This opinion article was written by Josiah Neeley, Senior Fellow at R Street Institute. You should follow him on X.

AI is so hot right now. Currently, over $5 trillion is being invested in AI-related endeavors in the United States, and much of that funding is going to the construction of large-scale data centers that can provide the computing power needed to run AI programs. Data center investment was responsible for 92 percent of U.S. GDP growth in the first half of 2025. 

These data centers require an enormous amount of energy to operate. Some individual data centers already use more power than a medium-sized city, and the growth of data centers is helping to upend long term dynamics in the electricity sector. 

After years of flat or declining electricity usage throughout most of the country, data centers (along with other factors such as electrification and reindustrialization) are projected to significantly increase electricity demand in the next decade. Some of the projections are eye-popping. One projection predicts 50 GW of data center capacity expansion by 2028. Another prominent group has projected that data centers demand could reach 123 GW by 2035. 

Privately, however, few in the electricity sector believe these projections can or will happen. This skepticism is not based on the idea that AI is a bubble, but rather on aspects and constraints in the energy system itself and the time it will take to accommodate growth in demand. 

Some of the skepticism has to do with how demand growth projections are calculated. Consider a company planning to build a new data center that is looking at several possible locations to site the data center. To expedite the process of getting connected to the grid, this company may apply for interconnection with utilities at all of these sites, even though only one site will ultimately be used. These requests then get built into utilities’ demand forecasts, which are aggregated to form the basis of some demand growth projections. This is known as duplicative measurement error and can result in a significant overstatement of how much demand growth there will ultimately be. 

However, even to the extent that projected growth in data center demand is real, achieving that growth faces a more serious set of obstacles. Put simply, even if a data center is built, in order to operate, it needs power, and our current regulatory system is incapable of building enough power and power infrastructure to accommodate such a large growth in demand over so short a time period. 

In most regions of the country, it takes five years for new power plants to simply receive approval to interconnect to the grid. Plants may also face other types of delays. Gas plants, for example, are facing supply chain constraints due to a backlog of orders for turbines. Nuclear plants have also traditionally faced decade-plus timelines for permitting and construction. Building enough power to supply anywhere near the demand in some of the projections listed above in the next five years would require major changes in how the current process operates. 

And building enough new power plants is only the first hurdle. Linking up this power to new demand will require upgrades to the transmission and distribution system, which can involve its own set of delays. To build a new high-voltage transmission line currently takes an average of ten years, the majority of which is taken up by planning and permitting processes. Certain critical grid components, such as large transformers, face their own supply chain issues. And local permitting processes in much of the country can create regulatory roadblocks that stymie needed buildouts. 

Some data centers are looking for creative ways around these obstacles. For example, centers can choose to locate at points on the grid that minimize the need for grid upgrades or buildouts to provide them with power. Some data centers are also choosing to “bring their own generation,” co-locating new generation with their centers that may or may not also be hooked into the larger grid. And data centers may agree with grid operators to be “flexible,” i.e. to reduce their power consumption during peak periods, thus reducing strain on the grid. 

But all of these measures are at best partial solutions. The fact remains that our permitting and regulatory system is not built for speed. In the movie Field of Dreams, a ghostly voice tells Kevin Costner, “if you build it, they will come.” But when it comes to data centers and energy, that’s just a bunch of hippie nonsense. Unless we change the way energy infrastructure is built in this country, you can build as many data centers as you want, but if you want to use them, you’ll have to wait.