AI Data Centers: Energy Demand on the Rise
The AI boom is shifting the focus from talent and software to fundamental infrastructure such as electricity, space, water, and networks. These physical limitations pose a new stress test for digitalization, as the energy hunger of AI data centers challenges existing capacities.
Source: Own Representation
AI Energy Demand
Data centers are already significant electricity consumers today. According to the International Energy Agency (IEA), they account for around 1.5% of global electricity consumption, equivalent to approximately 415 TWh per year ( energy.ec.europa.eu). In its base scenario, the IEA forecasts that this global electricity consumption by data centers will nearly double to about 945 TWh by 2030, which would then account for just under 3% of global electricity consumption ( (iea.org). The projected growth from 2024 to 2030 is around 15% per year, which is significantly higher than the electricity consumption growth in other sectors ( (iea.org).
This dynamic is particularly evident in the USA. A report by the US Department of Energy (LBNL report) shows that data centers accounted for about 4.4% of total US electricity consumption in 2023. By 2028, this share could rise to between 6.7% and 12% ( (energy.gov). Absolute consumption increased from 58 TWh in 2014 to 176 TWh in 2023 and could grow to between 325 and 580 TWh by 2028 ( (energy.gov). These developments are leading grid operators and regulators to increasingly view peak loads as a supply risk.
Source: navitassemi.com
The projected increase in electricity demand by data centers due to AI by 2030 shows a significant doubling of consumption.
However, energy demand is only part of the challenge. Many regions are struggling not primarily with a lack of electricity generation, but with insufficient transmission and connection capacities, as large new loads appear quickly and are clustered locally. The North American Electric Reliability Corporation warned in November 2025 that rising demand from data centers is reducing power reserves in the US and increasing the risk of bottlenecks during extreme weather ( (reuters.com). This highlights the system's sensitivity to new large consumers, even before all announced AI projects go online.
In addition to electricity consumption, cooling plays a crucial role. In the US context, direct water consumption by data centers is estimated at around 17 billion gallons for 2023, with hyperscalers and colocation providers accounting for the largest share ( (pewresearch.org). Although European locations sometimes use different cooling concepts, the sheer size of the new facilities is shifting the discussion from abstract efficiency goals to concrete location conflicts.
Energy Sources
Given the time required for new wind, solar, and grid projects, gas is again being considered as a quick backup in many strategies. This is reflected in the rhetoric of major LNG exporting countries. Qatar's Energy Minister Saad al-Kaabi emphasized on December 6, 2025, that the growing energy demand from AI will support gas demand, and predicted global LNG demand of 600 to 700 million tons per year by 2035 ( (reuters.com). Such statements are politically relevant as they link energy security and climate goals while justifying the expansion of fossil fuel infrastructure.
In the US, signals from two directions are intensifying: data centers as a new electricity load and LNG exports as an additional gas consumer. Reuters reported in October 2025 on a wave of deals in the US gas sector, justified in part by the demand from AI data centers and LNG ( (reuters.com). This shows how closely digital growth narratives and traditional energy assets are once again linked.
Source: statista.com
The comparison of electricity generation for data centers by energy source and region highlights the diverse energy mixes.
In parallel, nuclear energy is experiencing a new wave of attention, both politically and commercially, as it is low-carbon and capable of providing baseload power. Large technology companies are seeking long-term power purchase agreements that are not only 'green' but also predictable. This development is openly discussed in the industry, including large PPAs (Power Purchase Agreements) and investments in new projects ( (trellis.net). While the specifics vary by company and country, the underlying motive is clear: AI loads are high, constant, and expensive to interrupt, making nuclear options seem like a strategic hedge for some locations.
However, the crucial factor remains the timeline. New reactors or advanced SMR (Small Modular Reactor) concepts will not solve acute grid problems in the next one to three years. In the medium term, however, they can be an important building block, especially where political acceptance and approval processes are in place. This makes nuclear energy a part of the infrastructure debate, but not its sole solution.
Regional Impact
Europe is now officially addressing the issue. The EU refers to the IEA figures and describes data centers as a growing challenge for electricity systems and climate policy ( (energy.ec.europa.eu). This suggests stricter efficiency requirements, more transparent reporting obligations, and more intensive site regulations. In Switzerland, the debate will unfold similarly, but with an even stronger focus on grid bottlenecks, winter supply gaps, and competition for land. This is because the system is smaller, and large new loads have a greater impact more quickly. This conclusion is a direct transfer of European and North American findings to a compact power system.
For operators and municipalities, the decision-making logic is shifting. Electricity prices are no longer the sole determining factor; instead, the question is whether a location can be connected quickly enough, whether waste heat can be utilized, and how cooling and water availability can be secured. This is where the real competitive advantage emerges: those who plan infrastructure proactively can expand AI capacity without immediately encountering political resistance or grid limitations.
Strategic Planning
The AI boom is no longer a pure software issue. The IEA projections until 2030 show that data centers will grow noticeably, but not dominantly, in global electricity consumption, with particularly strong increases in the USA, China, and Europe ( (iea.org). In practice, however, local infrastructure decides, not the global percentage. Where AI clusters encounter weak grids, scarce cooling resources, or long approval processes, energy hunger becomes the decisive location factor.
Source: mdpi.com
Strategies for optimizing energy consumption and reducing peak loads in data centers are crucial for sustainable operation.
In the short term, gas and LNG remain the most robust hedge for many players, while nuclear energy and renewable capacities tend to address the medium to long term ( (reuters.com). A clear separation of these time horizons and consistent consideration of infrastructure as a central factor enable more realistic planning and more successful implementation of AI projects in the coming years.