By 2035, IDTechEx forecasts that the continued growth of artificial intelligence will result in over 2000 TWh of energy being consumed by data centres.
By Eve Pope, technology analyst at IDTechEx
The expected increase in CO2 emissions from powering these new data centres creates a dual challenge for both governments with net-zero targets and tech giants with internal carbon neutrality pledges alike: how can the unprecedented rise of AI be supported without causing major harm to the environment?
The new report published by IDTechEx, “Sustainability for Data Centres 2025-2035: Green Technologies, Market Forecasts, and Players”, characterises green data centre technologies, players, and markets, focusing on cost and commercial viability.
Rapid growth in the data centre construction is starting to stretch grid capacity to its limits in some regions. To expand in a way aligned with sustainability goals, data centre hyperscalers are increasingly playing a more active role in bringing new renewable energy projects online beyond standard power purchasing agreements (PPAs) and renewable energy certificates (RECs). For example, early microgrid projects exploring on-site off-grid power generation for data centres are emerging.
Wind and solar power have long been favored by data centre players due to a low LCOE (levelised cost of electricity) that is often below fossil fuel alternatives. However, the intermittency of these renewables means fossil fuel power generation is still needed for many hours per day. There is building momentum to change the way scope 2 power emissions are accounted for under the GHG Protocol to favor hourly time-based energy matching so that demand side signals can be sent for emerging low-carbon energy technologies to better facilitate the continuous power demand of data centres.
Existing policies surrounding data centre decarbonisation, such as the EU Energy Efficiency Directive, relate to the energy efficiency (PUE – power use efficiency) of data centres. If less power can be consumed per data centre through improved thermal efficiency, electrical efficiency, and IT efficiency, the environmental impact is minimized. Therefore, from purpose-built chips, memory modules, to cooling components and AC/DC converters, data centre players are racing to enhance energy efficiency.
As the data centre sector transfers over from traditional air cooling to direct-to-chip liquid cooling, bringing reductions in greenhouse gas emissions, water usage, and energy consumption, tradeoffs in other metrics, such as cost and complexity, must be considered. Further, with the constant emphasis on energy efficiency from leading component suppliers such as Nvidia, AMD, SK Hynix, and Infineon, the energy efficiency on the componentry level (e.g., GPUs, CPUs, memory modules, power converters, etc.) is also seeing growth.
Typically, scope 3 emissions represent the majority of CO2 emissions from data centres. Key factors contributing to scope 3 emissions include upstream manufacturing/assembly of servers and networking equipment used in data centres and emissions related to data centre construction. In 2023, Microsoft’s Scope 3 emissions were 30,9% higher than in 2020, which was attributed to the embodied carbon in building materials, as well as hardware components such as semiconductors, servers, and racks.
Because Scope 3 emissions are indirect emissions in a company’s value chain that are not caused by the company itself, it can be hard for data centre players to tackle scope 3 emissions. IDTechEx explores three different ways for companies to reduce scope 3 emissions in its report: purchasing carbon credits (specifically carbon removal credits) to counteract hard-to-avoid CO2 emissions; using low-carbon materials in data centre construction (green concrete, green steel, and timber) either physically or through attribute purchases (book and claim); and choosing IT hardware with lower embodied/manufacturing carbon over the lifetime of a data centre.