L'intelligence artificielle transforme l'\u00e9lectricit\u00e9 en contrainte majeure pour l'informatique. La demande en \u00e9nergie des centres de donn\u00e9es devrait doubler d'ici 2030, et les r\u00e9seaux \u00e9lectriques ne pourront pas absorber cette charge suffisamment rapidement. Les turbines \u00e0 gaz install\u00e9es \u00e0 proximit\u00e9 des centres de donn\u00e9es sont devenues l'un des rares moyens d'accro\u00eetre rapidement la production d'\u00e9lectricit\u00e9, ce qui explique les commandes record enregistr\u00e9es par des \u00e9quipementiers comme Baker Hughes. Voici pourquoi cette demande est r\u00e9elle, pourquoi le gaz en capte une part et o\u00f9 se situe d\u00e9sormais le goulot d'\u00e9tranglement.<\/p>\n
Les centres de donn\u00e9es, pilot\u00e9s par l'intelligence artificielle, font cro\u00eetre la demande en \u00e9lectricit\u00e9 plus rapidement que les r\u00e9seaux \u00e9lectriques ne peuvent la satisfaire. L'Agence internationale de l'\u00e9nergie pr\u00e9voit que la consommation mondiale d'\u00e9lectricit\u00e9 des centres de donn\u00e9es doublera approximativement d'ici 2030 pour atteindre environ 945 t\u00e9rawattheures, et les listes d'attente pour le raccordement au r\u00e9seau s'\u00e9tendent sur plusieurs ann\u00e9es. Les turbines \u00e0 gaz peuvent \u00eatre install\u00e9es \u00e0 proximit\u00e9 des centres de donn\u00e9es pour fournir une alimentation \u00e9lectrique stable et continue, et ce, plus rapidement qu'un raccordement au r\u00e9seau. C'est pourquoi les op\u00e9rateurs et les d\u00e9veloppeurs de centres de donn\u00e9es hyperscale en commandent en grande quantit\u00e9. Baker Hughes a, \u00e0 lui seul, r\u00e9serv\u00e9 2,7 gigawatts de turbines \u00e0 gaz pour ses centres de donn\u00e9es et l'alimentation mobile au cours d'un seul trimestre en 2026.<\/p>\n
L'ampleur de la demande fait de cette question un enjeu strat\u00e9gique plut\u00f4t qu'un sujet de niche. L'Agence internationale de l'\u00e9nergie, dans son analyse \u00c9nergie et IA 2025, pr\u00e9voit que la consommation mondiale d'\u00e9lectricit\u00e9 des centres de donn\u00e9es doublera approximativement pour atteindre environ 945 t\u00e9rawattheures d'ici 2030, soit un peu moins de 3 % de la consommation mondiale d'\u00e9lectricit\u00e9, et augmentera pour atteindre pr\u00e8s de 1\u00a0200 t\u00e9rawattheures d'ici 2035 (AIE, \u00e9nergie et IA<\/a>La consommation d'\u00e9lectricit\u00e9 des centres de donn\u00e9es d\u00e9di\u00e9s \u00e0 l'IA est celle qui conna\u00eet la croissance la plus rapide, triplant approximativement sur la p\u00e9riode.<\/p>\n Cette charge est non seulement importante, mais aussi concentr\u00e9e et constante. L'entra\u00eenement et l'inf\u00e9rence de l'IA fonctionnent 24 h\/24 et 7 j\/7 \u00e0 un taux d'utilisation \u00e9lev\u00e9, ce qui diff\u00e8re consid\u00e9rablement de la demande commerciale classique. Y r\u00e9pondre exige une alimentation \u00e9lectrique fiable, disponible en permanence, et non pas seulement lorsque le vent souffle ou que le soleil brille.<\/p>\n La solution \u00e9vidente, connecter le centre de donn\u00e9es au r\u00e9seau \u00e9lectrique, se heurte \u00e0 une contrainte majeure\u00a0: le temps. Sur de nombreux march\u00e9s, la liste d\u2019attente pour obtenir un nouveau raccordement au r\u00e9seau s\u2019\u00e9tend sur plusieurs ann\u00e9es, et la construction de nouvelles lignes de transport prend encore plus de temps. Pour un op\u00e9rateur hyperscale qui s\u2019efforce de d\u00e9ployer des capacit\u00e9s d\u2019IA, attendre plusieurs ann\u00e9es pour l\u2019alimentation \u00e9lectrique signifie attendre plusieurs ann\u00e9es avant de percevoir des revenus.<\/p>\n Ce d\u00e9calage temporel incite les d\u00e9veloppeurs \u00e0 privil\u00e9gier la production d'\u00e9lectricit\u00e9 sur site ou \u00e0 proximit\u00e9 imm\u00e9diate. Construire sa propre alimentation \u00e9lectrique pr\u00e8s du b\u00e2timent, en aval du compteur, peut s'av\u00e9rer plus rapide que d'attendre dans la file d'attente pour l'interconnexion. La rapidit\u00e9 d'acc\u00e8s \u00e0 l'\u00e9nergie, et non plus seulement son co\u00fbt, est d\u00e9sormais le facteur d\u00e9terminant pour l'emplacement et la vitesse de construction d'un centre de donn\u00e9es.