Powering AI: Which Energy Sources Stand Out?
- 17 hours ago
- 7 min read
Source: Sustainable Energy in America Factbook, Business Council for Sustainable Energy (BCSE).
Rapid growth in AI-driven electricity demand is reshaping the landscape of power generation. Nuclear, geothermal, natural gas, coal, and renewable energy sources each present distinct strengths and constraints in terms of scalability, reliability, economics, and deployment timelines. While some technologies offer firm, continuous power, others benefit from faster construction or lower emissions but face structural limitations. These trade-offs illustrate the complex balance required to meet emerging energy needs in an increasingly power-intensive digital economy.
Why This Matters
If energy investments are not carefully selected or balanced across complementary sources, AI’s full potential may remain constrained by electricity supply limitations.
If shifts in energy policy are not fully understood, the pace and geography of AI expansion could diverge sharply from expectations.
The rapid expansion of AI-driven data centers is pushing electricity demand upward at an unprecedented pace, while generation capacity is struggling to keep up. According to International Energy Agency (IEA) estimates, U.S. data center electricity consumption is projected to grow by 133% by 2030, widening the structural gap between supply and demand.
The surge in electricity demand from AI data centers raises critical questions about what types of power generation can scale fast enough, deliver firm output, and remain economically viable. In practice, the options are far more limited than they appear.
Source: IEA Energy supply for AI - Global electricity supply to meet data centre demand.
Nuclear
Nuclear power is the energy source most frequently cited in discussions around meeting AI-driven electricity demand. As International Atomic Energy Agency (IAEA) Director General Rafael Mariano Grossi has noted, nuclear is the “only energy source that can meet combined demands of low-carbon generation, 24/7 reliability, massive power density, grid stability and genuine scalability.”
This advantage is reflected in capacity-factor data (a measure of actual electricity generation compared to the plant’s maximum possible output over time): in 2024, the U.S. Energy Information Administration found that nuclear plants operate at an average capacity factor of 92.3%, far exceeding other generation technologies. The next-best performer, geothermal, averages around 65%, while coal stands at 42.6%. Hydropower, wind, and solar all typically operate below 35%.
Unsurprisingly, nuclear power remains a focal point of global energy planning. Today, 441 nuclear reactors are operating worldwide, with 71 additional units under construction. The United States alone hosts 94 operating reactors, more than any other country, and has ten new reactors planned. In addition, there are three nuclear plants in the process of being restarted in the U.S.: Holtec Palisades, Crane Clean Energy Center (Three Mile Island Unit 1) and Duane Arnold.
Name | Location | Off taker | Generation Capacity | Planned Restart Time |
Covert, Michigan | Not formally confirmed; expected merchant / potential corporate PPAs | ~800 MW | 2026 | |
Crane Clean Energy Center (Three Mile Island Unit 1) | Middletown, Pennsylvania | Microsoft (long-term power purchase agreement linked to AI/data center demand) | 880 MW | 2027 |
Palo, Iowa | Google (reported discussions linked to data center power needs) | 624 MW | 2029 |
Nuclear Plants in the Process of Being Restarted in the U.S. Source: Author’s compilation.
Due to safety concerns associated with large conventional plants, Small Modular Reactors (SMRs) are gaining attention. Unlike traditional gigawatt-scale facilities, SMRs feature smaller footprints, enhanced safety systems, and the potential to be deployed near industrial sites, including data-center campuses.
Looking ahead, projections remain optimistic. According to the IEA World Energy Outlook 2025, global nuclear capacity could expand by at least one-third, and potentially up to 70%, by 2035.
Yet nuclear power’s limitations are difficult to ignore.
First and foremost, the consequences of failure are severe, as demonstrated by historical accidents such as Chernobyl (1986), Fukushima (2011), and Three Mile Island Unit 2 (1979), which continue to shape public perception and regulatory caution.
More immediately, nuclear power is highly capital-intensive. Upfront construction costs account for a large share of total project expenditure, and development timelines are long: under well-planned conditions, the full process to start a new nuclear plant from licensing to commissioning typically takes 11–12 years, with the average construction time of a single reactor unit around seven years.
Part of the U.S. Department of Energy’s mission in this new administration is to speed up the development of domestic nuclear power, largely for the purpose of meeting AI-influenced power demand. This includes speeding up the permit process. The ten aforementioned new nuclear reactors have a goal of being built by 2030. Still, this is an ambitious timeline, and there is skepticism surrounding it.
Source: Importance of Advanced Planning of Manufacturing for Nuclear Industry, Shykinov, Rulko, and Mroz, DOI: 10.1515/mper-2016-0016.
As a result, while nuclear power remains the most technically attractive option, it is unlikely to scale fast enough on its own to close the short-term supply gap, making additional energy sources unavoidable.
Geothermal
The most promising non-nuclear source of reliable, clean power is geothermal energy. Geothermal power features a relatively high capacity factor, allowing it to deliver electricity on a continuous, around-the-clock basis rather than intermittently. However, its use is currently concentrated in a few countries with easily accessible and high-quality resources, including the United States, Iceland, Indonesia, Türkiye, Kenya, and Italy.
