_{According to the latest data from International Energy Agency (IEA), data centers consumed 415 terawatt-hours of electricity in 2024, equivalent to around 1.5% of global electricity consumption. This figure is expected to double by 2030 and surpass Japan’s total electricity consumption. In the United States, data centers alone will account for nearly half of the country’s growth in electricity demand through the end of this decade. The country will allocate more electricity to data centers than to all energy-intensive industries combined.}

_{Artificial intelligence is commonly perceived as software, algorithms, and cloud services. Yet scaling it is creating renewed dependence on the most traditional infrastructure of the industrial age, including power plants, high-voltage transmission lines, substations, transformers, cooling systems, land, and water. The most advanced digital technology of our era is once again making the oldest forms of physical infrastructure strategically important. What is at stake is technological capacity, sovereignty and geopolitical power.}

_{The digital economy was never truly immaterial. AI is not suddenly making it physical. Rather, it is making its material foundations far more visible, far more intensive and far more strategically decisive. These foundations should be understood as a layered stack consisting of energy sources, power generation, transmission grids, data centers, chips and computing capacity, models, and finally applications. Capabilities at the upper layers are constrained by the capacity and vulnerabilities of the layers beneath them.}

_{Constraint not ideas, but queue}

_{The conventional account of the race centers on chips, models, data, capital, and talent. Yet one input is consistently omitted from this list: electricity. Although the revolution is digital by nature, its binding constraint is physical.}

_{To understand what this means, it is enough to look at how a large data center is built. A company can buy the land, construct the building, and order billions of dollars’ worth of processors. But the facility cannot perform a single computation until it is connected to the power grid. Authorities assess whether a massive new load could disrupt the balance of the grid, plan the necessary investments in transmission lines and transformers, and obtain the required permits. Everyone seeking a connection must join a queue, and in some regions that queue barely moves for years. This is why the decisive metric in the sector is no longer the price of electricity, but how quickly a site can gain access to power.}

_{The second challenge is that announced capacity is not the same as actual capacity. According to analyses of grid connection queues in the U.S., more than 1,000 gigawatts of new power plant projects are awaiting connection before 2030. However, some of these projects are never completed, some of those that are completed merely replace retiring power plants, and because wind and solar do not generate the same amount of power every hour, installed capacity is not the same as reliable capacity. Once these factors are taken into account, only about 33 gigawatts of reliable capacity from the one-thousand-gigawatt pool can be added to the grid by 2030. When generation built by facilities on their own sites is included, the total rises to 82 gigawatts. There is abundance on paper, but scarcity on the grid. Therefore, we can say that the constraint is not ideas, but the queue. An idea for a new model can emerge within days; however, the power line needed to run that model takes years.}

_{Four models}

_{This constraint is reorganizing the geography of the race around four strategic models.}

_{The U.S. has a clear advantage in frontier models, advanced chips}

_{and capital markets. Yet in some regions, waiting times for grid connections can reach seven years, orders for transformers and gas turbines stretch over several years, and permitting processes remain lengthy. Due to streamlined permitting and faster grid connections, an AI data center in China can be completed six to sixteen months earlier than one in the U.S.}

_{China, meanwhile, is seeking to turn infrastructure mobilization into a technological advantage. The U.S. accounts for 45% of global data center capacity, while China accounts for 25%. Beijing has identified computing infrastructure as a strategic priority in its new five-year plan and is pursuing a spatial strategy that matches demand in the east with renewable energy generation in the west. While the U.S. is largely asking where it can find enough electricity for AI, China is increasingly asking how computing capacity can absorb electricity that the grid is unable to accommodate. This does not mean that the problem has been solved. In the first half of 2026, approximately 9% of solar and wind generation could not be integrated into the grid. Pricing, market design, and transmission bottlenecks also remain real constraints for China.}

_{The Gulf states represent a third model. Their strategy is to convert abundant energy and capital directly into computing capacity. The UAE has brought 5.6 gigawatts of capacity online through the region’s first civilian nuclear energy program and aims to expand its data center capacity to the gigawatt scale. The geopolitical influence created by hydrocarbon wealth could thus become a new source of technological leverage. Yet energy abundance alone does not translate into AI power. Because of extreme temperatures, cooling accounts for more than 40% of data center energy consumption in the country. Access to advanced chips also remains uncertain under the US chip diffusion regime. The UAE’s major overseas data center partnerships therefore serve as a hedging strategy.}

