1、 EUR 31841 EN ISSN 1831-9424 Kamiya,G.,Bertoldi,P.2024 Energy Consumption in Data Centres and Broadband Communication Networks in the EU JRC135926 EUR 31841 EN PDF ISBN 978-92-68-12554-0 ISSN 1831-9424 doi:10.2760/706491 KJ-NA-31-841-EN-N Luxembourg:Publications Office of the European Union,2024 Eur

2、opean Union,2024 The reuse policy of the European Commission documents is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents(OJ L 330,14.12.2011,p.39).Unless otherwise noted,the reuse of this document is authorised under the Creative Commons A

3、ttribution 4.0 International(CC BY 4.0)licence(https:/creativecommons.org/licenses/by/4.0/).This means that reuse is allowed provided appropriate credit is given and any changes are indicated.For any use or reproduction of photos or other material that is not owned by the European Union permission m

4、ust be sought directly from the copyright holders.How to cite this report:European Commission,Joint Research Centre,Kamiya,G.and Bertoldi,P.,Energy Consumption in Data Centres and Broadband Communication Networks in the EU,Publications Office of the European Union,Luxembourg,2024,https:/data.europa.

5、eu/doi/10.2760/706491,JRC135926.This document is a publication by the Joint Research Centre(JRC),the European Commissions science and knowledge service.It aims to provide evidence-based scientific support to the European policymaking process.The contents of this publication do not necessarily reflec

6、t the position or opinion of the European Commission.Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use that might be made of this publication.For information on the methodology and quality underlying the data used in this publication for whi

7、ch the source is neither Eurostat nor other Commission services,users should contact the referenced source.The designations employed and the presentation of material on the maps do not imply the expression of any opinion whatsoever on the part of the European Union concerning the legal status of any

8、 country,territory,city or area or of its authorities,or concerning the delimitation of its frontiers or boundaries.Contact information Name:Bertoldi Paolo Address:JRC,21027 Ispra(VA),Italia,TP 450 Email:paolo.bertoldiec.europa.eu Tel.:+39-0332 78 9299 EU Science Hub https:/joint-research-centre.ec.

9、europa.eu 1 Contents Abstract.2 Executive Summary.3 1.Introduction.4 1.1 Background.4 1.2 Objectives and approach.4 2.Literature review.6 2.1 Introduction.6 2.2 Modelling approaches.6 2.3 Data centres.7 2.4 Telecommunication networks.12 2.5 Digital technologies and services.12 3.Data collection and

10、modelling.15 3.1 Introduction.15 3.2 Data centres.15 3.3 Telecommunication networks.18 3.4 Summary and limitations.19 4.Results and discussion.21 4.1 Data centres.21 4.2 Telecommunication networks.23 4.3 Summary.25 5.Conclusions and recommendations.26 References.28 List of figures.35 List of tables.

11、36 Annex.37 2 Abstract Demand for digital services is rising rapidly,raising concerns about the energy use and environmental impacts of data centres and telecommunication networks.Despite the increasing public and policy interest in addressing these impacts,there is a lack of official statistics on

12、the energy use of digital infrastructure.This study reviews and uses existing literature and public data sources to estimate the energy consumption of data centres and telecommunication networks in the European Union(EU-27)in 2022.Data centres in the EU used an estimated 4565 TWh of electricity in 2

13、022(1.82.6%of total EU electricity use),while telecommunication networks used an estimated 2530 TWh of electricity(11.2%of total EU electricity use).Network energy use as a share of national electricity use was both lower and more uniform than data centres.Policymakers and companies should work toge

14、ther to improve data collection,quality,and availability in order to better understand trends and make informed policy decisions to manage the energy and environmental impacts of digital infrastructure.3 Executive Summary Demand for digital services is rising rapidly,raising concerns about the energ

15、y use and environmental impacts of data centres and telecommunication networks.Despite the increasing public and policy interest in addressing these impacts,there is a lack of official statistics on the energy use of digital infrastructure.This study reviews and uses existing literature and public d

16、ata sources to estimate the energy consumption of data centres and telecommunication networks in the European Union(EU-27)in 2022.Data centres in the EU used an estimated 4565 TWh of electricity in 2022,equivalent to 1.82.6%of total regional electricity consumption.The top four data centre markets G

17、ermany,France,the Netherlands,and Ireland accounted for nearly two-thirds of the regions data centre energy use,despite having less than 40%of the population.Data centres represent over 2%of national electricity use in Ireland(18%),the Netherlands(5.2%),Luxembourg(4.8%),Denmark(4.5%),and Germany(3%)

18、,Sweden(2.3%),and France(2.2%).Telecommunication networks used an estimated 2530 TWh of electricity,equivalent to 11.2%of total EU electricity use.The four largest Member States by population and GDP(Germany,France,Italy,and Spain)were also the four largest users of energy for telecommunication netw

19、orks,accounting for 65%of the total.Network energy use as a share of national electricity use was both lower and more uniform compared with data centres,ranging from 0.5%to 1.5%.In contrast,data centres as a share of national electricity use range from as low as 0.4%in some countries to as high as 1

20、8%in Ireland.The combined energy use of data centres and telecommunication networks in the EU was 7095 TWh in 2022,equivalent to 2.83.8%of total regional electricity use.The four largest Member States Germany,France,Italy,and Spain accounted for about 60%of total digital infrastructure energy use in

21、 the region.Digital infrastructure accounts for more than 5%of national electricity use in four countries,each with major data centre markets:Ireland(19%),the Netherlands(6%),Luxembourg(5.5%),and Denmark(5%).Policymakers and companies must work together to improve data collection,quality and availab

22、ility.While the estimates of this study represent a likely range of figures,it is critical to develop more robust estimates to better understand trends and make informed policy decisions to manage the energy and environmental impacts of digital infrastructure.Governments and statistical agencies sho

23、uld develop standardised definitions and classifications for data centres and networks,such as providing criteria and guidance on classifying different data centre types.Governments and companies should work together to improve data quality and availability regarding data centre energy consumption(b

24、y size and type),telecommunication network energy use(by type),as well as relevant activity indicators(e.g.connections,data traffic,data centre workloads).Data collection efforts should also seek to better understand energy use characteristics and implications of specific services and tasks such as

25、artificial intelligence.4 1.Introduction 1.1 Background Digital infrastructure data centres and telecommunication networks are at the heart of the digital transformation,underpinning all aspects of our increasingly digitalised and connected societies.Demand for their services is rising rapidly,raisi

26、ng concerns about their energy use and environmental impacts.Between 2015 and 2022,the number of internet users globally increased by 80%,internet traffic increased five-fold,while data centre workloads more than quadrupled(IEA,2023a;ITU,2023;Cisco,2019;TeleGeography,2022,2023;Cisco,2018;Masanet et

27、al.,2020).However,energy efficiency improvements in computation,data storage,and data transmission have helped to limit energy demand growth globally.Between 2015 and 2022,data centre and network energy demand grew at a much slower pace(+28%CAGR)than data traffic(+24%CAGR)or data centre workloads(+3

28、3%CAGR)(Table 1).Data centres and data transmission networks each accounted for 11.5%of global electricity use globally in 2022(IEA,2023a).Table 1.Global trends in digital and energy indicators,20152022 2015 2022 Change Internet users 3 billion 5.3 billion+78%Internet traffic 0.6 ZB 4.4 ZB+600%Data

29、centre workloads 180 million 800 million+340%Data centre energy use(excluding crypto)200 TWh 240340 TWh+2070%Crypto mining energy use 4 TWh 100150 TWh+23003500%Mobile subscriptions 7.1 billion 8.3 billion+17%Fixed broadband subscriptions 790 million 1.2 billion+51%Data transmission network energy us

30、e 220 TWh 260360 TWh+1864%Sources:IEA(2023a);Malmodin et al.(2023)In contrast to modest growth at the global level,there has been significant growth in some countries,especially for data centre energy use.In Ireland,for example,data centre energy use more than tripled between 2015 and 2022,growing t

31、o 18%of national electricity use in 2022(Central Statistics Office,Ireland,2022).In the European Union(EU),there is a lack of recent and comprehensive estimates of the energy consumption of data centres and telecommunication networks.In order to formulate effective policies to manage the energy use

32、of digital infrastructure,policymakers require better understanding of data centre and network energy consumption.1.2 Objectives and approach The objective of this study is to estimate the energy consumption of data centres and telecommunication networks in the European Union(EU-27).In order to deve

33、lop these estimates,a comprehensive literature assessment was conducted to compile available data to inform the development of a simplified model to estimate data centre and network energy use at the national and EU-27 levels.Section 2 summarises the literature review of ICT energy estimates from a

34、range of available sources,including government data and reports,peer-reviewed journal articles,industry data and reports,and other 5 grey literature.The section also reviews literature estimating the energy use of emerging digital technologies and services such as AI,blockchain and cryptocurrencies

35、,streaming and gaming.Section 3 summarises the modelling methodology used to estimate EU-27 energy consumption of data centres and networks in 2022,including key data sources and assumptions.Section 4 summarises the results and discusses key strengths and limitations of the analysis.Section 5 conclu

36、des the reports with key recommendations for future data collection efforts.6 2.Literature review 2.1 Introduction National statistical agencies and regional and intergovernmental organisations such as the European Commission(Eurostat,2023)and the International Energy Agency(IEA,2023b)collect and pu

37、blish official statistics on the energy use of many energy end-use sectors and services(e.g.steel,road transport,lighting).However,to date,there is a lack of official statistics on the energy use of the information and communications technology(ICT)sector1 at the national,regional,and global levels.

38、Since 2018,only a few studies have comprehensively estimated the energy consumption of the entire ICT sector(4E EDNA,2019,2021;Andrae,2019,2020;ITU,2020;Malmodin et al.,2023;Malmodin&Lundn,2018;The Shift Project,2019b,2020a,2021).However,as these studies do not include any regional disaggregation in

39、 their published materials,estimates for Europe or individual European countries are not available.The literature review(Task 1)identified and summarised the studies and estimates published since 2018,focusing on data centres and Europe.This section summarises the key findings from the literature re

40、view.2.2 Modelling approaches Studies that have estimated the energy use of data centres and telecommunication networks at the global,regional,and national levels have employed a variety of modelling approaches.These methodologies can be broadly categorised into one of three types(or a combination)b

41、ottom-up,top-down,and extrapolation each with their own advantages and disadvantages(Mytton&Ashtine,2022).Bottom-up studies use detailed data on technology such as equipment specifications(e.g.server power draw),data centre infrastructure characteristics(e.g.power usage effectiveness PUE)and install

42、ed base and equipment shipment values.The main advantage of these studies is that they have substantial explanatory power and they can be useful for exploring the potential effects of policies,technologies,and other trends.However,their substantial data requirements make these studies very resource-

43、and time-intensive to produce.Some data inputs such as proprietary market data can be very expensive and cannot be shared,limiting transparency.Some bottom-up studies have used high-level parameters such as data centre floor space or the number of data centres to estimate data centre energy use.Give

44、n the uncertain(and relatively weak)relationship between these parameters and energy use,studies using these methods are less certain and credible compared with those that use technology-level data such as the number of installed servers and their energy use characteristics.Top-down estimates compil

45、e measured or estimated energy consumption data from governments and companies.Their main advantage is that they are accurate,being based on reliable data,and easy to generate and update.But the limited availability of data from governments and companies currently means that only a portion of the ov

