Data Centres and Data Transmission Networks

Digital technologies have direct and indirect effects on energy use and emissions, with data centres connected to electricity grids with lower shares of generation based on fossil fuel producing less associated emissions, and hold enormous potential to help (or hinder) global clean energy transitions, including through the digitalisation of the energy sector.  

The data centres and data transmission networks¹ that underpin digitalisation accounted for around 330 Mt CO2 equivalent in 2020 (including embodied emissions), equivalent to 0.9% of energy-related GHG emissions (or 0.6% of total GHG emissions). Since 2010, emissions have grown modestly despite rapidly growing demand for digital services, thanks to energy efficiency improvements, renewable energy purchases by information and communications technology (ICT) companies and broader decarbonisation of electricity grids in many regions. However, to get on track with the NZE Scenario, emissions must halve by 2030.

Estimated global data centre electricity consumption in 2022 was 240-340 TWh¹, or around 1-1.3% of global final electricity demand. This excludes energy used for cryptocurrency mining, which was estimated to be around 110 TWh in 2022, accounting for 0.4% of annual global electricity demand. 

Since 2010, data centre energy use (excluding crypto) has grown only moderately despite the strong growth in demand for data centre services, thanks in part to efficiency improvements in IT hardware and cooling and a shift away from small, inefficient enterprise data centres towards more efficient cloud and hyperscale data centres.  

Despite strong gains in efficiency, the rapid growth in workloads handled by large data centres has resulted in a substantial increase in energy use in this segment over the past several years, growing by 20-40% annually. Combined electricity use³ by Amazon, Microsoft, Google, and Meta more than doubled between 2017 and 2021, rising to around 72 TWh in 2021. Overall data centre energy use (excluding crypto) appears likely to continue growing moderately over the next few years, but longer-term trends are highly uncertain.  

Although data centre electricity consumption globally has grown only slightly, some smaller countries with expanding data centre markets are seeing rapid growth. For example, data centre electricity use in Ireland has more than tripled since 2015, accounting for 18% of total electricity consumption in 2022, and data centres and other non-industrial large energy users could account for 28% of national demand by 2031 for data centres and other non-industrial large energy users unless generation capacity increases. In Denmark, data centre energy use is projected to rise six times by 2030 to account for almost 15% of the country’s electricity use.

1 IEA analysis based on Masanet et al. (2020), Malmodin (2020), Hintemann & Hinterholzer (2022) and reported energy use data from large data centre operators.

Globally, data transmission networks consumed 260-360 TWh in 2022² , or 1-1.5% of global electricity use. Mobile networks accounted for around two-thirds of total network energy consumption. The energy efficiency of data transmission has improved rapidly over the past decade: fixed-line network energy intensity has halved every two years in developed countries, and mobile-access network energy efficiency has improved by 10-30% annually in recent years. 

Internet traffic globally was up nearly 30% in 2022, lower than the 40-50% Covid-19 pandemic-driven surge in 2020. GSMA members reported that their network data traffic increased by 31% in 2021, while total electricity use by operators rose by only 5%. Data from major European telecom network operators analysed by Lundén et al. (2022) mirror these global efficiency trends. Electricity consumption by reporting companies – representing about 36% of European subscriptions and 8% of global subscriptions – increased by only 1% between 2015 and 2018, while data traffic tripled. 

In addition, waste heat from data centres is increasingly being used to help heat nearby commercial and residential buildings or supply industrial heat users, reducing energy use from other sources. 

2 - IEA analysis based on Coroamă (2021), ITU (2020), Malmodin and Lundén (2018), Malmodin (2020) and GSMA (2022).

Strong growth in demand for data network services is expected to continue, driven primarily by data-intensive activities such as video streaming, cloud gaming and augmented and virtual reality applications. However, these data-intensive services may only have limited impacts on energy use in the near term since energy use does not increase proportionally with traffic volumes. In addition, the average energy consumption of video streaming is fairly low compared with other everyday activities, with end-user devices such as televisions consuming the majority.  