_{Europe’s test}

_{The fourth model deserves particular attention because it provides the clearest illustration of the relationship between digital sovereignty and infrastructure sovereignty. Europe has placed technological sovereignty, European computing capacity, cloud autonomy, and domestic chip production at the center of its agenda. At the same time, the energy layer on which these ambitions must rest has experienced its most severe disruption in years. Since the war in Ukraine, liquefied natural gas has largely replaced Russian pipeline gas. This has been a gain in terms of supplier diversification, but it has also shifted Europe from dependence on a single pipeline to dependence on the volatility of a global spot market and primarily on U.S. exports. Industrial electricity prices reaching approximately two and a half times U.S. levels is another consequence of this structure. This raises a fundamental question: Can Europe establish meaningful AI sovereignty without a sufficiently secure energy layer beneath it?}

_{Events during the past summer revealed a second vulnerability. Successive heat waves pushed air-conditioning use to record levels. At the same time, rising river temperatures forced France to reduce nuclear output, while low water levels affected other power plants in Central Europe. Supply tightened precisely when demand peaked. Because France exports electricity to its neighbors, this became a continental problem rather than a national one. Solar generation covered much of the shortfall during the day, but the real strain emerged in the evening. The implications are clear for data centers, which require uninterrupted power around the clock. What determines whether an energy system can support AI infrastructure is not its annual average performance, but how it performs during the most difficult hours. This is also why differences within the continent are becoming decisive. France has a strong foundation, generating approximately 70% of its electricity from nuclear power and announcing a 109 billion euro ($218 billion) plan for AI infrastructure. In Ireland, by contrast, data centers now consume nearly one-quarter of the country’s electricity. The grid operator suspended new connections for years and now requires facilities seeking a connection to provide their own generation capacity.}

_{This internal divergence is creating a new geography of digital infrastructure in Europe. Computing capacity is shifting toward regions where electricity access, grid capacity, cost, and political approval converge. Europe cannot regulate its way to AI sovereignty. It must also build the infrastructure that makes sovereignty possible. The capacity to set rules remains Europe’s greatest strength, but meaningful technological sovereignty increasingly depends on the ability to physically host computing infrastructure and power it during the most difficult hours. Europe also demonstrates that sovereignty does not mean complete self-sufficiency. Because the continent’s countries operate through interconnected electricity markets, cross-border transmission lines, and shared institutions, none is building a fully self-sufficient system. The issue is not choosing between dependence and independence. It is distinguishing manageable interdependence from strategic vulnerability.}

_{Infrastructure sovereignty}

_{The meaningful question is not which country has more electricity, but which states can connect energy, power grids, data centers, chips, capital and political decision-making capacity. Energy has always been strategic because it powers industrial production, transportation and military capabilities. Today, it is acquiring another function. AI is turning electricity into a direct input for a broad set of capabilities, from model training and large-scale inference to autonomous systems and decision-support systems. This means that another function is being added to the strategic role of energy.}

_{This pressure is already shaping energy investment. In the short term, the additional load will be supported by natural gas, renewable energy, storage, existing nuclear capacity, and behind-the-meter generation built without waiting for a grid connection. In the longer term, nuclear power will play a larger role. Yet even if every nuclear agreement announced by technology companies to date is implemented, the resulting generation will not meet even one-fifth of expected demand. Regardless of which technology is selected, investment in transmission and the grid will remain essential. This makes the question of who bears the cost an inherently political issue. As the costs of new transmission lines, transformers, and power plants are passed on to existing consumers, public opposition is growing alongside concerns about water and land use. In short, the question is this: Who can generate and transmit the energy AI requires, how quickly, from which sources, at whose expense, and under whose control?}

_{AI sovereignty must therefore be understood in layers. Energy, generation, the grid, data centers, chips, models and applications are not independent components. Scarcity, vulnerability or external control in the lower layers determines the limits of capability and sovereignty in the upper layers. Ambitions for sovereign AI and national cloud infrastructure remain constrained by those limits. In this context, energy sovereignty does not mean complete independence. It means the capacity to secure sufficient, reliable, and affordable energy for strategically important technological infrastructure without accepting unacceptable external vulnerabilities. This capacity can also be built through alliances, imports, cross-border electricity connections, cloud partnerships and diversified supply chains.}

_{AI is not reversing digitalization. It is making visible and intensifying the physical foundations on which digitalization has always depended. The next stage of the race will not be limited to laboratory breakthroughs, chip production, model performance, and talent. The ability of states to build, finance, connect, secure, and manage the energy and infrastructure systems that sustain large-scale computing will also be decisive. AI may be digital. But the economic and geopolitical power that makes it possible is becoming increasingly physical.}