46、erall scope can be reliably estimated,requiring other complementary approaches to ensure comprehensive coverage.Extrapolation approaches combine high-level activity indicators such internet protocol(IP)traffic with energy intensity assumptions to project total energy use under different activity and

47、 efficiency improvement scenarios.They require a baseline energy consumption estimate from a bottom-up or top-down model,and decisions around assumed growth rates(e.g.energy efficiency improvement,data volume growth).These studies are typically more transparent and relatively easy to generate and up

48、date.Their main disadvantages are their low explanatory power and a higher risk of mis-use(e.g.developing exaggerated estimates from long-term projections).1 According to the ITU-T L.1450 Recommendation,for the purposes of assessing the environmental impact of the ICT sector,the sector boundary incl

49、udes:i)ICT end-user goods(puters and computer peripherals,consumer electronics for communications purposes such as mobile phones,tablets,laptop PCs and home network goods,and IoT devices);ii)ICT network goods(e.g.telecommunication core networks and access networks);iii)data centres;and iv)ICT servic

50、es(e.g.software development)(ITU,2018)7 Hybrid estimates combine a combination of approaches,typically combining available top-down data from governments and companies with bottom-up data or extrapolation approaches using available data.2.3 Data centres Global estimates Since 2018,several institutio

51、ns and researchers have estimated the global energy use of data centres2,employing a range of scopes,methodologies,assumptions,and data sources(Annex,Table A.1).There is a significant range in the estimates,ranging from around 200 TWh to nearly 1,000 TWh for the year 2020.If outlier figures from stu

52、dies that extrapolate outdated assumptions or analyses are excluded(Belkhir&Elmeligi,2018;The Shift Project,2019b,2021),this range narrows significantly to 200380 TWh in 2020(excluding cryptocurrencies),equivalent to 0.81.6%of global electricity consumption.Cryptocurrencies accounted for an addition

53、al 80100 TWh in 2020,or around 0.4%of global electricity use(IEA,2021).European estimates Several studies have estimated data centre energy consumption for Europe or the European Union over the past few years,primarily funded or co-authored by the European Commission and European Union.The three mos

54、t recent reports were published in 2020,prepared for different departments of the European Commission:A report co-authored by the Joint Research Centre Product Bureau and consultants(Viegand Maage,Operational Intelligence,Hansheng,Ballarat Consulting)estimated that data centres in the EU consumed 10

55、4 TWh in 2020(Dodd et al.,2020).Enterprise data centres accounted for 44%of the total,while colocation and managed service providers(e.g.cloud)accounted for 56%.They also projected energy consumption would grow to 134 TWh in 2025 and 160 TWh in 2030.A report prepared for DG ENER by consultants VHK a

56、nd Viegand Maage estimated that data centres in the EU27 consumed 39.5 TWh in 2020(VHK&Viegand Maage,2020).This estimate was based on the disaggregated estimate for Western Europe in Masanet et al.(2020).A report prepared for DG CONNECT by the Environment Agency Austria and the Borderstep Institute

57、found that data centre energy consumption in the EU28 increased from 53.9 TWh in 2010 to 76.8 TWh in 2018(Montevecchi et al.,2020).Based on current trends,they projected energy use would grow to 92.6 TWh in 2025 and 98.5 TWh in 2030.Table 2.Overview studies estimating the energy use of data centres

58、in Europe since 2018 Publication Approach Estimate Quality assessment Bashroush(2018)Methodology not disclosed.130 TWh in 2017 Low no details available to assess BloombergNEF et al.(2021)Bottom-up estimate of colocation and hyperscale data centres based on data from Eaton on installed data centre ca

59、pacity in each country,public announcements,assumed rack capacities,and lease rates.26 TWh in 2021 for Germany,Ireland,Netherlands,Norway,and the United Kingdom Medium-high high quality analysis but limited geographic scope and excludes small data centres Dodd et al.(2020)Estimate based on data from

60、 2013 data from DCD supplemented by surveys on data centre area,installed capacities,and other studies from the US and Europe including the Code of Conduct.74 TWh in 2015 104 TWh in 2020 for EU27 Projections:134 TWh in 2025 and 160 TWh in Low-medium 2 See Mytton&Ashtine(2022)for a comprehensive and

61、critical review of 258 data centre energy estimates published between 2007 and 2021.8 Publication Approach Estimate Quality assessment 2030 Masanet et al.(2020)Bottom-up estimate based on stock and shipment data for servers,drives,networking,their energy use characteristics and lifespans,combined wi

62、th assumptions for each type of data centre class and region-specific PUE.39.4 TWh in 2018 for Western Europe Medium-high Montevecchi et al.(2020)Bottom-up estimate based on data centre market developments,technical characteristics of servers,storage,and networking(energy use,age)and data centre inf

63、rastructure(air conditioning,power supply,UPS).76.8 TWh in 2018 for EU28(2.7%of EU28)Projections:92.6 TWh in 2025 and 98.5 TWh in 2030 Medium VHK&Viegand Maage(2020)Based on Masanet et al.(2020).39.5 TWh in 2020 for EU27 Same as Masanet et al.(2020)Source:JRC There is a large range in the results;fo

64、r example,estimates for 2020 range from 39.5 TWh to 104 TWh,and projections for 2030 range from 98.5 TWh to 160 TWh.The large range stems from substantial differences in data sources,assumptions,and methodologies,which are summarised in Table 2.However,the lack of details and documentation regarding

65、 assumptions and methodology makes it difficult to compare underlying differences,and how they contribute to diverging estimates.Older reports on the topic include 2014 reports prepared for DG CONNECT(Prakash et al.,2014)and DG GROW(Deloitte et al.,2014),as well as a 2017 report funded under the EU

66、Horizon 2020 programme(Bashroush,2018).Prakash et al.(2014)estimated that data centre energy use in the EU27 would increase from 52 TWh in 2011 to 70 TWh in 2020.Deloitte et al.(2014)estimated that data centres consumed 78 TWh in the EU28 in 2015.Bashroush(2018)estimated 130 TWh in 2017.The results

67、of the studies published since 2014 are summarised in Figure 1.Two other studies have estimated the energy consumption of data centres in Europe.The global study by Masanet et al.(2020)includes disaggregated estimates for“Western Europe”,with an estimate of 39.4 TWh in 2018.BloombergNEF,Statkraft,an

68、d Eaton published a joint study in 2021 focusing on colocation and hyperscale data centres in Germany,Ireland,Netherlands,Norway and the United Kingdom(BloombergNEF et al.,2021).9 Figure 1.Summary of European data centre energy estimates Note:Darker circles indicate estimates;lighter circles and dot

69、ted lines indicate projections.Source:JRC Country-level estimates There are country-level estimates of many of the largest data centre markets in Europe.The most credible estimates are based on reported electricity consumption and metering data(Ireland,the Netherlands,Finland)as well as bottom-up es

70、timates using robust models developed over many years(Germany).Most estimates focus on colocation and hyperscale data centres.Country-level estimates from within the EU-27 are summarised in Table 3,and include the following:Belgium:The Belgian Digital Infrastructure Association,based on research by

71、consultants Pb7,estimated that colocation and hyperscale data centres used 380 GWh in 2021(Belgian Digital Infrastructure Association,2022).Denmark:The Danish Energy Agency commissioned a study in 2021 which estimated that data centres consumed 0.88 TWh in 2020(COWI,2021),and have since published da

72、ta centre energy consumption estimates in their annual Energy and Climate Outlook,with an estimate of 1.1 TWh in 2021(Danish Energy Agency,2023a).Finland:The Research Institute of the Finnish Economy(ETLA)estimated that data centres consumed around 250 GWh in 2018 based on reported electricity consu

73、mption of the two-digit level industrial classification covering data centres(Hiekkanen et al.,2021).France:The French environment agency(Ademe)and telecommunications regulator(Arcep)jointly published a study in 2022 estimating the total energy consumption of data centres in France at 11.6 TWh in 20

74、20(Ademe&Arcep,2022).They found that colocation(49%)and enterprise data centres(36%)account for the vast majority of total data centre energy use.GreenIT.fr estimated that data Dodd et al.(2020)VHK&Viegand Maage(2020)Montevecchi et al.(2020)Bashroush(2018)Prakash et al.(2014)Deloitte et al.(2014)020

75、40608010012014016020102015202020252030TWh 10 centres in France consumed 5.2 TWh3 in 2020(Bordage et al.,2021).A report commissioned by the French Senate estimated that data centres in France consumed around 9 TWh in 2019(CITIZING,2020).Germany:The Borderstep Institute estimated that data centres in

76、Germany consumed 14 TWh in 2018,16.3 TWh in 2020,17 TWh in 2021,and 17.9 TWh in 2022(Hintemann et al.,2023;Hintemann&Hinterholzer,2020,2022).Based on current trends,they project data centre energy consumption could reach 27 TWh in 2030(Hintemann et al.,2023).BloombergNEF estimates that colocation an

77、d hyperscale data centres in Germany used 7.2 TWh in 2021(BloombergNEF et al.,2021).Ireland:The Central Statistics Office(CSO)analysed meter data from the Irish electricity utility ESB to estimate the combined electricity consumption of all meters that were primarily being used for data centre activ

78、ities.The CSO estimate that data centres used 2.5 TWh in 2019(9%of national electricity consumption),3 TWh in 2020(11%),4 TWh in 2021(14%),and 5.25 TWh in 2022(18%)(Central Statistics Office,Ireland,2021,2022,2023).The study found that a small number of large data centres accounted for most of the m

79、etered electricity consumption.BloombergNEF estimated that hyperscale and colocation data centres in Ireland consumed 4.7 TWh in 2021(BloombergNEF et al.,2021).Netherlands:Based on business registration data and electricity consumption data from the regional grid operator TenneT,Statistics Netherlan

80、ds estimated that data centres consumed 1.6 TWh in 2017,2.4 TWh in 2018,and 2.7 TWh in 2019(Statistics Netherlands,2021a).Larger data centres(consuming more than 7.5 GWh)accounted for 90%of the 2019 total.BloombergNEF estimated that colocation and hyperscale data centres in the Netherlands used 6.3

81、TWh in 2021(BloombergNEF et al.,2021).Sweden:The Swedish Energy Agency and the Research Institutes of Sweden(RISE)estimate that data centres in the country used 2.83.2 TWh in 2022.Radar,an IT consulting firm,estimated that Swedish data centres used 2.4 TWh of electricity in 2020.They estimate floor

82、space of 20 hectares,total capacity of 640 MW,and 43%power utilisation.Table 3.Overview studies estimating the energy use of data centres in Europe since 2018 Country Publication Approach Estimate Quality assessment Belgium Belgian Digital Infrastructure Association(2022)Based on research conducted

83、by Pb7(consultants);details of methodology not provided.0.38 TWh in 2021 Low-medium Denmark COWI(2021)Historical estimates based on data centre characteristics,data traffic,and other national and international studies.0.88 TWh in 2020 Medium Danish Energy Agency(2021;2022a;2022b;2023a)Based(COWI,202

84、1)and expected power utilisation data from the transmission system operator,Energinet.1.1 TWh in 2021 Medium Finland Hiekkanen et al.(2021)Based on the reported electricity consumption of the two-digit level industrial classification covering data centres(Computer and information service activities,