Mobile data traffic is also projected to continue growing quickly, quadrupling by 2028. 5G’s share of mobile data traffic is projected to rise to nearly 70% in 2028, up from around 17% in 2022. Although 5G networks are expected to be more energy efficient than 4G networks, the overall energy and emissions impacts of 5G are still uncertain due to their higher network density and energy consumption per site. Decommissioning of legacy networks such as 2G and 3G offers a major opportunity to reduce mobile network energy use. Managed phase-outs of legacy networks are needed to ensure existing services and devices such as smart meters can maintain functionality.  

Demand for data centre services is also poised to rise, driven in part by emerging digital technologies such as blockchain (particularly proof-of-work) and machine learning. For example, Bitcoin – the most prominent example of proof-of-work blockchain and most valuable cryptocurrency by market capitalisation – consumed an estimated 110 TWh in 2022, 20-times more than in 2016. Ethereum, second behind Bitcoin in terms of market capitalisation and energy use, consumed around 18 TWh over the first three quarters of 2022. In September 2022, Ethereum transitioned from a proof-of-work consensus mechanism to proof-of-stake, which is expected to slash energy use by 99.95%. As blockchain applications become more widespread, understanding and managing their energy use implications may become increasingly important for energy analysts and policy makers.  

Artificial intelligence (AI) and machine learning (ML) is another area of demand growth, with potentially significant implications for data centre energy use in upcoming years. While the amount of computing power needed to train the largest ML models is growing rapidly, it is unclear how quickly overall ML-related energy use in data centres is increasing.  

Early studies focused on the energy and carbon emissions associated with training large ML models, but recent data from Meta and Google indicate that the training phase only accounts for around 20–40% of overall ML-related energy use, with 60–70% for inference (the application or use of AI models) and up to 10% for model development (experimentation). Google estimates that ML accounted for 10-15% of its total energy use in 2019-2021, growing at a rate comparable with overall energy growth (+20-25% per year over the same period). 

The combination of rapidly growing size of models and computing demand are likely to outpace strong energy efficiency improvements, resulting in a net growth in total AI-related energy use in the coming years. Although AI itself can help reduce energy use in data centres, the rapid and mainstream adoption of AI chatbots like OpenAI’s ChatGPT and Google Bard are likely to accelerate growth in energy demand for AI. 

The nature of data centre demand is expected to evolve over the coming decade, with 5G, the Internet of Things and the metaverse likely to increase demand for low-latency computing, increasing demand for edge data centres. User devices such as smartphones – increasingly equipped with ML accelerators – are set to increase the use of ML with uncertain effects on overall energy demand.

ICT companies invest considerable sums in renewable energy projects to protect themselves from power price volatility, reduce their environmental impact and improve their brand reputation. Hyperscale data centre operators, in particular, lead in corporate renewable energy procurement, mainly through power purchase agreements (PPAs). Amazon, Microsoft, Meta and Google are the four largest purchasers of corporate renewable energy PPAs, having contracted almost 50 GW to date, equal to the generation capacity of Sweden.

Apple (2.8 TWh), Google (18.3 TWh), Meta (9.4 TWh) and Microsoft (13 TWh) purchased or generated enough renewable electricity to match 100% of their operational electricity consumption in 2021 (primarily in data centres). Amazon consumed 30.9 TWh (85% renewable) across their operations in 2021, with a goal of achieving 100% renewables by 2025.  

Although a few network operators have achieved 100% renewables (including BT, TIM and T-Mobile), data transmission network operators generally lag behind data centre operators in renewable energy purchase and use. Compared with data centres, which are typically large, centralised and more flexible in location, telecommunication network operators have many sites (with limited flexibility for site selection). As a result, accessing renewable energy can be a challenge in many markets, particularly in emerging and developing economies with less well-developed energy markets.

