The accelerated expansion of Artificial Intelligence is changing a less visible part of the cloud business: it’s no longer enough to have land, fiber, servers, or access to GPUs. Increasingly, a much more basic yet harder-to-secure factor in terms of timing is becoming critical: reliable, available electricity. In Europe, where the debate on digital sovereignty and computing capacity has gained momentum, the energy bottleneck is emerging as one of the key factors that could significantly influence the real growth rate of new data centers.
The clearest signal is coming from the United States. Oracle and Bloom Energy expanded their partnership in April 2026 to deploy up to 2.8 GW of capacity through fuel cells, with an initial 1.2 GW already contracted for Oracle projects within the US. This move is significant: it reflects that major operators are no longer only competing for chips and industrial land but also for securing power supply within timeframes compatible with AI infrastructure deployment.
Europe Looks to the Cloud, but the Power Grid Sets the Schedule
The International Energy Agency has warned that the electricity demand from data centers surged by 17% in 2025, far outpacing the global electricity consumption growth of 3%. At the same time, the organization emphasizes that AI-focused data centers are growing even faster, and total data center consumption could double by 2030. It’s not just about more servers being turned on: training and inference of models require dense clusters, more demanding cooling, and much more stable energy availability.
This is where Europe faces its biggest challenge. According to data gathered by Reuters from an Ember report, connecting a new data center to the grid in traditional hubs like Frankfurt, London, Amsterdam, Paris, or Dublin can take an average of seven to ten years, with some projects experiencing delays of up to thirteen years. The conclusion is uncomfortable for any European AI strategy: the problem is no longer just attracting investment; it’s ensuring that investment can access reliable power within a practical timeframe.
The IEA succinctly summarizes this temporal asymmetry: while a data center can be established in one to three years, planning, permitting, and building new network infrastructure can take between five and fifteen years. This lag makes electricity as much a location factor as connectivity or land availability. In practice, the grid is beginning to influence which markets will be able to absorb the next wave of cloud expansion and which will lag behind.
For this reason, the European debate is no longer solely focused on energy efficiency or sustainability but also on industrial competitiveness. Reuters reports that without improved grid planning, up to half of Europe’s data center capacity could be relocated outside the major traditional hubs by 2035. France appears somewhat less pressured due to a less strained network, while markets with shorter connection times, like Italy, are becoming more attractive relative to traditional nodes.
U.S. Accelerates with On-Site Generation and Asia Multiplies Capacity
The US is responding with a combination of grid expansion and local generation. In Texas, one of the emerging data center hubs, Reuters highlights that ERCOT has launched a $33 billion plan to strengthen the grid, though many new lines won’t be operational until after 2030. Meanwhile, the market is moving toward “behind the meter” solutions—generation facilities located very close to consumption points, supported by gas, renewables, and other firm sources.
The partnership between Oracle and Bloom fits precisely within this logic. Fuel cells don’t solely solve the sector’s energy challenge, but they do offer a way to reduce dependence on especially long connection timelines. Bloom also claims that its SOFC systems allow for modular deployment and, with heat recovery, can significantly improve overall efficiency—though the fuel mix and final costs remain critical variables in each project.
Meanwhile, Asia is growing rapidly but is not immune to energy challenges. Cushman & Wakefield estimates that the development pipeline of data centers in Asia-Pacific stands at 19.4 GW by the end of 2025, with 3.7 GW under construction and 15.7 GW planned. The report highlights that Southeast Asia accounts for 31% of capacity under development, and markets like Johor, Mumbai, or Bangkok have become priority nodes for cloud and AI loads. In essence, Asia continues to scale up, but it’s doing so by selecting locations capable of providing higher capacity and better access to available power.
Illustrative examples include Microsoft’s 2024 announcement of a $2.9 billion investment to expand its cloud and AI infrastructure in Japan. Singapore has reopened capacity through a new call for at least 200 MW, demanding high efficiency standards and a minimum of 50% from green sources. Malaysia exemplifies the regional boom’s other side: Reuters reported that by 2025, demand from data centers could require 19.5 GW of new generation capacity by 2035—equal to 52% of Peninsular Malaysia’s electrical consumption.
Energy Goes from Cost to Competitive Advantage
From a European perspective, the lesson is clear. AI cloud deployment will no longer be decided solely in the software realm, nor only in hardware. It will also depend on each region’s ability to supply megawatts where and when they are needed, at an affordable cost. Electric availability is starting to serve as a new market entry filter.
This shifts the strategic conversation for operators, hyperscalers, and governments. Access to reliable energy, grid improvements, long-term power purchase agreements, local generation, and renewable integration are no longer peripheral to cloud business—they are central to the industrial architecture of AI. Europe still has time to compete, but it will need to address its power grid with the same urgency as foundational models, sovereign data centers, or AI gigafactories. Because in this new phase, compute power starts well before the rack: it begins at the substation.