85、TOL 6263).0.25 TWh in 2018 High France Ademe&Arcep(2022)Based on publicly available data on colocation data centres and assumptions for data centre area(by type)and energy 11.6 TWh in 2020 Medium 3 The report estimates total electricity consumption for digital technologies in 2020 of 40 TWh,of which

86、 13%was consumed in the use phase by data centres.11 Country Publication Approach Estimate Quality assessment characteristics from previous studies.Bordage et al.(2021)Based on the number of servers in operation from University of Sherbrooke and some assumed energy“impact factor”and average PUE of 1

87、.7.5.2 TWh in 2020 Low CITIZING(2020)Extrapolation approach based on data traffic to and from data centres and assumed energy intensity.9 TWh in 2019 Low Germany Hintemann et al.(2023);Hintemann&Hinterholzer(2020;2022)Bottom-up estimate based on data centre market developments,technical characterist

88、ics of servers,storage,and networking(energy use,age)and data centre infrastructure(air conditioning,power supply,UPS).14 TWh in 2018 16.3 TWh in 2020 17 TWh in 2021 17.9 TWh in 2022 High BloombergNEF et al.(2021)Bottom-up estimate of colocation and hyperscale data centres(details in Table 2).7.2 TW

89、h in 2021 Medium-high Ireland Central Statistics Office,Ireland(2021;2022;2023)Based on meter data from electricity utility to estimate the combined electricity consumption of all meters that were primarily being used for data centre activities.2.5 TWh in 2019 3 TWh in 2020 4 TWh in 2021 5.3 TWh in

90、2022 High BloombergNEF et al.(2021)Bottom-up estimate of colocation and hyperscale data centres(details in Table 2).4.7 TWh in 2021 Medium-high Netherlands Statistics Netherlands(2021a)Based on business registration data and electricity consumption data from the regional grid operator TenneT.1.6 TWh

91、 in 2017 2.4 TWh in 2018 2.7 TWh in 2019 High BloombergNEF et al.(2021)Bottom-up estimate of colocation and hyperscale data centres(details in Table 2).6.3 TWh in 2021 Medium-high Sweden Swedish Energy Agency(2023)Based on available reports,interviews,and statistical data.2.83.2 TWh in 2022 Medium-h

92、igh Radar(2020)Analysis of quantitative data and previous reports complemented by qualitative information from interviews.2.4 TWh in 2020 Medium-high Source:JRC There are several published country-level estimates for important European data centre markets outside the EU-27,notably Norway and the Uni

93、ted Kingdom:Norway:The Norwegian Water Resources and Energy Directorate(NVE)estimate that data centres in Norway consumed 0.8 TWh in 2019(NVE,2020).The report also projects data centres to consume 49 TWh in 2040.BloombergNEF estimates that colocation and hyperscale data centres in Norway used 0.7 TW

94、h in 2021(BloombergNEF et al.,2021).United Kingdom:National Grid ESO,the national the UKs national electricity system operator,estimates that data centres used 47 TWh of electricity in 2020,equivalent to 1.32.5%of national electricity use(National Grid ESO,2022).BloombergNEF estimates that colocatio

95、n and hyperscale data centres in the United Kingdom used 7.2 TWh in 2021(BloombergNEF et al.,2021).12 Outside of Europe,there are national level estimates on data centre energy use covering the largest data centre markets in the world.In the United States,data centres consumed 70 TWh in 2014(1.8%of

96、national electricity use)and projected to rise to 73 TWh in 2020 based on current trends(Shehabi et al.,2016,2018).In China,some studies have estimated that data centres4 consumed 150200 TWh in 2020,or around 2%of national electricity use(Fan,2021;Greenpeace East Asia,2021;Greenpeace East Asia&North

97、 China Electric Power University,2019).In Japan,data centres used around 20 TWh in 2021,or 2%of national electricity consumption(Deloitte Tohmatsu MIC Research Institute,2022;Nikkei,2022).In Singapore,data centres accounted for around 7%of national electricity consumption in 2020(Singapore Ministry

98、of Communications and Information,2021).2.4 Telecommunication networks There are only a few estimates on the global estimates for telecommunication network energy use.Malmodin et al.(2023)estimate that networks consumed 244 TWh in 2020,or around 1%of global electricity use.Coroam(2021)reviewed botto

99、m-up and top-down studies to estimate that networks consumed 340 TWh in 2020(1.4%of global electricity).The GSMA estimated that telecommunication network operators used 293 TWh of electricity in 20215(GSMA,2022).The energy use of European network operators was estimated by Lundn et al.,(2022)based o

100、n publicly reported data on energy use of operators covering 36%of European subscriptions.They estimated that European network operators in used 38 TWh in 2018(EU28)and 29 TWh in 2020(EU27),equivalent to around 1.2%of total regional electricity use.Several national regulatory agencies in Europe have

101、 compiled and published estimated energy consumption of telecommunication networks in their countries(BEREC,2023):Belgium:The three major telecom operators consumed an estimated 624 GWh of electricity in 2021,of which 481 GWh was consumed in networks(0.7%of national electricity use)(BIPT,2022).Mobil

102、e networks accounted for 60%of total network energy use6.Finland:Communications networks consumed an estimated 650 GWh in 2021(0.75%of national electricity use)(Traficom,2022).Mobile networks accounted for 58%of total network energy use(Traficom,2023).France:Data networks consumed 3.9 TWh of electri

103、city in 2021,equivalent to 0.8%of national electricity use(Arcep,2022,2023a,2023b).Mobile networks accounted for nearly 60%of the total.Total network energy use was up 3%from 2020 levels.Data network energy use has increased 25%since 2017,representing a compound annual growth rate of 6%.2.5 Digital

104、technologies and services Emerging technologies such as artificial intelligence(AI),blockchain,and 5G are poised to boost demand for digital infrastructure.These technologies and trends could have different implications for energy use in data centres,networks,and devices,with some technologies such

105、as AI and blockchain primarily impacting data centres while 5G and IoT likely to affect networks and devices.Artificial intelligence Artificial intelligence(AI)is likely to have significant implications for data centre energy use in upcoming years.Early studies on the energy and carbon footprint of

106、AI and machine learning(ML)focused on the energy and carbon emissions associated with training large language models(Lacoste et al.,2019;Luccioni et 4 Some references to these reports note that the energy totals include 5G energy consumption.The exact scope and methods cannot be verified as the orig

107、inal report could not be located.5 This estimate includes all uses of electricity by network operators,including networks,data centres,offices,and stores.Based on available company-level data regarding electricity use between these end uses,it is likely that networks typically accounted for 7090%of

108、the total for most operators,with the exception of those with substantial data centre businesses.6 This figure is based on data from Orange and Telenet,which together account for 43%of energy use by network operators,as disaggregated data for the largest operator(Proximus)was not available.13 al.,20

109、20;Schwartz et al.,2019;Strubell et al.,2019).But training a single ML model represents only a small fraction of the overall energy use of AI.Recent data from Meta(Wu et al.,2022)and Google(Patterson et al.,2022)indicate that the training phase only accounts for around 2040%of overall ML-related ene

110、rgy use,with 6070%for inference(application/use and up to 10%for model development(experimentation).A Danish researcher estimated that ChatGPT used around 4 GWh in January 2023(Ludvigsen,2023a,2023b),about three times more electricity than was used to train GPT-3(Patterson et al.,2022),the model tha

111、t provides the basis for ChatGPT.Only a fraction of total ICT energy use is attributable to AI and ML,but its exact share is not known due to challenges in boundary definition and a lack of data and established methodology(Kaack et al.,2022).Based on estimates of global ICT energy use(IEA,2023a;Malm

112、odin et al.,2023)and shares of data centre workloads and data centre IP traffic attributed to AI(Cisco,2018;Compton,2018),Kaack et al.(2022)estimated that AI likely accounted for less than 0.2%of global electricity use in 2021(50 TWh).Other researchers have estimated future potential AI energy consu

113、mption based on energy use characteristics and projected shipments of servers(Vries,2023).Google estimated that ML accounted for 1015%of their total energy use in recent years(i.e.23 TWh in 2021),but noted that it is growing at a similar rate as overall company-wide energy use around 2030%per year(G

114、oogle,2022;Patterson et al.,2022).Computing demand for ML training and inference at Meta have increased by more than 100%per year in recent years,while overall data centre energy consumption grew about 40%per year(Meta,2022;Naumov et al.,2020;Park et al.,2018).The combination of the rapid growth in

115、the size of the largest models(OpenAI,2018)and ML compute demand(Wu et al.,2022)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(DeepMind,2016;Luo et

116、al.,2022),the rapid and mainstream adoption of AI chatbots like OpenAIs ChatGPT and Google Bard are likely to accelerate energy demand growth for AI.Blockchain and cryptocurrencies Blockchain and other distributed ledger technologies are major energy users.Bitcoin the most prominent example of proof

117、-of-work blockchain and most valuable cryptocurrency by market capitalisation consumed around 95 TWh in 2022,equivalent to 0.4%of global electricity use(Cambridge Centre for Alternative Finance,2023).While this was similar to its electricity use in 2021,it is 17-times higher than its energy use in 2

118、016.Ethereum,second behind Bitcoin in terms of market capitalisation and energy use,consumed around 18 TWh over the first three quarters of 2022(McDonald,2022).In September 2022,Ethereum transitioned from a proof-of-work consensus mechanism to proof-of-stake,which slashed energy use by more than 99.

119、95%(CCRI,2022;ethereum.org,2023).Gallersdrfer et al.(2020)estimated that Bitcoin and Ethereum accounted for 80%of all crypto-related energy use in 2020.The same authors estimated that cryptocurrencies as a whole consumed around 150 TWh in 2022(CCRI,2023).Streaming media and cloud gaming The delivery

120、 of streaming videos,music,and gaming from a content provider to the viewer is associated with energy consumption across the ICT system,including in data centres,telecommunication networks,customer premises equipment(e.g.routers),and end user devices.The energy and carbon footprint of streaming vide

121、o has attracted significant attention from researchers,companies,and the media over the past few years.Much of this attention can be traced back to media headlines from 2019 quoting the Shift Project claiming that half an hour of streaming emitted as much CO2 as driving four miles(6,100 Watt-hours(W

122、h)per viewing hour)(AFP,2019;Sparks,2019;The Shift Project,2019a),which was later revised downwards by 88%after correcting a unit conversion error(The Shift Project,2020b).Marks et al.(2020)estimated that streaming 35 hours of HD video consumes 382 kWh(11,000 Wh per hour),but have since revised thei

123、r estimate downwards by over 90%to 780980 Wh(Makonin et al.,2022).These studies overestimate energy use particularly by networks by using outdated energy intensity values(kWh/GB)and assuming that energy use is proportional to data traffic(i.e.doubling bitrate doubles energy use).14 These figures are

124、 substantially higher than more recent estimates published by the IEA(80180 Wh per hour)and the Carbon Trust(220 Wh)(Kamiya,2020b,2020a;The Carbon Trust,2021).A July 2023 study published by the European Commission estimated that a typical hour of streaming in Europe uses 50 Wh(EC DG Energy et al.,20

125、23).Outdated energy intensity assumption and the use of faulty methodologies have resulted in some studies significantly overestimating the energy and carbon footprint of streaming video,particularly from data transmission(Moulierac et al.,2023).Based on typical viewership patterns today,the vast ma