<\/p>\n Les turbines \u00e0 gaz r\u00e9pondent aux besoins des centres de donn\u00e9es selon trois axes. Elles sont pilotables, c'est-\u00e0-dire qu'elles fonctionnent \u00e0 la demande et fournissent la puissance stable et constante n\u00e9cessaire au calcul de l'IA. Elles peuvent \u00eatre install\u00e9es \u00e0 proximit\u00e9 des points de consommation, \u00e9vitant ainsi les goulots d'\u00e9tranglement du r\u00e9seau de transport. Et, \u00e0 grande \u00e9chelle, elles peuvent \u00eatre d\u00e9ploy\u00e9es plus rapidement que la plupart des autres solutions offrant la m\u00eame capacit\u00e9 stable. L'AIE pr\u00e9voit une augmentation de la production de gaz naturel d'environ 175 t\u00e9rawattheures pour contribuer \u00e0 satisfaire la demande des centres de donn\u00e9es, notamment aux \u00c9tats-Unis.AIE, Demande \u00e9nerg\u00e9tique li\u00e9e \u00e0 l'IA<\/a>).<\/p>\n Il ne s'agit pas d'une opposition entre le gaz et les \u00e9nergies renouvelables. Ces derni\u00e8res devraient r\u00e9pondre \u00e0 la plus grande part de la demande de nouveaux centres de donn\u00e9es, gr\u00e2ce \u00e0 des d\u00e9lais de construction courts et \u00e0 des co\u00fbts r\u00e9duits. Cependant, le d\u00e9veloppement des \u00e9nergies renouvelables doit \u00eatre consolid\u00e9, et les turbines \u00e0 gaz constituent l'une des technologies qui jouent ce r\u00f4le aujourd'hui. Il en r\u00e9sulte une forte augmentation des commandes de turbines. Baker Hughes a enregistr\u00e9 2,7 gigawatts de turbines \u00e0 gaz pour les centres de donn\u00e9es et l'alimentation mobile au cours d'un seul trimestre 2026, sur un total record de 7,1 milliards de dollars de commandes dans le secteur des technologies industrielles et \u00e9nerg\u00e9tiques.Rigzone<\/a>).<\/p>\n La demande s'\u00e9tant d\u00e9plac\u00e9e vers les turbines, la contrainte a suivi cette m\u00eame tendance. L'AIE constate que les d\u00e9lais de livraison des nouvelles turbines \u00e0 gaz s'\u00e9tendent d\u00e9sormais sur plusieurs ann\u00e9es, une p\u00e9nurie d'approvisionnement qui pourrait retarder la mise en service de certaines centrales \u00e0 gaz au-del\u00e0 de 2030. Autrement dit, commander une turbine aujourd'hui ne garantit pas la production d'\u00e9lectricit\u00e9 demain.<\/p>\n Pour les fabricants d'\u00e9quipements, cette raret\u00e9 repr\u00e9sente une opportunit\u00e9. C'est pourquoi Baker Hughes, GE Vernova, Siemens Energy et Mitsubishi Power augmentent leurs capacit\u00e9s de production et que leurs carnets de commandes s'\u00e9tendent d\u00e9sormais sur plusieurs ann\u00e9es. Baker Hughes vise un chiffre d'affaires annuel d'environ 5 milliards de dollars pour ses syst\u00e8mes d'\u00e9nergie d'ici 2029 et pr\u00e9voit de doubler sa capacit\u00e9 de production de turbines \u00e0 gaz par rapport \u00e0 2026 d'ici fin 2028. Celui qui sera capable de produire des turbines le plus rapidement captera la demande.<\/p>\n Pour les acteurs de ce march\u00e9, la le\u00e7on \u00e0 retenir est que la ressource rare est l'\u00e9lectricit\u00e9 fiable et rapidement fournie, et non l'\u00e9nergie en soi. Les acheteurs sont pr\u00eats \u00e0 payer un prix plus \u00e9lev\u00e9 pour une capacit\u00e9 qui r\u00e9duit leur d\u00e9lai de mise en service et \u00e0 s\u00e9curiser leur approvisionnement des ann\u00e9es \u00e0 l'avance. Cela modifie profond\u00e9ment la mani\u00e8re dont les \u00e9quipements, la fourniture de gaz et les services \u00e9nerg\u00e9tiques doivent \u00eatre positionn\u00e9s et tarif\u00e9s.<\/p>\n Cela explique aussi pourquoi une entreprise de services p\u00e9troliers comme Baker Hughes parle d\u00e9sormais d'\u00e9conomie num\u00e9rique. Pour comprendre comment une grande entreprise \u00e9nerg\u00e9tique se repositionne pr\u00e9cis\u00e9ment pour r\u00e9pondre \u00e0 cette demande, consultez notre analyse. Baker Hughes et le pivot de puissance<\/a>, et comment Chevron renforce la puissance des centres de donn\u00e9es avec le projet Kilby.<\/a>.<\/p>","protected":false},"excerpt":{"rendered":" L'intelligence artificielle (IA) fait doubler la demande en \u00e9lectricit\u00e9 des centres de donn\u00e9es d'ici 2030, tandis que les files d'attente sur les r\u00e9seaux \u00e9lectriques s'allongent pendant des ann\u00e9es. Les turbines \u00e0 gaz permettent de fournir rapidement une puissance stable, ce qui explique les commandes record enregistr\u00e9es par Baker Hughes et ses concurrents. Voici pourquoi.