3 emissions associated with the facility and IT infrastructure materials and equipment (controlled by the cloud provider), and for planning their eventual reduction (in collaboration with their suppliers).
Given this complex interaction between IT operators and their cloud service providers, many operators believe cloud providers are the party responsible for accounting for and reducing GHG emissions. In reality, IT operators play an important role in efficiently operating IT infrastructure and taking steps to minimise energy use and emissions through site selection and workload placement, while recognising the geographical limits imposed by response time and reliability requirements.
Reduction of operational emissions
Each operator should set utilisation and operational goals that minimise IT infrastructure resource use and energy use and reduce associated carbon emissions. One goal should be a three- to five-year energy-use and emissions reduction target that supports achieving a net-zero-emissions cloud infrastructure by a specific future date.
To achieve the necessary emissions reductions, a strategy must be established with three key components:
Minimise the energy use of current operations,
Optimise the operation of new workloads at inception, and;
Locate the workloads at the most carbon-efficient location(s) that meet the workload response time and reliability needs.
The discussion of GHG emissions management is facility-focused and is based on location-based, not market-based, accounting. An IT operator should focus on decarbonising the energy supply that powers their infrastructure without relying on offsets.
Minimise the emissions of existing operations
Many operators select cloud infrastructure capability and capacity to match the expected demand and operational requirements of their IT workloads. Once they are deployed, they track peak loads, response time, and other key operational metrics to ensure customer needs are met while ignoring utilisation metrics.
Actively managing infrastructure capacity and utilisation can have huge returns.
Discussions with companies that offer capacity and utilisation tracking and optimisation services indicate that operators who right-size CPU and memory infrastructure and maximise its utilisation can achieve significant cost savings and reduce energy use and associated carbon emissions by up to 40%. This strategy offers a direct path to reductions that will make significant contributions toward an interim carbon reduction or net zero goal.
Energy savings and emissions reductions can also be achieved by deploying workloads to next-generation infrastructure. An IT operator should regularly review their cloud provider’s infrastructure offerings and move to more efficient work-per-watt systems as they are deployed and where it makes business sense.
These operational reductions can be documented and validated by capturing before-and-after measurements of energy use and emissions quantities. Documentation should include changes in purchased infrastructure capacity and before-and-after energy use (where available) and emissions quantities.
Optimise infrastructure for new workload deployments
IT operators should have a standard workload deployment process that includes pre-testing workloads to determine the CPU-to-memory ratio that optimises infrastructure performance and utilisation while minimising energy use. The process should also periodically recheck that ratio and infrastructure utilisation to reallocate workloads and infrastructure to maintain performance and efficiency. The process to document and validate reductions described above should be used when re-allocation is performed.
Operators should catalogue and publicise energy and emissions reductions in their annual environmental report. Reductions are reported on a project basis and can be contextualised as percentage reductions against the cloud operation’s current annual total or baseline year energy use and emissions. A baseline is not necessarily required, as the ultimate goal is to power the infrastructure with carbon-free energy (CFE).
Utilising carbon-free energy
Datacentre systems run on electrical power. So, the final step in decarbonising public cloud operations requires that hosting datacentre(s) maximise CFE consumption and have a plan to increase the percentage of CFE used over time.
Site selection is a critical step in managing GHG emissions from public cloud deployments. Each facility has an associated electricity emissions factor based on the contracted, on-site, and grid-based generation that supplies energy to the site. That factor will change monthly (the typical emissions reporting period for public cloud providers) based on the season, weather conditions, and grid performance. For a given geographical area, some facilities may have a lower emissions factor and a better plan/outlook for further reduction. The facility emissions factor should be part of the procurement criteria, though other factors like IT performance and cost are likely to predominate in facility selection.
Operators can consider carbon-aware workload placement software for non-critical workloads or time-insensitive batch jobs. Most available software systems are in early-stage development, not widely proven, and not widely deployed. These systems offer a means to achieve further emissions reductions, but their immaturity introduces unattractive operational risk and limits their usefulness.
Managing material and equipment embedded emissions
Public cloud service providers are responsible for design, construction, and material and equipment procurement for their datacentres. They set procurement specifications and can collaborate on developing low-carbon materials, such as concrete and steel. While many initiatives are underway to develop low-carbon materials, to achieve volume production and incorporate them into equipment is a decades-long process. Because they hold relationships with suppliers, cloud providers should account for, report, and manage reduction efforts with their suppliers for these materials.
Accounting for these Scope 3 emissions offers IT operators little actionable value. They do not control facility procurement or supplier decisions, and providers often report emissions before operators contract for infrastructure. While some individuals and groups have advocated for “depreciating” Scope 3 emissions, this practice gives operators no actionable data and adds complexity that only confuses emissions reporting.
Reducing cloud emissions will be a journey
Datacentres depend on electrical power. In almost all cases, an electricity grid carries a carbon emissions factor. Only two major geographical regions, Latin America (65%) and Europe (71%), supply more than 50% CFE in their grid networks. The Middle East has the lowest percentage at 8%. Decarbonising datacentre operations will require continuing efforts to develop and deploy CFE generation sources and energy storage systems.
Reducing public cloud emissions is a shared responsibility. IT operators must act on the operations they control. That means right-sizing IT resources, improving resource utilisation, and selecting locations supplied by low carbon electricity where performance and reliability constraints allow.
Cloud providers must improve infrastructure efficiency, decarbonise electricity supplies, and work with their suppliers to address embedded emissions from materials and equipment. Emissions estimates need not be perfect to support these actions, but claimed reductions should be backed by documented and validated before-and-after energy use and GHG emissions estimates.
The priority is to connect carbon accounting to operational decisions, rather than treat reporting as an end in itself. Operators should establish reduction goals, validate and report project-level results for their IT infrastructure, and assess and support cloud providers’ progress toward carbon-free energy and low-carbon materials. Focusing on these practical steps gives operators a path to reducing the GHG emissions associated with their cloud operations.
Jay Dietrich is research director for energy and sustainability at the Uptime Institute