126、jority of total end-to-end energy use(i.e.from the data centre to the viewing devices)is consumed by end user devices and home networking equipment(The Carbon Trust,2021).Online gaming typically consumes more energy than streaming video,both from higher data intensity and from using more energy inte

127、nsive devices.Aslan(2020)compared the carbon intensity of different gaming methods,and found that downloading was the least carbon intensive(47g CO2e per hour),followed by disc(55g)and cloud(149g).Mills et al.(2019)also found that cloud gaming is up 30-300%more energy intensive than local gaming,con

128、suming 1001000 kWh annually per user(Cardoso,2020).They also estimate that gaming could consume 34 TWh in 2016 in the United States.5G and the Internet of Things Mobile data traffic is projected to continue growing quickly,more than tripling between 2023 and 2028(Ericsson,2022).5Gs share of global m

129、obile data traffic is projected to rise to 70%by 2028,up from 27%in 2023,driven by early adopters of 5G,including the United States,China,the Republic of Korea and European countries.5G networks are expected to be more energy efficient than 4G networks per unit of traffic and benefit from improved s

130、leep modes(Johnson,2018;Orange,2022;STL Partners,2019).But higher traffic volumes and a higher number of base stations are likely to increase overall energy and emissions,as indicated by studies from developed countries like France,Switzerland and the United Kingdom(Bieser et al.,2020;Haut Conseil p

131、our le Climat,2020;Williams et al.,2022).IoT adoption is expected to grow rapidly over the next five years,reaching 35 billion connections by 2028(Ericsson,2022).The low latency and high data throughput of 5G is also expected to accelerate cellular IoT adoption,which could double to 5.5 billion conn

132、ections.IoT devices are generally expected to be energy efficient,but the sheer growth in the number of IoT devices could have important implications for standby energy use and embodied energy and material in IoT devices.15 3.Data collection and modelling 3.1 Introduction A mixed-method approach was

133、 used to develop a simplified spreadsheet model to estimate country-level energy use.The model uses country-level data centre and network energy use estimates where available.For countries without country-level data,estimates are derived using a variety of relevant indicators.This section summarises

134、 the development of the simplified model(Task 2),including data sources,assumptions,and methodologies.3.2 Data centres Data collection Data centre energy use estimates were collected from government reports,data,and other reputable sources(see 2.3 Data centres).Other relevant bottom-up data on data

135、centres(e.g.IT capacity,data centre floorspace,PUE)were collected from a range of reputable sources,including governments,national data centre industry associations,and data centre market research companies.These additional data sources are summarised in Table 4.Other relevant country-level indicato

136、rs were collected from Eurostat and the ITU,including population,GDP,national electricity use,industrial electricity prices,household internet access,mobile internet access rates,and mobile broadband subscription.Modelling approach Depending on the available data for a given country,different modell

137、ing approaches were used:If country-level data centre energy use estimates were available from a reputable source using credible methods,this was used as the primary basis for developing country-level estimates.In some cases where coverage was incomplete,for example,where the estimate only covers co

138、location and hyperscale data centres,this was supplemented with assumptions from other bottom-up studies.Eight countries were estimated using this approach(Belgium,Denmark,Finland,France,Germany,Ireland,Netherlands,and Sweden),collectively covering nearly 80%of EU-27 data centre capacity,62%of GDP,5

139、5%of electricity use,and 46%of population.If energy use estimates were not available,a modelled estimate was derived using data on installed IT capacity or data centre IT floorspace,combined with appropriate assumptions on utilisation rates,PUE,and power density from other credible studies.Five coun

140、tries were estimated using this approach(Austria,Italy,Luxembourg,Poland,and Spain)covering for around 18%of EU-27 data centre capacity,nearly 30%of GDP and electricity use,and one-third of population.For countries without any available IT-related data,relevant economic,energy,and digital indicators

141、 were used to derive a first-order estimate.For example,assuming a given share of national service sector electricity use for data centres based on other countries with similar economic and digital indicators.In some cases,the available energy use estimates or IT capacity data did not cover all size

142、s and types of data centres,for example,including only colocation data centres.In these cases,estimates on energy use of other data centre types were derived by adapting assumptions from bottom-up studies(e.g.Ademe&Arcep,2022;Hintemann et al.,2023;Hintemann&Hinterholzer,2022;Montevecchi et al.,2020)

143、.If 2022 data were not available,historical average growth rates(where available and appropriate)were applied.Given the uncertainties in chosen assumptions,central estimates were complemented with optimistic and pessimistic assumptions to develop a range of estimates for each country.16 Table 4.Data

144、 sources and modelling approaches taken for data centres Country Available data and sources Approach taken Assessment quality Austria Total IT capacity(Mordor Intelligence,2023);colocation IT capacity(Baxtel,2023).14/02/2024 15:13:00 Estimated based on IT capacity estimates(Baxtel,2023a;Mordor Intel

145、ligence,2023)and assumptions from other studies and countries(Ademe&Arcep,2022;Hintemann et al.,2023).Low-medium Belgium Energy use(colocation and hyperscale only);IT capacity and floorspace(Belgian Digital Infrastructure Association,2022)Energy use estimate for colocation and hyperscale from BDIA;e

146、nterprise estimate based on IT capacity from BDIA and assumptions for utilisation and PUE based on bottom-up studies(Ademe&Arcep,2022;Hintemann et al.,2023).High Bulgaria None Assumed to consume 2%of national service sector electricity.Low Croatia None Assumed to consume 2%of national service sector

147、 electricity.Low Cyprus None Assumed to consume 2%of national service sector electricity.Low Czechia None Assumed to consume 4%of national service sector electricity based on estimate for Poland.Low Denmark Energy use for colocation and hyperscale(COWI,2021;Danish Energy Agency,2021,2022a,2022b,2023

148、a,2023b).Energy use estimate for colocation and hyperscale from Danish Energy Agency;energy use for enterprise DCs estimated based on assumptions from bottom-up studies(Ademe&Arcep,2022;Hintemann et al.,2023).Medium-high Estonia None Assumed to consume 4.5%of national service sector electricity base

149、d on estimate for Finland.Low Finland Energy use based on two-digit level industrial classification for data centres(Hiekkanen et al.,2021;Statistics Finland,2022)Energy use estimate for colocation and hyperscale from Statistics Finland;energy use for enterprise DCs estimated based on assumptions fr

150、om bottom-up studies(Ademe&Arcep,2022;Hintemann et al.,2023).Medium-high France Energy use,IT capacity,floorspace,utilisation rate,and PUE for all data centre types(Ademe&Arcep,2022).Based primarily on Ademe&Arcep(2022),with optimistic(lower energy use)estimates developed using input from other coun

151、try-level studies and lower utilisation and PUE assumptions.Medium Germany Energy use(total),IT capacity(by size),and PUE(Hintemann et al.,2023);energy use(colocation and hyperscale)(BloombergNEF et al.,2021).Based primarily on Hintemann et al.(2023),with optimistic(lower energy use)estimates develo

152、ped using energy use estimates for large data centres from(BloombergNEF et al.,2021).High Greece None Assumed to consume 2%of national service sector electricity.Low Hungary None Assumed to consume 2%of national service sector electricity.Low Ireland Electricity meter data of all sites primarily use

153、d for data centre activities(Central Statistics Office,Ireland,2023).Energy use data for colocation and hyperscale data centres(Central Statistics Office,Ireland,2023);energy use for enterprise DCs estimated based on assumptions from bottom-up studies(Ademe&Arcep,2022;Hintemann et al.,2023)adjusted

154、for lower assumed enterprise capacity.High Italy Data centre IT capacity Estimated based on estimated IT capacity Low-medium 17 Country Available data and sources Approach taken Assessment quality(Fernandez,2023;Politecnico di Milano,School of Management,2022).(Fernandez,2023;Politecnico di Milano,S

155、chool of Management,2022)and assumptions for utilisation and PUE(Ademe&Arcep,2022;Hintemann et al.,2023).Latvia None Assumed to consume 2%of national service sector electricity.Low Lithuania None Assumed to consume 2%of national service sector electricity.Low Luxembourg Data centre IT capacity(Baxte

156、l,2023b).Estimated based on IT capacity estimate from(Baxtel,2023b)and assumptions for utilisation and PUE(Ademe&Arcep,2022;Hintemann et al.,2023).Low Malta None Assumed to consume 2%of national service sector electricity.Low Netherlands Data centre energy use(Statistics Netherlands,2021a,2021b);ene

157、rgy use and IT capacity and area(Dutch Data Center Association,2022,2023).Estimated based on historical estimates from Statistics Netherlands(2017-2020)and Dutch Data Center Association(2021-2022)on energy use,IT capacity,and implied utilisation and PUEs.High Poland Data centre IT capacity(Atman,202

158、2;PMR,2023).Estimated based on IT capacity estimate(Atman,2022;PMR,2023)and adjusted assumptions for utilisation and PUE based on other studies and countries(Ademe&Arcep,2022;Hintemann et al.,2023).Low-medium Portugal None Assumed to consume 2%of national service sector electricity.Low Romania None

159、Assumed to consume 2%of national service sector electricity.Low Slovakia None Assumed to consume 2%of national service sector electricity.Low Slovenia None Assumed to consume 2%of national service sector electricity.Low Spain Data centre IT capacity(SpainDC,2023)Estimated based on IT capacity estima

160、te from SpainDC and adjusted assumptions for utilisation and PUE based on Ademe&Arcep(2022)and Hintemann et al.(2023).Low-medium Sweden Data centre energy use(Swedish Energy Agency,2023);IT capacity(Node Pole&CBRE,2022).Results taken directly from Swedish Energy Agency(2023).High Source:JRC 18 3.3 T

161、elecommunication networks Data collection Country-level telecommunication network energy use estimates were collected from available government sources(see 2.4 Telecommunication networks).Other relevant country-level indicators were collected from Eurostat and the ITU,including population,population

162、 density,GDP,national electricity use,industrial electricity prices,household internet access,mobile internet access rates,and mobile broadband subscriptions.Modelling approach Depending on the available data for a given country,different modelling approaches were used(Table 5):For countries with na

163、tional estimates of telecommunication network energy use from reputable sources using credible methods,this was assumed to be the best available estimate.The estimate year and source are indicated in Table 5,column“National estimate year and source”.Four countries(Belgium,Finland,France,Germany)are

164、estimated using this approach,covering around 33%of mobile broadband subscriptions in the EU-27,as well as 38%of the population and 46%of GDP.If national energy use estimates were not available,a modelled estimate was derived using publicly available data from telecommunication network operators in

165、that country,including corporate sustainability reports and publicly available corporate sustainability disclosures.The approximate coverage of the operator data based on publicly estimated market shares is indicated in Table 5,column“Market share covered by operator data”.Total energy use was extra

166、polated based on the market share covered and reduced by 10%to exclude non-network energy use(e.g.data centres,offices).19 countries were estimated using this approach,covering about two-thirds of mobile broadband subscriptions and 60%of population.If country-level operator data was not available fo

167、r a particular Member State,we used available economic,energy,and digital indicators to derive the estimate.For example,using indicators such as electricity use per internet user or mobile connection from other similar countries.As the energy use of data networks per connection is relatively consist

