Data centres: must we choose between digital tech and ecology?

In Europe, data cen­tres are becom­ing hubs of dig­i­tal devel­op­ment; dri­ven by the wide­spread adop­tion of cloud tech­nol­o­gy, a rise in online ser­vices and the grow­ing inte­gra­tion of arti­fi­cial intel­li­gence into pub­lic and pri­vate ser­vices. Their pro­lif­er­a­tion is accom­pa­nied with an unprece­dent­ed lev­el of invest­ment: accord­ing to Euronews, plans to set up data cen­tres have increased by 168% in one year, focus­ing around hubs in Lon­don, Frank­furt and Paris¹. The phe­nom­e­non is linked to dig­i­tal sov­er­eign­ty and eco­nom­ic and fis­cal issues, with each Mem­ber State seek­ing to attract these strate­gic foundations.

At the same time, their impact on local resources is begin­ning to show. The Euro­pean Com­mis­sion esti­mates that data cen­tres account for near­ly 3% of the elec­tric­i­ty con­sumed in the Euro­pean Union². In Ire­land, this share will reach 21% accord­ing to fig­ures from the nation­al ener­gy author­i­ty for 2023³. Oth­er ten­sions are emerg­ing, par­tic­u­lar­ly in Madrid and Mar­seille, where sev­er­al projects have raised con­cerns about their water con­sump­tion for cool­ing, giv­en the fre­quent droughts in the region⁴.

Faced with these changes, the Euro­pean Com­mis­sion is impos­ing trans­paren­cy require­ments on ener­gy and water con­sump­tion through the Ener­gy Effi­cien­cy Direc­tive, to con­tribute to the tar­get of an 11.7% reduc­tion by 2030⁵. In response, some indus­tri­al ini­tia­tives are seek­ing to reduce their foot­print through heat recov­ery for dis­trict heat­ing, pas­sive cool­ing and soft­ware opti­mi­sa­tion, among oth­er mea­sures. In Saint-Denis, for exam­ple, heat from a data cen­tre sup­plies the urban net­work, while oth­er projects are look­ing at link­ing it to trans­port or leisure sys­tems⁶.

In order to sort out the facts from the fic­tion regard­ing these inter­twined con­cerns of tech­no­log­i­cal devel­op­ment, envi­ron­men­tal con­straints and ter­ri­to­r­i­al strate­gies, two researchers offer some explana­to­ry insights. Cécile Diguet, an urban plan­ner at Stu­dio Dégel who head­ed the urban plan­ning and devel­op­ment depart­ment at the Paris Region Insti­tute for nine years, co-authored Sous le feu numérique. Spa­tial­ités et éner­gies des data cen­tres (Under dig­i­tal fire: Spa­tial­i­ties and ener­gies of data cen­tres), pub­lished in the jour­nal Géo­car­refour⁷. Clé­ment Mar­quet, a soci­ol­o­gist at the CNRS, is inter­est­ed in the mate­ri­al­i­ty of dig­i­tal infra­struc­ture. He is the author of Ce nuage que je ne saurais voir. Pro­mou­voir, con­tester et réguler les data cen­ters à Plaine Com­mune (The cloud I can­not see: pro­mot­ing, con­test­ing and reg­u­lat­ing data cen­tres in Plaine Com­mune), pub­lished in the jour­nal Tracés, in which he analy­ses the social and polit­i­cal dynam­ics linked to data cen­tres in Plaine Commune.

In France, the annu­al elec­tric­i­ty con­sump­tion of data cen­tres was esti­mat­ed at 10 TWh in 2023. Accord­ing to pro­jec­tions by the French Elec­tric­i­ty Trans­mis­sion Net­work (RTE), this con­sump­tion could reach 28 TWh per year by 2035. If all the projects announced go ahead and data cen­tres use their full capac­i­ty, con­sump­tion could even rise to 80 TWh per year, or around 15% of France’s cur­rent nuclear pro­duc­tion⁸.

“The cloud relies on data cen­tres,” points out urban plan­ner Cécile Diguet, “these build­ings that need to be con­struct­ed, con­sume a lot of elec­tric­i­ty and are nec­es­sary to oper­ate and cool the servers. This cool­ing is essen­tial because servers gen­er­ate a lot of heat.” She adds: “The vocab­u­lary of dig­i­tal tech­nol­o­gy – “imma­te­r­i­al”, “wire­less”, “dema­te­ri­alised” – delib­er­ate­ly obscures the mate­r­i­al real­i­ty and impact of dig­i­tal tech­nol­o­gy.” Soci­ol­o­gy researcher Clé­ment Mar­quet adds: “There is noth­ing imma­te­r­i­al about the cloud: all data exchanges rely on phys­i­cal infra­struc­ture.” Fur­ther­more, “in the indus­try, it is com­mon to say that the cloud is some­one else’s data centre.”