However, matching 100% of annual demand with renewable energy purchases or certificates does not mean that data centres and data transmission networks are powered exclusively by renewable sources. The variability of wind and solar sources may not match a data centre’s demand profile, and the renewable energy may be purchased from projects in a different grid or region from where demand is located. Renewable energy certificates, in particular, are unlikely to lead to additional renewable energy production, resulting in uncertainty of real-world emissions mitigation. 

Google and Microsoft have announced 2030 targets, and Iron Mountain a 2040 target, to source and match zero-carbon electricity on a 24/7 basis within each grid where demand is located. A growing number of organisations are working towards 24/7 carbon-free energy to match their electricity demand on an hourly basis, which could stimulate deployment of a wider portfolio of flexible technologies needed for net zero transitions in the power sector.

While broader electricity decarbonisation policies are key to reducing operational emissions from data centres and networks, there are currently few policies and regulations primarily focused on reducing their energy consumption or emissions footprint. In data centres, these include: 

• Regulatory and voluntary schemes to improve energy efficiency at the component level (e.g. servers, data storage, heating, ventilation and air conditioning [HVAC]) such as ENERGY STAR and EU Ecodesign Regulations for servers and data storage products. 
• Buildings-based data centre energy efficiency guidance, standards, ratings, certifications and labelling schemes such as the EU Code of Conduct on Data Centre Energy Efficiency, CLC/TS 50600-5-1, BREEAM SD 5068 (United Kingdom) and IGBC Green Data Center Rating System (India). 

 
Other recent policy developments target improved data collection and transparency: 

• The Corporate Sustainability Reporting Directive (CSRD) was adopted in 2022 by the European Commission to come into force from 2024 onwards. It will require large organisations, including technology companies, to report sustainability indicators as well as energy and carbon emissions. In the United States, similar reporting mandates are underway at the state level in Oregon, while at the federal level the Ensuring Responsible Development of Digital Assets executive order was signed to explore the potential for future reporting requirements for crypto assets. 
• The recast Energy Efficiency Directive (EED), will introduce energy and sustainability reporting requirements for data centres based in the European Union from May 2024. This will require data centres with installed capacity greater than 500 kW to report total energy consumption – including the share derived from renewable energy – water usage, and waste heat utilisation. The European Commission will evaluate this data to determine whether further measures such as minimum performance standards for data centres are warranted. 
• In China, the government has called for average power use effectiveness (PUE) of 1.25 in the east and 1.2 in the west of the country as part of its Eastern Data and Western Computing project. Major cities have minimum PUE requirements for new data centres, including Beijing (1.4), Shanghai (1.3) and Shenzhen (no subsidies above 1.4).
• Other jurisdictions announced temporary restrictions on new hyperscale data centre developments in 2022, including the Netherlands. Singapore recently lifted its 2019 moratorium on new data centres.

• In January 2021 data centre operators and industry associations in Europe launched the Climate Neutral Data Centre Pact, pledging to make data centres climate-neutral by 2030 with intermediate (2025) targets for power usage effectiveness and carbon-free energy. 
• The Open Compute Project is a collaborative community focused on redesigning hardware technology to efficiently support the growing demands on computing infrastructure. 
• The 24/7 Carbon-free Energy Compact, coordinated by Sustainable Energy for All and the United Nations, includes three data centre operators: Google, Microsoft and Iron Mountain. 
• DIMPACT is a collaborative project convened by Carnstone and the University of Bristol to measure and report the carbon footprint of digital services. DIMPACT participants include some of the largest media companies in the world, including Netflix, the BBC and the Economist.
  

The authors would like to thank Steven Beletich (IEA TCEP 4E Electronic Devices and Networks Annex)Brian Denvir (Google), Devon Swezey (Google), Mark Caine (Google), George Kamiya (GSMA), Vlad Coroama (TU Berlin), Simon Hinterholzer (Borderstep Institute),Daniel Schien (University of Bristol), Jake Oster (Amazon), andStijn Grove (Dutch Data Center Association) for their helpful comments on earlier drafts of this report.