How energy innovation is reshaping infrastructure for AI workloads
AI is redrawing the global energy and infrastructure map. In the past year alone, energy demand from hyperscale AI data centers has surged so dramatically that innovators, operators, and governments are turning to unconventional solutions: redeveloping the site of America’s worst nuclear accident into an energy and data hub, building server farms deep inside Norwegian mines, and leaning into hydropower to operate a supercomputer in the Swiss Alps. As AI models become increasingly complex and power-hungry, the question of where and how companies run them is emerging as one of the defining industrial questions of the decade.
AI’s Appetite for Energy
Data centers have long consumed significant amounts of electricity, but the surge in generative AI has heightened that demand. Training a large-scale model like GPT-4 or Google Gemini requires tens of megawatt-hours for a single run, and inference workloads for millions of users drive constant strain on power grids. The International Energy Agency estimates that electricity consumption from data centers, cryptocurrencies, and AI could more than double by 2026, potentially reaching 1,000 terawatt-hours annually.
Other recent investigations have underscored this surge in energy usage. Research from Deloitte forecasts that AI-related data center power demand in the U.S. alone could grow 30-fold by 2035, prompting closer coordination between operators and utilities as hybrid energy-compute hubs emerge as the future model. Analysts at Bain & Company project that the global data center industry’s energy consumption could more than double by 2027, surpassing 1 million gigawatt-hours (GWh) annually, requiring the integration of on-site or adjacent power systems much like power plants themselves.
AI’s appetite for power is a matter of scale and intensity. Where traditional data centers could stagger workloads and tolerate latency, AI data centers require dense, high-performance computing clusters running GPUs and AI accelerators at near-constant utilization, often 24/7.
These projections have prompted leaders to consider new sources for secure, abundant, and ideally carbon-free power. The result is a wave of approaches that blend energy innovation with next-generation data center design, including nuclear, geological, and hydro-based options.
Nuclear Power for AI: From Disaster Site to Data Hub
Nuclear provides reliable, around-the-clock baseload power with no direct carbon emissions, making it attractive in an era where ESG concerns increasingly drive data center site selection. And SMRs, smaller and more flexible than traditional reactors, promise faster deployment and potentially lower regulatory hurdles. Yet the pairing of nuclear power with AI also revives familiar challenges: safety concerns, regulatory complexity, and public skepticism, particularly when projects are tied to legacy disaster sites.
In the search for alternative power sources, this dynamic is unfolding at Three Mile Island, the site of the worst nuclear accident in U.S. history, where new proposals aim to repurpose it as a next-generation hub for nuclear power and AI infrastructure. Developers envision pairing advanced atomic power—potentially Small Modular Reactors (SMRs)—with AI workloads, creating a closed-loop ecosystem of constant power and constant compute.
Going Deep: Norway’s Underground Advantage
If nuclear power is one path, nature itself is another. In Norway, the Lefdal Mine Data Center presents a vision of how geographic features can provide sustainable, scalable infrastructure. Built inside a former olivine mine on the edge of a fjord, Lefdal uses cold seawater for highly efficient cooling, achieving power usage effectiveness (PUE) ratings as low as 1.15.
Norway’s abundant renewable hydropower, combined with the mine’s inherent physical security and stable climate, makes it a magnet for companies seeking large-scale, sustainable AI infrastructure. With the European hyperscale cloud market growing rapidly, sites like Lefdal position themselves as prime real estate for AI-focused expansion.
Supercomputing Meets Hydro-Cooling: Switzerland’s Alps Project
This drive for sustainable power is reshaping not just where, but also how, high-performance systems operate. The Swiss Center for Scientific Computing (CSCS) Alps supercomputer in Lugano is a case in point, designed to meet AI-workload demands while embedding energy efficiency into its operations.
The Alps facility is powered by 100% hydropower, giving it a minimal carbon footprint. Electricity generated by Swiss hydropower plants (including pumped-storage and run-of-river types) feeds the data center’s massive power demands.
Cold water from Lake Lugano is also used as a natural cooling resource. It’s circulated through heat exchangers to absorb the heat from the supercomputer’s infrastructure (instead of using energy-hungry mechanical chillers). After picking up heat, the water is returned to the lake at a controlled temperature, and some of the recovered heat is used for district heating (warming buildings in the city).
Where Power Meets Compute
Together, these projects reveal that energy choices are no longer secondary to technology; they’re integral. The future of AI infrastructure depends as much on energy innovation as on compute performance.
AI is reshaping the physical landscape and prompting leaders to rethink how they build compute facilities and supply energy. As modern workloads continue to push resource and infrastructure boundaries, the sites being developed today offer a glimpse of a future where data centers are not only information hubs but also energy hubs, designed from the ground up for performance, efficiency, and sustainability. In this evolving landscape, energy itself may become the ultimate competitive advantage for AI infrastructure.