The mate­r­i­al dimen­sion goes beyond build­ings alone. Clé­ment Mar­quet points out that “the man­u­fac­ture of com­put­ers requires the extrac­tion of min­er­als, with destruc­tive effects on ecosys­tems, soil, water and air.” He points out that “chip engrav­ing is very water-inten­sive” and “elec­tron­ic waste caus­es sig­nif­i­cant pol­lu­tion.” These effects, although indi­rect, are struc­tural­ly linked to the func­tion­ing of the cloud.

Ter­mi­nals, includ­ing com­put­ers, tele­vi­sions, smart­phones, con­nect­ed devices, remain the largest source of dig­i­tal ener­gy con­sump­tion, account­ing for more than 50% of glob­al demand. Net­works and data cen­tres share the rest: approx­i­mate­ly one-third for the for­mer and just under 15% for the lat­ter⁹. In 2022, data cen­tres con­sumed between 240 and 340 TWh, or 1 to 1.3% of glob­al elec­tric­i­ty demand¹⁰.

This dis­tri­b­u­tion is chang­ing rapid­ly under the influ­ence of gen­er­a­tive AI, cloud com­put­ing and stream­ing, which are sig­nif­i­cant­ly increas­ing the con­sump­tion of dig­i­tal fac­to­ries¹¹. Clé­ment Mar­quet dis­tin­guish­es between dis­trib­uted and cen­tralised infra­struc­ture: “Ter­mi­nals con­sume more, but in a dis­persed man­ner. Data cen­tres con­cen­trate their demand, which cre­ates spe­cif­ic ten­sions on net­works.” This pool­ing caus­es local pres­sures that are invis­i­ble in over­all bal­ances. Cécile Diguet points out that “dig­i­tal tech­nol­o­gy relies on three major infra­struc­tures: net­works (around one-third of the car­bon impact), data cen­tres (around 40%) and equip­ment. They also con­tain hard­ware and net­works, mak­ing them a cen­tral and vis­i­ble infra­struc­ture, unlike net­works, which are often buried underground.”

“Data cen­tres there­fore cause localised envi­ron­men­tal prob­lems that can­not be reduced to a glob­al com­par­i­son between ter­mi­nals, net­works and data cen­tres,” notes Clé­ment Mar­quet. “The rapid rise of gen­er­a­tive AI is exac­er­bat­ing this pres­sure: these infra­struc­tures, which are very ener­gy-inten­sive but less depen­dent on the net­work, tend to be scat­tered across rur­al areas. Their size and inten­si­ty nev­er­the­less pose many challenges.”

The loca­tion of data cen­tres is pri­mar­i­ly based on eco­nom­ic cri­te­ria: tax­a­tion, ener­gy costs, avail­able land, polit­i­cal sta­bil­i­ty and legal cer­tain­ty. Envi­ron­men­tal argu­ments, often high­light­ed in com­mu­ni­ca­tion strate­gies, are not deci­sive at Euro­pean or glob­al lev­el. Clé­ment Mar­quet points out that “attrac­tive­ness poli­cies play an impor­tant role,” cit­ing North­ern Vir­ginia as an exam­ple. Dublin and Ams­ter­dam have ben­e­fit­ed from favourable tax­a­tion, and Spain is posi­tion­ing itself thanks to active sup­port from the pub­lic authorities.

In France, cer­tain mea­sures are enhanc­ing attrac­tive­ness: “For local author­i­ties, the prop­er­ty tax is what makes these infra­struc­tures so attrac­tive,” explains Cécile Diguet. “For inter-munic­i­pal com­mu­ni­ties, the ben­e­fits are lim­it­ed. It is main­ly the State that inter­venes through reduc­tions in elec­tric­i­ty tax­es, in par­tic­u­lar the TICFE, which is an impor­tant lever for operators.”