168、ent across countries(after adjusting for demographic and digital factors),this approach is considered to be a reasonable estimate.The most recent data or estimates available in most cases were from 2021.Adjustments were made based on historical trends to extrapolate these figures to generate 2022 es

169、timates.Table 5.Data sources and modelling approaches taken for telecommunication networks Country Source of national estimate Approach Market share covered by operator data Quality assessment Austria-Operator data 90%High Belgium BIPT(2022)Government estimate 70-80%High Bulgaria-Operator data 60-70

170、%Medium Croatia-Operator data 70-80%Medium-high Cyprus-Based on other indicators-Low-medium Czechia-Operator data 60-70%Medium Denmark-Operator data 80-90%Medium-high Estonia-Operator data 90%High Finland Traficom(2022;2023)Government estimate 90%High France Arcep(2023b)Government estimate 50-60%Hig

171、h Germany German Bundestag Government estimate 90%High 19 Country Source of national estimate Approach Market share covered by operator data Quality assessment(2022)Greece-Operator data 70-80%Medium-high Hungary-Operator data 90%High Ireland-Operator data 70-80%Medium-high Italy-Operator data 70-80%

172、Medium-high Latvia-Operator data 70-80%Medium-high Lithuania-Operator data 60-70%Medium Luxembourg-Based on other indicators-Low-medium Malta-Based on other indicators-Low-medium Netherlands-Operator data 90%High Poland-Operator data 50-60%Medium Portugal-Operator data 60-70%Medium Romania-Operator

173、data 80-90%Medium-high Slovakia-Operator data 80-90%Medium-high Slovenia-Based on other indicators-Medium Spain-Operator data 80-90%Medium-high Sweden-Operator data 90%High Source:JRC 3.4 Summary and limitations The modelling approaches employed in this study are summarised in Table 6.The largest ma

174、rkets rely primarily on published country-level estimates,complemented by modelled estimates based on operator data on electricity use(for networks)and IT capacity estimates(for data centres).Table 6.Overview of modelling approaches Approach Data centres Telecommunication networks Countries Coverage

175、 Countries Coverage Country-level estimates Belgium,Denmark,Finland,France,Germany,Ireland,Netherlands,Sweden 77%of data centre capacity 46%of population 62%of GDP Belgium,Finland,France,Germany 33%of mobile broadband subscriptions 38%of population 46%of GDP Modelled estimate based primarily on IT c

176、apacity(data centres)or operator data(networks)Austria,Italy,Luxembourg,Poland,Spain 19%of data centre capacity 34%of population 28%of GDP Austria,Bulgaria,Croatia,Czechia,Denmark,Estonia,Greece,Hungary,Ireland,Italy,Latvia.Lithuania,Netherlands,Poland,Portugal,Romania,Slovakia,Spain,Sweden 66%of mo

177、bile broadband subscriptions 61%of population 53%of GDP Modelled estimate based Bulgaria,Croatia,Cyprus,Czechia,4%of data centre capacity Cyprus,Luxembourg,Malta,Slovenia 1%of mobile broadband 20 Approach Data centres Telecommunication networks Countries Coverage Countries Coverage on other indicato

178、rs Estonia,Greece,Hungary,Latvia,Lithuania,Malta,Portugal,Romania,Slovakia,Slovenia 19%of population 11%of GDP subscriptions,population,and GDP Source:JRC Data centres Published country-level estimates were available for most of the largest data centre markets covering around three-quarters of the r

179、egions data centre capacity.Country-level IT capacity data(typically covering colocation and hyperscale data centres)were used to estimate around 20%of the overall market.Although the published country-level estimates are considered to be best-available estimates,most of the country-level estimates

180、are modelled.Only the estimates from Ireland and the Netherlands are based primarily on measured(electricity meter)data,covering larger data centres.There are also significant uncertainties in the number and installed capacity of smaller data centres.Telecommunication networks For telecommunication

181、networks,four country-level estimates were available,covering one-third of mobile broadband subscriptions.Data from operators covering the majority of each market were used to develop country-level estimates for countries covering two-thirds of mobile broadband subscriptions.Publicly reported operat

182、or data collated for this analysis accounted for around 20 TWh of electricity use,covering nearly three-quarters of the estimated total EU-27 network energy use.Only 1%of the market was modelled using other indicators.The published country-level estimates are generally considered very robust,since t

183、hey primarily rely on measured electricity data from operators(except Germany).The main limitation of the modelled estimates using operator data stems from challenges in validating the approach used to extrapolate energy use of non-disclosing operators.In addition,this analysis assumes that 90%of op

184、erator electricity use goes to networks.This analysis also excludes energy use in diesel backup generators,estimated to be less than 2%of overall network energy use in Europe based on reported data from several operators in the region.21 4.Results and discussion 4.1 Data centres Data centres in the

185、EU-27 used an estimated 4565 TWh of electricity in 2022,or 1.82.6%of total electricity consumption.This estimate is slightly lower than Montevecchi et al.(2020),which estimated that data centres accounted for 2.7%of total EU28 electricity use in 2018.Large data centres,including colocation and hyper

186、scale data centres,accounted for about 65%of the total,while enterprise data centres accounted for around 35%.These shares are similar to Dodd et al.(2020),which estimated a 56/44 split in 2020 and projected a 66/34 split in 2025.Most of the regions data centre energy use is concentrated in the larg

187、est markets.The top four data centre markets Germany,France,the Netherlands,and Ireland account for nearly two-thirds of the regions data centre energy use,despite having less than 40%of the population(Figure 2).The top twelve markets7,which this study estimated based on country-level data,account f

188、or around 95%the regions data centre energy use.In Germany,data centres used an estimated 15 TWh in 2022,equivalent to around 3%of national electricity use.In France,data centres used around 10 TWh of electricity,equivalent to 2.2%of national electricity use.Data centres represent a significant shar

189、e of national electricity use in Ireland(18%),the Netherlands(5.2%),Luxembourg(4.8%),Denmark(4.5%),and Germany(3%),Sweden(2.3%),and France(2.2%).In all other countries,data centres represent less than the EU-27 average of 2.2%.7 Germany,France,the Netherlands,Ireland,Italy,Spain,Sweden,Poland,Belgiu

190、m,Denmark,Austria,and Finland.22 Figure 2.Estimated data centre energy use by country,2022 Note:Error bars indicate range of estimates.Source:JRC 05101520GermanyFranceNetherlandsIrelandItalySwedenSpainPolandBelgiumDenmarkAustriaFinlandCzechiaGreeceLuxembourgPortugalRomaniaBulgariaHungarySlovakiaEsto

191、niaCroatiaLithuaniaSloveniaLatviaCyprusMaltaShare of national electricity use(%)Data centre electricity use(TWh)23 4.2 Telecommunication networks Telecommunication networks in the EU-27 used an estimated 2530 TWh of electricity in 2022,or 11.2%of total electricity consumption.Based on disaggregated

192、data from a few countries,mobile networks likely accounted for about 60%of the total.These results are in-line with Lundn et al.(2022),which estimated 29 TWh in 2020(EU-27),equivalent to 1.2%of electricity use.The four most populous Member States are also the four largest users of network energy:Ger

193、many,France,Italy,and Spain(Figure 3).These four countries,representing about 58%of the population,account for 65%of the regions telecommunication network energy use.Compared with data centres,energy use by networks as a share of national electricity use was more uniform,ranging from 0.51.5%.In cont

194、rast,data centre energy use ranged from as low as 0.4%in some countries to as high as 18%in Ireland.Data centres accounted for at least 2%of electricity use in eight countries,while telecommunication networks accounted for less than 2%of national electricity use in all countries.24 Figure 3.Estimate

195、d telecommunication network energy use by country,2022 Note:Error bars indicate range of estimates.Source:JRC 012345678GermanyItalyFranceSpainPolandSwedenAustriaNetherlandsGreeceRomaniaFinlandCzechiaBelgiumPortugalHungaryBulgariaCroatiaIrelandSlovakiaDenmarkLithuaniaSloveniaEstoniaLatviaCyprusLuxemb

196、ourgMaltaShare of national electricity use(%)Telecommunication network electricity use(TWh)25 4.3 Summary Data centres in the EU-27 used an estimated 4565 TWh of electricity in 2022,or 1.82.6%of total electricity consumption.Telecommunication networks used an estimated 2530 TWh of electricity in 202

197、2,or 11.2%of total electricity consumption.The relatively wide range for data centres is indicative of the considerable uncertainty in data centre energy estimates stemming from the lack of available data.Data centres and networks digital infrastructure together consumed an estimated 7095 TWh in the

198、 EU-27,equivalent to 2.83.8%of total regional electricity use.The four largest Member States by population and GDP Germany,France,Italy,and Spain account for about 60%of total digital infrastructure energy use in the region(Figure 4).Digital infrastructure accounts for more than 5%of national electr

199、icity use in five countries:Ireland(19%),the Netherlands(6%),Luxembourg(5.5%),and Denmark(5%).Figure 4.Estimated digital infrastructure energy use by country,2022 Source:JRC 26 5.Conclusions and recommendations Summary This study reviews,assesses,and uses published analyses and other public data sou

200、rces to estimate the energy consumption of data centres and telecommunication networks in the EU-27 in 2022.Data centres used an estimated 4565 TWh of electricity in 2022,or 1.82.6%of total EU-27 electricity consumption.Telecommunication networks used an estimated 2530 TWh of electricity in 2022,or

201、11.2%of total electricity consumption.At the country level,the shares of national electricity use varies significantly for data centres,ranging from as low as 0.4%in some countries to as high as 18%in Ireland.Telecommunication network energy use was much more uniform,ranging from 0.5%to 1.5%.While t

202、here are promising developments on data availability and transparency,notably from the largest data centre markets and telecommunication regulators in a growing number of countries,there is an overall lack of rigorous country-level data and studies available across the EU.As digital technologies and

203、 services are evolving quickly,policymakers and companies must work together to improve data collection,quality and availability to develop more robust estimates to make informed policy decisions to manage the energy and environmental impacts of digital infrastructure.Recommendations for future data

204、 collection and modelling Governments and statistical agencies should develop standardised definitions and classifications.For example,developing clear definitions of what is considered a data centre,and providing criteria and guidance on classifying different data centre types.Standardised definiti

205、ons and classifications are essential to compare and combine energy consumption estimates from different studies and countries.Governments and companies should work together to improve data quality and availability to improve the quality of future estimates.For example,the following data should be c

206、ollected at the country level:Data centre energy consumption by size and type(e.g.enterprise,colocation,hyperscale,telecom,edge,etc.);Telecommunication network energy consumption by type(e.g.mobile,fixed,core);Relevant activity indicators,e.g.fixed connections,mobile connections,mobile network cover

207、age,network data traffic,data centre workloads(and type of tasks).Governments,companies,and researchers should work together to develop and standardise a suite of appropriate energy intensity indicators and metrics to inform future modelling efforts.For example,energy intensity of mobile networks ba

208、sed on data traffic,subscriptions,and coverage area.Further work is needed to standardise measurement methodologies and indicators.Data collection efforts should also seek to better understand energy use characteristics and implications of specific services and tasks,such as AI,streaming media,and a

209、ugmented and virtual reality.Improved understanding of data centre and network energy use for specific services can help improve understanding of end-to-end energy use of specific services such as streaming video,and help prioritise measures to reduce energy use(e.g.actual impacts of lower bitrates