<\/p>","protected":false},"author":12,"featured_media":3989,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"p54_article_data":"{\"meta\":{\"kicker\":\"Answer \u00b7 Data Center Power\",\"topics\":[\"Energy\",\"Analysis\"],\"title\":\"Why Are Data Centers Driving Demand for Gas Turbines?\",\"dek\":\"Artificial intelligence is turning electricity into the binding constraint on computing. Data center power demand is set to roughly double by 2030, and grids cannot connect that load fast enough. Gas turbines sited next to the data center have become one of the few ways to add firm power quickly, which is why equipment makers like Baker Hughes are booking record orders. Here is why the demand is real, why gas is winning a share of it, and where the bottleneck now sits.\",\"date\":\"30\/07\/2026\",\"readTime\":\"7 min read\",\"author\":\"Project 54 Research & Strategy\"},\"quickAnswer\":{\"q\":\"Why are data centers driving demand for gas turbines?\",\"a\":\"Data centers, driven by artificial intelligence, are adding electricity demand faster than grids can supply it. The International Energy Agency projects global data center electricity use will roughly double to around 945 terawatt hours by 2030, and connection queues for new grid capacity stretch for years. Gas turbines can be installed next to a data center to deliver firm, always on power on a shorter timeline than a grid connection, so hyperscale operators and developers are ordering them in volume. Baker Hughes alone booked 2.7 gigawatts of gas turbines for data centers and mobile power in a single quarter in 2026.\"},\"takeaways\":[\"The International Energy Agency projects data center electricity demand will roughly double to about 945 terawatt hours by 2030, with AI focused data centers growing fastest.\",\"Grid connection queues run for years, so on site or near site generation is often the fastest way to power a new data center.\",\"Natural gas is expected to supply a large share of the new load, expanding by around 175 terawatt hours to meet data center demand, notably in the United States.\",\"Gas turbines deliver firm, dispatchable power that matches the constant, high load profile of AI computing, unlike weather dependent renewables on their own.\",\"The bottleneck has shifted to the turbines themselves: lead times now stretch several years, which is why equipment makers are booking record order books and expanding capacity.\"],\"sections\":[{\"id\":\"demand\",\"q\":\"How much new electricity do data centers actually need?\",\"h\":\"A demand curve that doubles\",\"p\":[\"The scale of the demand is what makes this a strategic issue rather than a niche one. The International Energy Agency, in its 2025 Energy and AI analysis, projects that global electricity consumption from data centers will roughly double to around 945 terawatt hours by 2030, just under 3 percent of world electricity, and rise toward 1,200 terawatt hours by 2035 (IEA, Energy and AI<\/a>). Electricity use from AI focused data centers grows fastest of all, roughly tripling over the period.\",\"That load is not just large, it is concentrated and constant. AI training and inference run around the clock at high utilisation, which is a very different profile from ordinary commercial demand. Meeting it requires firm power that is available every hour, not just when the wind blows or the sun shines.\"]},{\"id\":\"grid\",\"q\":\"Why not just connect to the grid?\",\"h\":\"The queue is the problem\",\"p\":[\"The obvious answer, connect the data center to the grid, runs into a hard constraint: time. In many markets the queue to secure a large new grid connection stretches for several years, and new transmission takes even longer to build. For a hyperscale operator racing to deploy AI capacity, a multi year wait for power is a multi year wait for revenue.\",\"That timing gap is what pushes developers toward on site or near site generation. Building your own power next to the building, behind the meter, can be faster than waiting in the interconnection queue. Speed to power, not just cost of power, is now the deciding factor for where and how fast a data center can be built.\"]},{\"id\":\"gas\",\"q\":\"Why gas turbines specifically?