How­ev­er, this attrac­tive­ness does not guar­an­tee coher­ent local inte­gra­tion. Clé­ment Mar­quet notes that “these poli­cies often neglect land use plan­ning: they take lit­tle account of inte­gra­tion or local con­di­tions.” They may also be accom­pa­nied by spe­cif­ic sta­tus­es (crit­i­cal archi­tec­ture, nation­al inter­est), inhibit­ing envi­ron­men­tal require­ments. Ire­land is a case in point: its favourable tax regime has attract­ed many cen­tres, but their con­cen­tra­tion has sat­u­rat­ed the net­works, lead­ing to a mora­to­ri­um in Dublin in 2021¹². In the Nether­lands, some projects have been sus­pend­ed due to their exces­sive impact on land and energy.

Data cen­tres host essen­tial ser­vices in major sec­tors such as health­care, defence, gov­ern­ment and finance. Their secu­ri­ty has become a mat­ter of dig­i­tal sov­er­eign­ty and nation­al secu­ri­ty, with their pro­tec­tion con­sid­ered a pri­or­i­ty for eco­nom­ic and social sta­bil­i­ty, “espe­cial­ly when they host sen­si­tive data (health, gov­ern­ment depart­ments, mil­i­tary),” says Cécile Diguet. “Some are clas­si­fied as OVIs. The mil­i­tary often uses its own data cen­tres and net­works. Data cen­tres are essen­tial to guar­an­tee dig­i­tal sov­er­eign­ty, and the state can cre­ate them to secure strate­gic data.”

The des­ig­na­tion of an oper­a­tor of vital impor­tance (OVI) is intend­ed to pro­tect these crit­i­cal facil­i­ties. In Europe, gov­ern­ments are seek­ing to secure strate­gic data in a con­text of grow­ing threats, par­tic­u­lar­ly with the rise of gen­er­a­tive AI. But Clé­ment Mar­quet points out that “this sta­tus is often high­light­ed, par­tic­u­lar­ly in the con­text of com­pe­ti­tion around AI,” adding that “this has prompt­ed coun­tries such as the Unit­ed King­dom, and soon France via Arti­cle 15 of the Eco­nom­ic Sim­pli­fi­ca­tion Act, to recog­nise data cen­tres as struc­tures of nation­al impor­tance.” This recog­ni­tion has con­crete effects: “On the pow­er of local elect­ed offi­cials, land arti­fi­cial­i­sa­tion and envi­ron­men­tal pro­ce­dures.” This is a dif­fer­ent log­ic from that of OVIs, which are ‘linked to the pro­tec­tion of state data and more mod­est projects.’

The Euro­pean ener­gy mix, such as nuclear pow­er in France and hydro­elec­tric­i­ty in Swe­den, may seem con­ducive to reduc­ing car­bon foot­prints. But this does not guar­an­tee a low­er over­all impact. Oth­er fac­tors come into play, such as facil­i­ty design, ther­mal man­age­ment, use of mate­ri­als and rapid growth in dig­i­tal usage.

Cécile Diguet high­lights the lim­i­ta­tions of this rea­son­ing: “There are major dif­fer­ences between coun­tries: France high­lights its use of nuclear pow­er, but this also has its own impacts, par­tic­u­lar­ly on waste. Gains in ener­gy effi­cien­cy are quick­ly off­set by increased usage. Greater effi­cien­cy does not nec­es­sar­i­ly mean low­er con­sump­tion. At the local lev­el, data cen­tres can com­pete with oth­er elec­tric­i­ty needs, such as trans­port or indus­try. The impact is there­fore also felt at the region­al level.”

Ener­gy opti­mi­sa­tion cer­tain­ly leads to sav­ings, but these are often off­set by grow­ing demand linked to tech­nolo­gies such as gen­er­a­tive AI. This rebound effect part­ly can­cels out the envi­ron­men­tal ben­e­fits. Resource use also remains an issue, regard­less of the ener­gy source¹³.

Clé­ment Mar­quet shares this view: “The ener­gysource mix is cer­tain­ly an impor­tant fac­tor, but it varies from coun­try to coun­try. It is not enough to guar­an­tee a low impact. The envi­ron­men­tal impact of dig­i­tal tech­nol­o­gy is not lim­it­ed to CO₂ emis­sions: it also includes resource con­sump­tion, water, waste, etc.” Fur­ther­more, “the grow­ing needs of dig­i­tal tech­nol­o­gy, par­tic­u­lar­ly in rela­tion to gen­er­a­tive AI, are adding to oth­er low-car­bon elec­tric­i­ty needs (mobil­i­ty, indus­try, heat­ing). This com­pe­ti­tion may slow down the ener­gy tran­si­tion and pre­vent the decar­bon­i­sa­tion of oth­er sec­tors at the local level.’

Aicha Fall