210、or increased compression on networks as well as end-user devices).Increased use of AI is likely to be an important driver of growing demand for data centre services over the coming years.AI likely accounts for less than one-quarter of data centre energy use today(Kaack et al.,2022),but the widesprea

211、d adoption of AI by businesses and consumers could see this share rise(Minde,2023).The greater use of graphics processing units(GPU)and application-specific integrated circuits(ASIC)for AI are expected to increase the power density of data centre racks,increasing cooling needs.There is a very limite

212、d understanding of AI-related energy use in data centres,including the energy use in different 27 stages of the model life cycle(development,training,and inference),as well as the share of energy use in data centres associated with AI.Additional data from companies and methodology development from r

213、esearchers is needed.Future studies that estimate the energy use of data centres and telecommunication networks should clearly and comprehensively document and disclose data sources,assumptions,and other methodological details.The lack of available details in previous studies makes it difficult to a

214、ssess the quality of analysis and compare results.Increasing the consistency of baseline data and methodologies can also improve comparability to understand energy trends over time.28 References 4E EDNA.(2019).Total Energy Model for Connected Devices.https:/www.iea-4e.org/wp-content/uploads/publicat

215、ions/2019/06/A2b_-_EDNA_TEM_Report_V1.0.pdf 4E EDNA.(2021).Total Energy Model V2.0 for Connected Devices.https:/www.iea-4e.org/document/429/total-energy-model-for-connected-devicesahttps:/www.iea-4e.org/wp-content/uploads/publications/2021/02/EDNA-TEM2.0-Report-V1.0-Final.pdf Ademe&Arcep.(2022).Eval

216、uation de limpact environnemental du numrique en France et analyse prospective.https:/www.arcep.fr/uploads/tx_gspublication/etude-numerique-environnement-ademe-arcep-volet02_janv2022.pdf AFP.(2019,October 28).Chill your Netflix habit,climate experts say.France 24.https:/ Andrae,A.(2019).Comparison o

217、f Several Simplistic High-Level Approaches for Estimating the Global Energy and Electricity Use of ICT Networks and Data Centers.International Journal of Green Technology,5,5063.https:/doi.org/10.30634/2414-2077.2019.05.06 Andrae,A.(2020).New perspectives on internet electricity use in 2030.Engineer

218、ing and Applied Science Letter,3(2),1931.https:/doi.org/10.30538/psrp-easl2020.0038 Andrae,A.,&Edler,T.(2015).On Global Electricity Usage of Communication Technology:Trends to 2030.Challenges,6(1),117157.https:/doi.org/10.3390/challe6010117 Arcep.(2022).Enqute annuelle Pour un numrique soutenable Ed

219、ition 2022.https:/www.arcep.fr/fileadmin/cru-1677573101/user_upload/observatoire/enquete-pns/edition-2022/enquete-annuelle-pour-un-numerique-soutenable_edition2022.pdf Arcep.(2023a).Achieving digital sustainability:Arcep publishes the second edition of its annual inquiry.https:/en.arcep.fr/fileadmin

220、/cru-1677573101/user_upload/22-23-english-version.pdf Arcep.(2023b).Enqute annuelle Pour un numrique soutenable 2me dition.https:/www.arcep.fr/fileadmin/cru-1677573101/user_upload/observatoire/enquete-pns/edition-2023/enquete-annuelle-pour-un-numerique-soutenable_edition2023.pdf Aslan,J.(2020).Clima

221、te change implications of gaming products and services University of Surrey.https:/doi.org/10.15126/thesis.00853729 Atman.(2022,June 23).Current state and forecasts for data center market in Poland.https:/www.atman.pl/en/blog-post/current-state-and-forecasts-for-data-center-market-in-poland/Bashrous

222、h,R.(2018).Datacenter EURECA Project Final Report.https:/ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e5bb754090&appId=PPGMS Baxtel.(2023a).Austria Data Centers&Colocation.https:/ Baxtel.(2023b).Luxembourg Data Centers&Colocation.https:/ Belgian Digital Infrastructur

223、e Association.(2022).State of the Belgian Data Centers 2022.https:/bdia.be/en/insights/state-of-the-belgian-data-center-2022/Belkhir,L.,&Elmeligi,A.(2018).Assessing ICT global emissions footprint:Trends to 2040&recommendations.Journal of Cleaner Production,177,448463.https:/doi.org/10.1016/j.jclepro

224、.2017.12.239 BEREC.(2023).Draft BEREC Report on Sustainability Indicators for Electronic Communications Networks and Services.https:/www.berec.europa.eu/system/files/2023-03/BoR%20%2823%29%2046%20Draft%20Report%20on%20sustainability%20indicators%20for%20ECN%20ECS%20%20%281%29.pdf Bieser,J.,Salieri,B

225、.,Hischier,R.,&Hilty,L.M.(2020).Next generation mobile networks:Problem or opportunity for climate protection?BIPT.(2022).Sustainability of telecommunication networks and operators in Belgium.https:/www.bipt.be/consumers/publication/sustainability-of-telecommunication-networks-and-operators-in-belgi

226、um 29 BloombergNEF,Statkraft,&Eaton.(2021).Data Centers and Decarbonization.https:/ Bordage,F.(2019).The environmental footprint of the digital world.GreenIT.fr.https:/www.greenit.fr/wp-content/uploads/2019/11/GREENIT_EENM_etude_EN_accessible.pdf Bordage,F.,de Montenay,L.,&Vergeynst,O.(2021).Impacts

227、 environnementaux du numrique en France.https:/www.greenit.fr/wp-content/uploads/2021/02/2021-01-iNum-etude-impacts-numerique-France-rapport-0.8.pdf Cambridge Centre for Alternative Finance.(2023).Cambridge Bitcoin Electricity Consumption Index(CBECI).https:/www.cbeci.org/Cardoso,A.(2020).Reducing t

228、he energy use of video gaming:Energy efficiency and gamification.The Copenhagen Centre on Energy Efficiency.https:/c2e2.unepccc.org/wp-content/uploads/sites/3/2020/10/reducing-the-energy-use-of-video-gaming-energy-efficiency-and-gamification-en.pdf CCRI.(2022).The MergeImplications on the Electricit

229、y Consumption and Carbon Footprint of the Ethereum Network.https:/carbon- CCRI.(2023).CCRI Indices.https:/indices.carbon- Statistics Office,Ireland.(2021).Data Centres Metered Electricity Consumption 2020.CSO.https:/www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsum

230、ption2020/Central Statistics Office,Ireland.(2022).Data Centres Metered Electricity Consumption 2021.CSO.https:/www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2021/keyfindings/Central Statistics Office,Ireland.(2023).Data Centres Metered Electricity Consumpt

231、ion 2022.CSO.https:/www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2022/Cisco.(2018).Cisco Global Cloud Index:Forecast and Methodology,20162021.Cisco.(2019).Cisco Visual Networking Index:Forecast and Trends,20172022 White Paper.http:/ CITIZING.(2020).Emprein

232、te carbone du numrique en France Snat.https:/www.senat.fr/fileadmin/Fichiers/Images/commission/Developpement_durable/MI_empreinte_environnementale/r19-555-annexe.pdf Compton,K.(2018).Ciscos Global Cloud Index Study:Acceleration of the Multicloud Era Cisco.https:/ Coroam,V.C.(2021).Investigating the

233、Inconsistencies among Energy and Energy Intensity Estimates of the Internet:Metrics and Harmonising Values.Swiss Federal Office of Energy.https:/www.aramis.admin.ch/Default?DocumentID=67656&Load=true COWI.(2021).Udviklingen af datacentre og deres indvirkning p energisystemet.https:/ens.dk/sites/ens.

234、dk/files/Basisfremskrivning/cowi_-_udviklingen_for_datacentre_og_deres_indvirkning_paa_energisystemet.pdf Danish Energy Agency.(2021).Analyseforudstninger til Energinet 2021 Datacentre Baggrundsnotat.https:/ens.dk/sites/ens.dk/files/Hoeringer/baggrundsnotat_-_datacentre.pdf Danish Energy Agency.(202

235、2a).Denmarks Climate Status and Outlook 2022.https:/ens.dk/sites/ens.dk/files/Forskning_og_udvikling/cso22_-_english_translation_of_kf22_hovedrapport.pdf Danish Energy Agency.(2022b).Klimastatus og fremskrivning 2022(KF22):Datacentre Forudstningsnotat nr.6A.https:/ens.dk/sites/ens.dk/files/Analyser/

236、6a_kf22_forudsaetningsnotat_-_datacentre.pdf Danish Energy Agency.(2023a).Klimastatus og fremskrivning 2023.https:/ens.dk/sites/ens.dk/files/Basisfremskrivning/kf23_hovedrapport.pdf 30 Danish Energy Agency.(2023b).Klimastatus og fremskrivning 2023(KF23):Serviceerhverv.https:/ens.dk/sites/ens.dk/file

237、s/Basisfremskrivning/kf23_sektornotat_5a_serviceerhverv.pdf DeepMind.(2016).DeepMind AI Reduces Google Data Centre Cooling Bill by 40%.https:/ for Internal Market,I.,Fraunhofer IZM,Bayramoglu,S.,Nissen,N.,Berwald,A.,Faninger,T.,Stobbe,L.,Mudgal,S.,&Tinetti,B.(2014).Ecodesign preparatory study on ent

238、erprise servers and data equipment.Publications Office of the European Union.https:/data.europa.eu/doi/10.2873/14639 Deloitte Tohmatsu MIC Research Institute.(2022).省化向市場実態將來予測【2022年度版】|経済研究所.https:/mic-r.co.jp/mr/02480/Dodd,N.,Alfieri,F.,De,O.G.C.M.N.,Maya-Drysdale,L.,Viegand,J.,Flucker,S.,Tozer,R.