\",\"h\":\"Firm power on a shorter timeline\",\"p\":[\"Gas turbines fit the data center problem on three dimensions. They are dispatchable, meaning they run on demand and provide the firm, constant power AI computing needs. They can be sited next to the load, avoiding the transmission bottleneck. And at scale they can be deployed faster than most alternatives that deliver the same firm capacity. The IEA expects natural gas to expand by around 175 terawatt hours to help meet data center demand, notably in the United States (IEA, Energy demand from AI<\/a>).\",\"This is not a story of gas versus renewables. Renewables are expected to supply the largest single share of new data center demand, helped by short build times and low cost. But renewables need firming, and gas turbines are one of the technologies filling that role today. The result is a surge in turbine orders. Baker Hughes booked 2.7 gigawatts of gas turbines for data centers and mobile power in a single quarter of 2026, part of a record 7.1 billion dollars of Industrial and Energy Technology orders (Rigzone<\/a>).\"]},{\"id\":\"bottleneck\",\"q\":\"Where is the bottleneck now?\",\"h\":\"The turbines themselves\",\"p\":[\"As demand has shifted to turbines, the constraint has moved with it. The IEA notes that lead times for new gas turbines now stretch several years, a supply crunch that could delay commissioning of some gas fired capacity beyond 2030. In other words, ordering a turbine today does not guarantee power tomorrow.\",\"For equipment makers, that scarcity is the opportunity. It is why Baker Hughes, GE Vernova, Siemens Energy and Mitsubishi Power are expanding manufacturing capacity and why their order books now reach years into the future. Baker Hughes is targeting roughly 5 billion dollars of annual power systems revenue by 2029 and doubling its gas turbine capacity from 2026 levels by the end of 2028. Whoever can build turbines fastest captures the demand.\"]},{\"id\":\"means\",\"q\":\"What does this mean for energy suppliers and sellers?\",\"h\":\"Sell speed to power\",\"p\":[\"For anyone selling into this market, the lesson is that the scarce resource is firm power delivered quickly, not energy in the abstract. Buyers will pay a premium for capacity that shortens their time to power, and they will lock in supply years ahead to secure it. That reshapes how equipment, gas supply and power services should be positioned and priced.\",\"It also explains why an oilfield services name like Baker Hughes now talks about the digital economy. To understand how a major energy company is repositioning around exactly this demand, read our analysis of Baker Hughes and the power pivot<\/a>, and how Chevron is building power for data centers with Project Kilby<\/a>.\"]}],\"media\":{\"image\":{\"src\":\"https:\/\/projectfifty4.com\/wp-content\/uploads\/2026\/07\/data-center-power-demand-operations.jpg\",\"label\":\"Engineers reviewing live data on an operations wall, the kind of always on computing load that is driving data center power demand.\",\"credit\":\"Project 54\"},\"infographicLabel\":\"FIG.54 \/ DATA CENTER POWER \/ GAS TURBINE DEMAND \/ REV 1.0\"},\"poll\":{\"q\":\"What will be the biggest constraint on data center growth to 2030?\",\"options\":[{\"id\":\"power\",\"label\":\"Access to firm power\",\"insight\":\"The IEA sees data center demand doubling by 2030 while grid queues stretch for years. Firm power, not chips, is emerging as the binding constraint.\"},{\"id\":\"turbines\",\"label\":\"Equipment lead times\",\"insight\":\"Gas turbine lead times now run several years. The bottleneck has moved from the grid to the factory floor that builds the turbines.\"},{\"id\":\"grid\",\"label\":\"Grid and transmission\",\"insight\":\"New transmission takes many years to permit and build, which is exactly why developers are turning to behind the meter generation.\"},{\"id\":\"cost\",\"label\":\"Cost of energy\",\"insight\":\"Cost matters, but for AI operators racing to deploy, speed to power is often worth more than the cheapest electron.\"}],\"note\":\"The common thread is time. Every constraint on the list is really a constraint on how fast firm power can be delivered.\"},\"faq\":[{\"q\":\"Why are data centers driving demand for gas turbines?