When you ask AI a question on a smartphone, an answer appears almost at once. But the phone is not doing all the calculation. Thousands of high-performance computers are running inside large buildings far away. These facilities are called data centers.
A data center uses electricity for its computers and for equipment that cools them when they get hot. More computation is needed as more people use AI and ask it to create not only text but also images and video. Every time we use AI, electricity is being consumed somewhere we cannot see.
How should that electricity be supplied? Along with solar and wind power, nuclear power and smaller reactors are receiving attention. Smaller reactors, however, are not ready for immediate, large-scale use. This article separates the reasons for the interest from the problems that still stand in the way.
How much more electricity will AI use?
The International Energy Agency is an international organization that studies energy around the world. It is known as the IEA. The IEA projects that electricity used by the world’s data centers will rise from 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030.
A terawatt-hour is a unit used for a very large amount of electricity. The IEA expects 950 terawatt-hours to be about 3% of all the electricity used worldwide in 2030. This number does not predict the future with certainty. It can change depending on how widely AI is used, how energy-efficient machines become, and how much construction is delayed.
Data centers need stable electricity during both the day and the night. When computation suddenly increases, electricity use can change quickly too. An outage could stop services, so backup equipment is also necessary.
Building power stations is not enough. Transmission lines are needed to carry electricity over long distances. Substations are needed to change electricity into a form that can be used at the destination. If a power source is far from a planned data center, the equipment between them can become the shortage.
Why is nuclear power one of the options?
A nuclear power plant obtains heat from fuel such as uranium and uses that heat to turn water into steam and generate electricity. It can keep generating for long periods without depending directly on the weather or the time of day. This is why it is seen as a possible match for data centers that need power all the time.
Nuclear fuel provides a large amount of energy, so a nuclear facility may use less land than another facility producing the same amount of electricity. But it would be wrong to look only at the reactor building and call it compact. Space is also needed for safety systems, security, cooling, grid equipment, and management of used fuel.
One design now attracting particular interest is the small modular reactor, or SMR. The idea is to make each reactor smaller than a conventional large nuclear plant and repeatedly manufacture parts of the same design in factories. Several units could be combined when more electricity is needed. Their size, manufacturing method, and power output differ from one design to another.
Producing electricity near a data center
Researchers are studying combinations of small reactors and batteries placed near data centers. A system that links generators, batteries, and buildings using electricity within a small area is called a microgrid. It aims to keep essential equipment running even when the larger outside grid has a problem.
The U.S. Department of Energy has published a research plan for combining small reactors, heat-storage equipment, and other kinds of generation. The plan says that researchers need to test resilience during outages and whether stable power can actually be delivered. In other words, this is not a finished system already being sold widely. Many questions still need to be tested.
Small does not mean quick or easy to build
In its 2025 report, the IEA expects small reactors to begin contributing to data-center electricity around 2030. Until then, renewables such as solar and wind, natural gas, and existing grids are expected to play large roles in meeting growth. The plan is not for small reactors alone to provide the extra electricity needed over the next few years.
The first problem is the process of checking safety. Specialists must examine a reactor’s design, construction, operation, and preparations for accidents. In the United States, the Nuclear Regulatory Commission has created a new review process. Even so, submitting an application does not give permission to operate. Checks for each site and reactor remain necessary.
The second problem is building the first few units. Making the same thing many times may become cheaper as workers and factories gain experience. That benefit requires enough orders and experience first. Early projects still have to pay for design changes, construction management, and worker training.
The third problem is gathering everything that is needed. Facilities must produce reactor fuel. Large metal parts and materials whose safety can be demonstrated are required. People with specialized skills are needed too. The system that keeps delivering these materials and people where they are needed is called a supply chain. The U.S. Department of Energy’s support program also lists design, safety review, supply chains, and site preparation among the gaps that must be addressed.
Cost remains important. A small reactor does not automatically produce cheap electricity. A fair comparison must include not only construction but also borrowing money, operation, and dismantling the facility after it is retired.
What should we watch on the way to the 2030s?
It is too early either to declare that AI absolutely requires nuclear power or to say that small reactors are pointless because they cannot be built immediately. In the near term, strengthening existing grids can help sooner. More solar, wind, and battery storage are other options. Data centers can be built in regions with spare electricity, and some computation can be moved to different times.
Small reactors may then become one option for the 2030s. The number of projects announced by companies is not enough to judge progress. Did a design finish its safety review? Did construction begin and finish on schedule? What did it actually cost? Did it operate reliably? Can fuel continue to be supplied?
It has not yet been decided whether AI will need small reactors. Their ability to provide stable electricity is only one part of the answer. Safety, time, cost, and the ability to gather everything a reactor requires must also be demonstrated.