The artificial intelligence (AI) sector is increasingly influencing global energy consumption. With AI advancing, demand for generation and grid capacities is growing. Key to AI implementation is a stable supply of low-carbon or, ideally, even zero-carbon electricity, which nuclear power can provide. This is what the Energy and Artificial Intelligence report by the International Energy Agency deals with.
In today’s technological landscape, information processing involves using energy. Physically, a 1 or a 0 means the presence or absence of energy. Machine learning is also energy-dependent. For example, in a Hopfield neural network, named after the Nobel laureate who received the prize for foundational discoveries and inventions in machine learning, reference values are programmed to achieve the lowest energy state, or “valley,” towards which the system strives during training.
To date, technological breakthroughs, combined with lower computing costs and access to massive amounts of data, have propelled the growth of AI so dramatically that it has evolved from a scientific pursuit into an industry influencing production processes, everyday life, politics, art—and the energy industry.
The report notes that determining AI’s exact share in the overall data center loads is challenging since data centers handle more than just AI tasks. Conversely, AI is not limited to data centers. Nevertheless, the authors of the report use the term “data centers.”
According to the IEA, data centers account for about 1.5% of global electricity consumption in 2024 (415 TWh). The United States led the data center sector (45%), followed by China (25%) and Europe (15%).
On average, electricity consumption growth by data centers accelerated from 3% annually between 2005 and 2015 to 10% between 2015 and 2024. Since 2017, global electricity consumption by data centers has grown by roughly 12% on the annual average. This is more than four times faster than the overall growth of electricity demand. In China, the average annual increase was 15% from 2015 to 2024, while the US saw a 12% growth.
In developing economies where energy use is growing across various sectors, data centers contribute about 5% of total growth. In countries where energy consumption has long remained almost flat, data centers account for about 20% of growth.
However, it should be noted that data centers are not the fastest-growing electricity consumers globally. According to IEA estimates, they lag behind heavy industry and other manufacturing sectors, household appliances, space cooling, heating, and electric transport. Still, at the local level, data center connections can cause problems related to power generation and grid transmission capabilities.
The IEA outlines four scenarios, or cases, for electricity consumption by data centers by 2030. In the Base Case, electricity consumption will more than double compared to current levels, reaching 945 TWh. As noted in the report, this is slightly higher than Japan’s current consumption of electricity.
In the Lift-Off Case, power consumption will exceed 1,260 TWh by 2030. The High Efficiency Case projects consumption to reach about 800 TWh by the same year thanks to energy conservation. The Headwinds Case forecasts around 670 TWh. The IEA also extended projections to 2035, but uncertainty was so high that estimates ranged from 700 TWh to 1,720 TWh depending on the case.
Industries employing AI reduce their energy use by optimizing production processes. “The industry of the future will be increasingly digitalized and automated; countries and companies that take the lead in integrating AI into manufacturing will jump ahead. AI applications can accelerate product development, lower costs and increase quality. Widespread adoption of existing AI applications to optimize processes in industry can lead to energy savings equivalent to more than the total energy consumption of Mexico today,” the report says. This translates to roughly 8 EJ (exajoules) by 2035, or over 222 TWh.
Thus, AI-driven efficiency gains may curb the overall growth in electricity demand but only partially. Some downward influence on electricity consumption by data centers is also achieved through improving server and cooling equipment energy efficiency.
30% of global data center electricity needs are met by coal-fired power plants.
Data centers are comparable in electricity consumption to metallurgical plants, including energy-intensive ones like aluminum smelters. Moreover, data centers are a rapidly growing and increasingly significant electricity consumer. This raises the question: How can their power supply be reliably ensured? Locating a data center depends heavily on a reliable power source, competitive electricity prices, and adequate grid capacity.