239、,Whitehead,B.,Wu,A.,&Brocklehurst,F.(2020,June 8).Development of the EU Green Public Procurement(GPP)Criteria for Data Centres,Server Rooms and Cloud Services.JRC Publications Repository.https:/doi.org/10.2760/964841 Dutch Data Center Association.(2022).State of the Dutch Data Centers 2022.https:/ww

240、w.dutchdatacenters.nl/publicaties/sodd-2022/Dutch Data Center Association.(2023).State of the Dutch Data Centers 2023.https:/www.dutchdatacenters.nl/publicaties/state-of-the-dutch-data-centers-2023/EC DG Energy,Louguet,A.,Caspani,M.,Pytel,D.,Pirlot,A.,Faura Rosendo,M.,&Blanadet,H.(2023).Assessment o

241、f the energy footprint of digital actions and services.Publications Office of the European Union.https:/data.europa.eu/doi/10.2833/478689 Ericsson.(2022).Ericsson Mobility Report November 2022.https:/ ethereum.org.(2023).The Merge.https:/ethereum.org/en/upgrades/merge/Eurostat.(2023).Energy statisti

242、csAn overview.Energy Statistics-an Overview.https:/ec.europa.eu/eurostat/statistics-explained/index.php?title=Energy_statistics_-_an_overview Fan,F.(2021).Green data centers in focus.China Daily.https:/ Fernandez,R.(2023,January 27).Tech leaders launch the Italian Data Center Association for growing

243、 European market.TechRepublic.https:/ Consumption of Cryptocurrencies Beyond Bitcoin.Joule,4(9),18431846.https:/doi.org/10.1016/j.joule.2020.07.013 German Bundestag.(2022).Energy consumption of ICT infrastructure.KIT.https:/doi.org/10.5445/IR/1000152733 Google.(2022).Google Environmental Report 2022

244、.https:/ Greenpeace East Asia.(2021).China 5G and Data Center Carbon Emissions Outlook 2035.https:/www.greenpeace.org/static/planet4-eastasia-stateless/2021/05/a5886d59-china-5g-and-data-center-carbon-emissions-outlook-2035-english.pdf Greenpeace East Asia&North China Electric Power University.(2019

245、).Powering the Cloud:How Chinas Internet Industry Can Shift to Renewable Energy.https:/www.greenpeace.org/static/planet4-eastasia-stateless/2019/11/7bfe9069-7bfe9069-powering-the-cloud-_-english-briefing.pdf GSMA.(2022).Mobile Net Zero:State of the Industry on Climate Action 2022.https:/ Haut Consei

246、l pour le Climat.(2020).Matriser limpact carbone de la 5G.https:/www.hautconseilclimat.fr/publications/maitriser-limpact-carbone-de-la-5g/31 Hiekkanen,K.,Seppala,T.,&Ylhinen,I.(2021).Energy and Electricity Consumption of the Information Economy Sector in Finland.The Research Institute of the Finnish

247、 Economy(ETLA).https:/www.econstor.eu/bitstream/10419/251075/1/ETLA-Raportit-Reports-107.pdf Hintemann,R.,&Hinterholzer,S.(2020).Data centers 2018.Borderstep Institute.https:/www.borderstep.de/wp-content/uploads/2020/04/Borderstep-Datacenter-2018_en.pdf Hintemann,R.,&Hinterholzer,S.(2022).Data cente

248、rs 2021.Borderstep Institute.https:/www.borderstep.org/wp-content/uploads/2022/08/Borderstep_Rechenzentren_2021_eng.pdf Hintemann,R.,Hinterholzer,S.,&Seibel,H.(2023).Studie Rechenzentren in Deutschland:Aktuelle Marktentwicklungen Update 2023.Berlin:Borderstep Institut.Bitkom.https:/www.bitkom.org/si

249、tes/main/files/2023-05/BitkomStudieRechenzentreninDeutschland2023.pdf IEA.(2017).Digitalization&Energy.https:/www.iea.org/reports/digitalisation-and-energy IEA.(2021).Data Centres and Data Transmission Networks(2021).https:/web.archive.org/web/20211119195429/https:/www.iea.org/reports/data-centres-a

250、nd-data-transmission-networks IEA.(2023a).Data Centres and Data Transmission Networks.https:/www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks IEA.(2023b).Data and statistics.Data and Statistics.https:/www.iea.org/data-and-statistics ITU.(2018).L.1450:Methodologies for

251、the assessment of the environmental impact of the information and communication technology sector.https:/www.itu.int/rec/T-REC-L.1450-201809-I/en ITU.(2020).L.1470:Greenhouse gas emissions trajectories for the information and communication technology sector compatible with the UNFCCC Paris Agreement

252、.https:/www.itu.int/rec/T-REC-L.1470-202001-I/en ITU.(2023).Key ICT indicators for the world and special regions(totals and penetration rates).https:/www.itu.int:443/en/ITU-D/Statistics/Pages/stat/default.aspx Johnson,D.(2018).The 5G Dilemma:More Base Stations,More AntennasLess Energy?IEEE Spectrum.

253、https:/spectrum.ieee.org/energywise/telecom/wireless/will-increased-energy-consumption-be-the-achilles-heel-of-5g-networks Kaack,L.H.,Donti,P.L.,Strubell,E.,Kamiya,G.,Creutzig,F.,&Rolnick,D.(2022).Aligning artificial intelligence with climate change mitigation.Nature Climate Change,12(6),Article 6.h

254、ttps:/doi.org/10.1038/s41558-022-01377-7 Kamiya,G.(2020a).Factcheck:What is the carbon footprint of streaming video on Netflix?Carbon Brief.https:/www.carbonbrief.org/factcheck-what-is-the-carbon-footprint-of-streaming-video-on-netflix Kamiya,G.(2020b).The carbon footprint of streaming video:Fact-ch

255、ecking the headlines.IEA.https:/www.iea.org/commentaries/the-carbon-footprint-of-streaming-video-fact-checking-the-headlines Koot,M.,&Wijnhoven,F.(2021).Usage impact on data center electricity needs:A system dynamic forecasting model.Applied Energy,291,116798.https:/doi.org/10.1016/j.apenergy.2021.1

256、16798 Lacoste,A.,Luccioni,A.,Schmidt,V.,&Dandres,T.(2019).Quantifying the Carbon Emissions of Machine Learning.https:/mlco2.github.Luccioni,A.,Lacoste,A.,&Schmidt,V.(2020).Estimating carbon emissions of artificial intelligence.IEEE Technology and Society Magazine,39(2),4851.https:/doi.org/10.1109/MT

257、S.2020.2991497 Ludvigsen,K.G.A.(2023a,March 5).ChatGPTs electricity consumption,pt.II.Medium.https:/ Ludvigsen,K.G.A.(2023b,July 12).ChatGPTs Electricity Consumption.Medium.https:/ Lundn,D.,Malmodin,J.,Bergmark,P.,&Lvehagen,N.(2022).Electricity Consumption and Operational Carbon Emissions of Europea

258、n Telecom Network Operators.Sustainability 2022,Vol.14,Page 2637,14(5),Article 5.https:/doi.org/10.3390/SU14052637 Luo,J.,Paduraru,C.,Voicu,O.,Chervonyi,Y.,Munns,S.,Li,J.,Qian,C.,Dutta,P.,Davis,J.Q.,Wu,N.,Yang,X.,Chang,C.-M.,Li,T.,Rose,R.,Fan,M.,Nakhost,H.,Liu,T.,Kirkman,B.,Altamura,F.,Mankowitz,D.J

259、.(2022).32 Controlling Commercial Cooling Systems Using Reinforcement Learning(arXiv:2211.07357).arXiv.http:/arxiv.org/abs/2211.07357 Makonin,S.,Marks,L.U.,Przedpeski,R.,Rodriguez-Silva,A.,&ElMallah,R.(2022).Calculating the Carbon Footprint of Streaming Media:Beyond the Myth of Efficiency.LIMITS.Com

260、puting within Limits.https:/computingwithinlimits.org/2022/papers/limits22-final-Makonin.pdf Malmodin,J.(2020).The power consumption of mobile and fixed network data servicesThe case of streaming video and downloading large files.Electronics Goes Green 2020+,8796.https:/online.electronicsgoesgreen.o

261、rg/wp-content/uploads/2020/10/Proceedings_EGG2020_v2.pdf Malmodin,J.,Lvehagen,N.,Bergmark,P.,&Lundn,D.(2023).ICT Sector Electricity Consumption and Greenhouse Gas Emissions 2020 Outcome(SSRN Scholarly Paper 4424264).https:/ Malmodin,J.,&Lundn,D.(2018).The Energy and Carbon Footprint of the Global IC

262、T and E&M Sectors 20102015.Sustainability,10(9),Article 9.https:/doi.org/10.3390/su10093027 Marks,L.U.,Clark,J.,Livingston,J.,Oleksijczuk,D.,&Hilderbrand,L.(2020).Streaming Medias Environmental Impact.Media+Environment,2(1).https:/doi.org/10.1525/001c.17242 Masanet,E.,Shehabi,A.,Lei,N.,Smith,S.,&Koo

263、mey,J.(2020).Recalibrating global data center energy-use estimates.Science,367(6481),Article 6481.https:/doi.org/10.1126/science.aba3758 McDonald,K.(2022).Ethereum Emissions:A Bottom-up Estimate.https:/kylemcdonald.github.io/ethereum-emissions/Meta.(2022).2021 Sustainability Report.https:/ Mills,E.,

264、Bourassa,N.,Rainer,Leo,Mai,J.,Shehabi,A.,&Mills,N.(2019).Toward Greener Gaming:Estimating National Energy Use and Energy Efficiency Potential Energy Access View project Energy modeling View project Toward Greener Gaming:Estimating National Energy Use and Energy Efficiency Potential.https:/doi.org/10

265、.1007/s40869-019-00084-2 Minde,Tor Bjrn.(2023,October 8).Generative AI does not run on thin air!RISE.https:/www.ri.se/en/news/blog/generative-ai-does-not-run-on-thin-air Montevecchi,F.,Stickler,T.,Hintemann,R.,&Hinterholzer,S.(2020).Energy-efficient cloud computing technologies and policies for an e

266、co-friendly cloud market:Final study report.Publications Office.https:/data.europa.eu/doi/10.2759/3320 Mordor Intelligence.(2023).Austria Data Center Market Size&Share Analysis.https:/ Moulierac,J.,Urvoy-Keller,G.,Dinuzzi,M.,&Ma,Z.(2023).What is the carbon footprint of one hour of video streaming?Un

267、iversit Cte dAzur.https:/hal.science/hal-04069500 Mytton,D.,&Ashtine,M.(2022).Sources of data center energy estimates:A comprehensive review.Joule,6(9),20322056.https:/doi.org/10.1016/j.joule.2022.07.011 National Grid ESO.(2022).Data Centres:What are data centres and how will they influence the futu

268、re energy system?https:/ Naumov,M.,Kim,J.,Mudigere,D.,Sridharan,S.,Wang,X.,Zhao,W.,Yilma,S.,Kim,C.,Yuen,H.,Ozdal,M.,Nair,K.,Gao,I.,Su,B.Y.,Yang,J.,&Smelyanski,M.(2020).Deep learning training in facebook data centers:Design of scale-up and scale-out systems.110.https:/arxiv.org/abs/2003.09518 Nikkei.