Advances in enhanced geothermal technologies have significantly expanded its theoretical potential. Estimates suggest that the full technical potential of next-generation geothermal systems ranks second only to solar PV among renewable technologies and, in theory, could exceed global electricity demand 140-times over. Using thermal resources at depths below 8km can deliver almost 600 TW of geothermal capacity with an operating lifespan of 25 years. With continued technological improvements and reductions in project costs, geothermal power could meet up to 15% of global electricity demand growth to 2050. This would mean the cost-effective deployment of as much as 800 GW of geothermal power capacity worldwide, producing almost 6,000 TWH per year-- equivalent to the current electricity demand of the U.S. and India combined.
Commercial-scale geothermal power generation is becoming a realistic option. For example, Fervo Energy’s flagship project, Cape Station, is currently under construction in Beaver County, Utah. The project’s initial 100 MW phase is expected to begin delivering power to the grid in October 2026, marking one of the first commercial-scale Enhanced Geothermal Systems (EGS) to reach operational status globally. An additional 400 MW expansion is planned for 2028. Earlier pilot developments, such as Fervo’s Project Red in Nevada, supported by a pioneering partnership with Google, demonstrated the technical feasibility of EGS and helped pave the way for commercial-scale deployment.
Despite this promise, geothermal power faces clear limitations.
Firstly, it remains highly dependent on geological conditions. In some regions, geothermal resources are readily apparent, signposted by surface features such as geysers and hot springs. In many others, however, usable heat resources lie thousands of feet underground, making exploration technically challenging and uncertain.
Secondly, like nuclear energy, geothermal projects are constrained by high upfront capital costs and lengthy permitting and approval processes. Deployment replication is slow, limiting the speed at which geothermal can be scaled to meet surging electricity demand from AI data centers.
Finally, compared with other energies, geothermal remains relatively immature, with many projected capacity targets still largely theoretical.
Source: Sustainable Energy in America Factbook, Business Council for Sustainable Energy (BCSE).
Natural Gas and Coal
Gas-fired generation can be built and commissioned far more quickly than nuclear or large-scale geothermal projects, often within a few years. It benefits from decades of existing infrastructure, including pipelines, storage, and grid connections, as well as mature technology. Natural gas is the third-largest source today, meeting 26% of the data center’s electricity demand globally. For the U.S., the number is over 40%.
Electricity generation for data centres by fuel in the United States, Base Case, 2020-2035. Source: IEA Energy supply for AI - Global electricity supply to meet data centre demand.
But these advantages come with significant drawbacks. Firstly, expanded reliance on gas sits uneasily with decarbonization goals: while gas burns more cleanly than coal, it is still a fossil fuel, and methane leakage across the supply chain can substantially erode its climate advantage. Secondly, gas supply itself may struggle to keep pace with the scale of projected demand growth. Rising electricity consumption from AI and data centers, combined with increasing gas use in Asia and continued European demand, could fundamentally alter global gas markets. Saad al-Kaabi, CEO of QatarEnergy, has warned that these trends could turn an expected global liquefied natural gas (LNG) supply glut into a shortage by 2030.
Coal, most obviously, also poses a fundamental challenge due to its high greenhouse gas emissions. While the IEA expects coal use in U.S. data-center power supply to decline over time, recent policy signals suggest a more complicated outlook. The current U.S. administration has taken a critical stance toward renewables and supported extending or restarting coal plants, keeping coal in the conversation as a short-term capacity option. Coal continues to play a dominant role in China’s data-center electricity mix, where it currently accounts for nearly 70% of supply. Between 2024 and 2030, coal is also expected to remain the largest source of electricity for Chinese data centers.
Other Renewables: Solar, Wind, and Hydropower
When it comes to other renewables, each presents its own mix of advantages and constraints.
Source: Enegy and AI.
Solar photovoltaic (PV) power offers one of the fastest deployment timelines among low-carbon energy sources and can often be developed within relatively short timeframes. Wind energy also benefits from relatively rapid construction, making both technologies attractive options for expanding generation capacity. However, several limitations reduce their suitability for meeting AI-driven electricity demand.
Firstly, while deployment timelines are generally short, supply-chain bottlenecks, grid interconnection challenges, and lengthy permitting processes can significantly delay project delivery in practice.
Secondly, both solar and wind are intermittent sources of power, producing variable output that does not consistently align with the continuous, high-load requirements of hyperscale data centers without substantial investments in storage or complementary firm generation.
Thirdly, large-scale renewable installations require extensive land footprints, creating siting challenges, particularly in regions with competing land-use priorities.
Finally, evolving policy dynamics and regulatory preferences introduce additional uncertainty, which may further slow renewable deployment and investment decisions.
Hydropower typically faces the longest development timespan. This makes it more suitable as a long-term strategic infrastructure rather than a rapid response to immediate electricity shortages driven by AI expansion.
Take Action
Technology companies and data-center operators: Diversify power procurement strategies by combining long-term contracts for firm energy with renewables and storage to ensure reliability and manage energy price risks.
Policymakers: Prioritize streamlined permitting and long-term regulatory clarity to accelerate deployment of firm, low-carbon power sources.
Comments