\",\"a\":\"AI is pushing data center electricity demand up faster than grids can supply it, with the IEA projecting a roughly doubling to around 945 terawatt hours by 2030. Grid connection queues stretch for years, so operators use gas turbines sited next to the data center to add firm power on a shorter timeline. Baker Hughes booked 2.7 gigawatts of such turbines in a single quarter of 2026.\"},{\"q\":\"How much electricity will data centers use by 2030?\",\"a\":\"The International Energy Agency projects that global data center electricity consumption will roughly double to around 945 terawatt hours by 2030, just under 3 percent of world electricity, and rise toward 1,200 terawatt hours by 2035. AI focused data centers grow fastest, tripling over the period.\"},{\"q\":\"Why not use renewables to power data centers?\",\"a\":\"Renewables are expected to supply the largest single share of new data center demand because they are cheap and quick to build. But AI computing needs firm, around the clock power, and renewables on their own are weather dependent. Gas turbines provide the firming capacity that keeps the load supplied at all hours, so the two are used together rather than as alternatives.\"},{\"q\":\"How long does it take to get a gas turbine for a data center?\",\"a\":\"Lead times have lengthened significantly. The IEA notes that turbine deliveries for new gas fired plants now face lead times of several years, which could delay some commissioning beyond 2030. That scarcity is why equipment makers are expanding manufacturing capacity and booking multi year order backlogs.\"},{\"q\":\"Which companies make gas turbines for data centers?\",\"a\":\"The main suppliers include Baker Hughes, GE Vernova, Siemens Energy and Mitsubishi Power. Baker Hughes booked a record 7.1 billion dollars of Industrial and Energy Technology orders in the second quarter of 2026, including 2.7 gigawatts of gas turbines for data centers and mobile power, and is targeting roughly 5 billion dollars of annual power systems revenue by 2029.\"}],\"newsletter\":{\"eyebrow\":\"The P54 Energy Growth Brief\",\"title\":\"Revenue architecture, engineered, for energy B2B\",\"body\":\"Join energy leaders getting our weekly analysis of how the industry's biggest players build growth, and what it means for how you sell and market. No noise, just the strategy behind the moves.\",\"cta\":\"Subscribe\"},\"related\":[{\"title\":\"Baker Hughes and the Power Pivot\",\"topic\":\"Industry Leader\",\"href\":\"\/insights\/baker-hughes-data-center-power-strategy-2026\/\"},{\"title\":\"Chevron Starts Selling Power, Not Barrels: Inside Project Kilby\",\"topic\":\"Industry Leader\",\"href\":\"\/insights\/chevron-microsoft-project-kilby\/\"},{\"title\":\"SLB's Digital Reinvention\",\"topic\":\"Industry Leader\",\"href\":\"\/insights\/slb-schlumberger-digital-oilfield-services-strategy-2026\/\"},{\"title\":\"ADNOC and XRG: A Bet on Gas, Chemicals and AI Power\",\"topic\":\"Industry Leader\",\"href\":\"\/insights\/adnoc-xrg-gas-chemicals-ai-bet\/\"}],\"listenTime\":\"Listen: 12 min\"}","p54_faq":"","p54_media":"","p54_comments_enabled":"","footnotes":""},"categories":[92,125],"tags":[],"class_list":["post-3990","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-analysis","category-strategy"],"acf":[],"_links":{"self":[{"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/posts\/3990","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/users\/12"}],"replies":[{"embeddable":true,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/comments?post=3990"}],"version-history":[{"count":1,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/posts\/3990\/revisions"}],"predecessor-version":[{"id":3991,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/posts\/3990\/revisions\/3991"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/media\/3989"}],"wp:attachment":[{"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/media?parent=3990"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/categories?post=3990"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/projectfifty4.com\/fr\/wp-json\/wp\/v2\/tags?post=3990"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Le probl\u00e8me, c'est la file d'attente.<\/h2>\n
Une puissance ferme dans un d\u00e9lai plus court<\/h2>\n
Les turbines elles-m\u00eames<\/h2>\n
Vendre la vitesse \u00e0 la puissance<\/h2>\n