269、(2022,October 19).経済研究所、省化向市場実態將來予測2022年度版発刊.日本経済新聞.https:/ Pole&CBRE.(2022).Data Centres in Sweden.https:/8866495.fs1.hubspotusercontent- NVE.(2020).Kartlegging og vurdering av potensial for effektivisering av oppvarming og kjling i Norge.https:/www.regjeringen.no/contentassets/f3f73a50bf32441a8485

270、cce8f4ce3bcf/rapport-fra-oslo-economics-og-asplan-viak-l1140225.pdf 33 OpenAI.(2018).AI and compute.https:/ Orange.(2022).5G and energy efficiency:New mechanisms for progress.Orange Hello Future.https:/ Learning Inference in Facebook Data Centers:Characterization,Performance Optimizations and Hardwa

271、re Implications.https:/arxiv.org/abs/1811.09886 Patterson,D.,Gonzalez,J.,Hlzle,U.,Le,Q.,Liang,C.,Munguia,L.-M.,Rothchild,D.,So,D.,Texier,M.,&Dean,J.(2022).The Carbon Footprint of Machine Learning Training Will Plateau,Then Shrink(arXiv:2204.05149).arXiv.https:/doi.org/10.48550/arXiv.2204.05149 PMR.(

272、2023).Data center market in Poland 2023.https:/mypmr.pro/products/data-centre-market-in-poland Politecnico di Milano,School of Management.(2022).I Data Center per lo sviluppo dellecosistema digitale italiano.https:/ Prakash,S.,Baron,Y.,Liu,R.,Proske,M.,&Schlsser,A.(2014).Study on the practical appli

273、cation of the new framework methodology for measuring the environmental impact of ICT cost/benefit analysis(SMART 2012/0064).https:/ Radar.(2020).Datacenter i Sverige 2020.https:/www.almega.se/app/uploads/sites/2/2022/09/radar-datacenterlandskapet-sverige-2020.pdf Schneider Electric.(2021).Digital E

274、conomy and Climate Impact.https:/ Schwartz,R.,Dodge,J.,Smith,N.A.,&Etzioni,O.(2019).Green AI.Communications of the ACM,63(12),5463.https:/doi.org/10.1145/3381831 Shehabi,A.,Smith,S.J.,Masanet,E.,&Koomey,J.(2018).Data center growth in the United States:Decoupling the demand for services from electric

275、ity use.Environmental Research Letters,13(12),124030.https:/doi.org/10.1088/1748-9326/aaec9c Shehabi,A.,Smith,S.,Sartor,D.,Brown,R.,Herrlin,M.,Koomey,J.,Masanet,E.,Horner,N.,Azevedo,I.,&Lintner,W.(2016).United States Data Center Energy Usage Report(LBNL-1005775,1372902;p.LBNL-1005775,1372902).https:

276、/doi.org/10.2172/1372902 Singapore Ministry of Communications and Information.(2021).MCIs response to PQ on data on current and expected 2021 total carbon emissions by data centres in Singapore and efforts to reduce emissions for data centres.https:/www.mci.gov.sg/pressroom/news-and-stories/pressroo

277、m/2021/7/mci-response-to-pq-on-data-on-current-and-expected-2021-total-carbon-emissions-by-data-centres-in-singapore-and-efforts-to-reduce-emissions-for-data-centres SpainDC.(2023).Ecosistema espaol de data centers.https:/ 28).Climate change activists are coming for your binge watch.New York Post.ht

278、tps:/ Finland.(2022).Energy Accounts by Year,Industries,households and other items,Energy product and Information dataset.https:/pxdata.stat.fi:443/PxWebPxWeb/pxweb/en/StatFin/StatFin_entp/statfin_entp_pxt_11wx.px/Statistics Netherlands.(2021a).Electricity supplied to data centres,2017-2019 Webpagin

279、a.https:/www.cbs.nl/en-gb/custom/2020/51/electricity-supplied-to-data-centres-2017-2019 Statistics Netherlands.(2021b).Elektriciteit geleverd aan datacenters,2017-2020 Webpagina.https:/www.cbs.nl/nl-nl/maatwerk/2021/50/elektriciteit-geleverd-aan-datacenters-2017-2020 STL Partners.(2019).Curtailing c

280、arbon emissions can 5G help?34 Strubell,E.,Ganesh,A.,&McCallum,A.(2019).Energy and Policy Considerations for Deep Learning in NLP.Computation and Language(cs.CL).https:/arxiv.org/abs/1906.02243v1 Swedish Energy Agency.(2023).Energianvndning i datacenter och digitala system.https:/www.energimyndighet

281、en.se/492f27/contentassets/054d98cfdcd54cb5a802e24b53779452/energianvandning-i-datacenter-och-digitala-system.pdf TeleGeography.(2022).The State of the Network 2022 Edition.https:/ TeleGeography.(2023).The State of the Network 2023.https:/ The Carbon Trust.(2021).Carbon impact of video streaming.htt

282、ps:/ The Shift Project.(2019a).Climate Crisis:The Unsustainable User of Online Video.https:/theshiftproject.org/en/article/unsustainable-use-online-video/The Shift Project.(2019b).Lean ICT:Towards Digital Sobriety.The Shift Project.https:/theshiftproject.org/en/article/lean-ict-our-new-report/The Sh

283、ift Project.(2020a).Dployer la sobrit numrique.https:/theshiftproject.org/article/deployer-la-sobriete-numerique-rapport-shift/The Shift Project.(2020b).Did The Shift Project really overestimate the carbon footprint of online video?The Shift Project.(2021).Impact environnemental du numrique:Tendance

284、s 5 ans et gourvernance de la 5G.https:/theshiftproject.org/article/impact-environnemental-du-numerique-5g-nouvelle-etude-du-shift/Traficom.(2022).First study on the energy consumption of communications networks.https:/www.traficom.fi/en/news/first-study-energy-consumption-communications-networks Tr

285、aficom.(2023).Traficom Sustainability Report 2022.https:/www.traficom.fi/sites/default/files/media/file/TRAFICOM_Vastuullisuusraportti-2022_ENG.pdf Vereecken,W.,Deboosere,L.,Simoens,P.,Vermeulen,B.,Colle,D.,Develder,C.,Pickavet,M.,Dhoedt,B.,&Demeester,P.(2010).Energy Efficiency in Thin Client Soluti

286、ons.In A.Doulamis,J.Mambretti,I.Tomkos,&T.Varvarigou(Eds.),Networks for Grid Applications(pp.109116).Springer.https:/doi.org/10.1007/978-3-642-11733-6_12 VHK&Viegand Maage.(2020).ICT Impact Study.European Commission.https:/susproc.jrc.ec.europa.eu/product-bureau/sites/default/files/2020-11/IA_report

287、-ICT_study_final_2020_(CIRCABC).pdf Vries,A.de.(2023).The growing energy footprint of artificial intelligence.Joule,7(10),21912194.https:/doi.org/10.1016/j.joule.2023.09.004 Williams,L.,Sovacool,B.K.,&Foxon,T.J.(2022).The energy use implications of 5G:Reviewing whole network operational energy,embod

288、ied energy,and indirect effects.Renewable and Sustainable Energy Reviews,157,112033.https:/doi.org/10.1016/j.rser.2021.112033 Wu,C.-J.,Raghavendra,R.,Gupta,U.,Acun,B.,Ardalani,N.,Maeng,K.,Chang,G.,Aga,F.,Huang,J.,Bai,C.,Gschwind,M.,Gupta,A.,Ott,M.,Melnikov,A.,Candido,S.,Brooks,D.,Chauhan,G.,Lee,B.,L

289、ee,H.-H.,Hazelwood,K.(2022).Sustainable AI:Environmental Implications,Challenges and Opportunities.Proceedings of Machine Learning and Systems,4,795813.https:/proceedings.mlsys.org/paper/2022/hash/ed3d2c21991e3bef5e069713af9fa6ca-Abstract.html 35 List of figures Figure 1.Summary of European data cen

290、tre energy estimates.9 Figure 2.Estimated data centre energy use by country,2022.22 Figure 3.Estimated telecommunication network energy use by country,2022.24 Figure 4.Estimated digital infrastructure energy use by country,2022.25 36 List of tables Table 1.Global trends in digital and energy indicat

291、ors,20152022.4 Table 2.Overview studies estimating the energy use of data centres in Europe since 2018.7 Table 3.Overview studies estimating the energy use of data centres in Europe since 2018.10 Table 4.Data sources and modelling approaches taken for data centres.16 Table 5.Data sources and modelli

292、ng approaches taken for telecommunication networks.18 Table 6.Overview of modelling approaches.19 37 Annex Table A.1.Overview of studies estimating the global energy use of data centres since 2018 Institution Publications Approach Estimates*Borderstep Institute Hintemann&Hinterholzer(2020;2022)Botto

293、m-up estimate based on data centre market developments(primarily in Europe),technical characteristics of servers,storage,and networking(energy use,age)and data centre infrastructure(air conditioning,power supply,UPS).310-330 TWh in 2018(400 TWh including crypto)270-380 TWh in 2020(350-500 TWh includ

294、ing crypto)Ericsson;Telia Malmodin(2020);Malmodin et al.(2023);Malmodin&Lundn(2018)Hybrid estimate based on bottom-up estimates based on hardware shipments,complemented by benchmarking to other studies and reported company data.220 TWh in 2015(245 TWh including enterprise networks)223 TWh in 2020 Gr

295、eenIT.fr Bordage(2019)Based on the number of servers in operation and LCAs of three different data centres.312 TWh in 2019 Huawei Andrae(2019;2020)Extrapolation based on Andrae&Edler(2015)with data centre IP traffic extrapolations and energy intensity per unit of IP traffic under updated efficiency

296、improvement scenarios.211 TWh in 2018 196-299 TWh in 2020 International Energy Agency(IEA)IEA(2023a)and previous versions in 2018-2022 Hybrid estimate based on the bottom-up modelling in IEA(2017),Masanet et al.(2020),and Hintemann&Hinterholzer(2022)complemented with reported energy consumption data

297、 from large data centre operators.200 TWh in 2018 200-250 TWh in 2020 220-320 TWh in 2021 International Telecommunications Union(ITU)ITU(2020)Based on the modelling of Malmodin&Lundn(2018)and input from Andrae(2019).14/02/2024 15:13:00 220 TWh in 2015 230 TWh in 2020(projection)Lawrence Berkeley Nat

298、ional Laboratory;Northwestern University;UC Santa Barbara Masanet et al.(2020)Bottom-up estimate based on shipment data for servers,drives,networking,their energy use characteristics and lifetimes,combined with assumptions for each type of data centre class and region-specific PUE.205 TWh in 2018 Mc

299、Master University Belkhir&Elmeligi(2018)Extrapolation of data centre energy use estimate for 2008 from Vereecken et al.(2010)increasing by 12%per year.704 TWh in 2017 990 TWh in 2020(projection)Schneider Electric Schneider Electric(2021)Bottom-up estimate based on workloads,data storage requirements

300、,and global average PUE.341 TWh in 2020 The Shift Project The Shift Project(2019b;2021)Based on the model developed by Andrae&Edler(2015)with updated assumptions and scenarios.559-593 TWh in 2017 393 TWh in 2019(438 TWh including crypto)University of Twente Koot&Wijnhoven(2021)Hybrid approach combin

301、ing top-down indicators and bottom-up data(e.g.workloads per application).286 TWh in 2016 240-275 TWh in 2020 Source:JRC GETTING IN TOUCH WITH THE EU In person All over the European Union there are hundreds of Europe Direct centres.You can find the address of the centre nearest you online(european-u

302、nion.europa.eu/contact-eu/meet-us_en).On the phone or in writing Europe Direct is a service that answers your questions about the European Union.You can contact this service:by freephone:00 800 6 7 8 9 10 11(certain operators may charge for these calls),at the following standard number:+32 22999696,

303、via the following form:european-union.europa.eu/contact-eu/write-us_en.FINDING INFORMATION ABOUT THE EU Online Information about the European Union in all the official languages of the EU is available on the Europa website(european-union.europa.eu).EU publications You can view or order EU publicatio

304、ns at op.europa.eu/en/publications.Multiple copies of free publications can be obtained by contacting Europe Direct or your local documentation centre(european-union.europa.eu/contact-eu/meet-us_en).EU law and related documents For access to legal information from the EU,including all EU law since 1

305、951 in all the official language versions,go to EUR-Lex(eur-lex.europa.eu).Open data from the EU The portal data.europa.eu provides access to open datasets from the EU institutions,bodies and agencies.These can be downloaded and reused for free,for both commercial and non-commercial purposes.The portal also provides access to a wealth of datasets from European countries.