Sufficient generation capacity is the most critical factor in making a decision to build a data center. As noted in the report, 30% of global electricity needs for data centers are met by coal-fired power plants, followed by renewable energy (27%), gas (26%), and nuclear (15%). After 2030, the share of coal generation in powering data centers will somewhat decline, gas will remain steady, while the shares of renewables and nuclear will grow.
The report says that grids are a challenge in many regions: “Grid connection queues for both supply and consumption projects, including data centers, are long and complex. Building new transmission lines can take four to eight years in advanced economies and wait times for critical grid components such as transformers and cables have doubled in the past three years.”
To ensure reliable and manageable electricity supply, technology and energy companies are partnering. In the United States, several major data center operators have teamed with power generation and distribution companies building new gas-powered facilities.
In 2024, similar agreements were signed between large American tech companies and nuclear power plants. In September, Oracle announced plans to use three small modular reactors (SMRs) to power its 1 GW data center. In mid-October, Amazon reached an agreement with Energy Northwest to build four advanced power units in the state of Washington. The nuclear station will have a capacity of 320 MW expandable to 960 MW. Technology for the station will be provided by X-energy, a company in which Amazon also invests. The project is expected to be finished in the early 2030s. Additionally, Amazon is working with Dominion Energy on the plans to build a 300 MW unit. Google signed an agreement with Kairos Power to purchase electricity from its upcoming SMRs, with a 500 MW fleet planned by 2035. The company is developing a demonstration molten salt reactor, Hermes, in Oak Ridge (Tennessee), scheduled to become operational in 2027. The next step will be constructing a two-unit station. “To date, plans to build up to 25 GW of SMR capacity associated with supplying the data center sector have been announced worldwide, almost all of them in the United States, although projects are at varying stages of maturity and certainty. The first projects are expected to start to materialize only towards the end of this decade,” the report says.
Many technology companies and major data center operators aim to cut emissions and use clean energy by purchasing electricity from renewable sources, as noted in the report. Often these are financial agreements. This means that other sources – natural gas or coal – are used in reality to satisfy actual electricity needs. However, some tech companies still strive for real emission reductions. Google and Microsoft sign agreements for electricity supplies from low-carbon energy sources – hydropower, nuclear power, geothermal power plants, and natural gas stations with carbon capture. For instance, in September 2024, Microsoft and Constellation Energy signed a 20-year contract covering the restart of Unit 1 at the Three Mile Island Nuclear Generating Station.
Assessing the prospects of nuclear power plants supplying electricity to data centers, the authors of the IEA report focus attention on small modular reactors, but large plants are also in play. In addition to the restart of a unit at Three Mile Island, there is also a contract between Amazon Web Services and Talen Energy for providing 960 MW of capacity from the Susquehanna Steam Electric Station, a nuclear facility in Pennsylvania consisting of two units rated at 1,257 MW each. Furthermore, in May, Google announced partnership with Elementl Power on three nuclear projects due to a sharp rise in AI-driven electricity consumption. Google will provide finance, with each of the three projects to have a capacity of 600 MW and classified as medium-sized nuclear plants under the IAEA criteria. These examples show that nuclear power plants of all sizes can be employed for connecting AI-related data centers in the US and beyond.
In Russia, the “data center + nuclear plant” model was pioneered in September 2019 as the Kalinin Data Center was put into operation near the Kalinin Nuclear Power Plant, supplying the data center with electricity. Since then, Rosatom has developed its own geographically distributed data center network, currently ranking among the top three in Russia by installed capacity. The Russian nuclear corporation also collaborates with AI developers.
Rosatom designs, constructs, and adapts small, medium, and large nuclear power plants for any projects, all of which can be used to supply data centers with electricity. With this expertise, Russia’s nuclear industry can offer integrated nuclear generation solutions for building and powering data centers in Russia and worldwide.
In Russia, the “data center + nuclear plant” model was pioneered in September 2019 as the Kalinin Data Center was put into operation near the Kalinin Nuclear Power Plant.
Photo by: Wikipedia, Unsplash, Kalinin NPP
