The Need for Electricity Abundance

I n addition to water use, the leading worry around data centers is that they consume a great deal of electricity. Unlike with water, this concern is not overblown. But the issue demands a serious policy response, not thoughtless obstruction.

That response can be summed up in two words: electricity abundance.

Data centers currently consume approximately 5 percent of all electricity generated in the United States. That figure is projected to grow exponentially amid the AI buildout, doubling or tripling in the next few years. The National Electrical Manufacturers Association estimates that demand from data centers may push overall power consumption up 55 percent by 2050. That is an enormous amount of juice.

The leading fear is that this surging demand will place great strain on electricity prices for everyone else, especially residential users.

To this point, there is little evidence that data centers have broadly increased electricity rates, even though future price effects remain far from clear. An analysis by S&P Global found “no observable correlation between the five-year increase in data center power demand and change in residential electricity rates at the state level.” The two states with the largest growth in data-center demand, Virginia and Texas, saw residential electricity prices rise in line with the national average.

In fact, there is some evidence that data centers could reduce prices. One causal study finds that data centers “caused average retail electricity rates to fall modestly in the United States from 2015 to 2024” by spreading the fixed utility costs of transmission and distribution across a larger pool of customers. “For every 10% increase in data center capacity,” the authors calculated, “average residential retail prices fell by approximately 0.4% on average,” causing rates to be 6 percent lower than what they would have been without new data centers.

The national rise in electricity rates since 2020, therefore, has far more to do with general inflation and costly maintenance. Yet the scholarly consensus is that the explosion in AI data centers is likely to boost prices going forward, as extraordinary demand outstrips available power-generation capacity.

That scenario won’t necessarily be the case. Shawn Regan of the Manhattan Institute notes a negative correlation among states between load growth, or rise in electricity demand, and utility rates from 2019 to 2025 due to the aforementioned economies of scale. States that saw demand decline during this period, such as California and New York, experienced the highest rate hikes. Meanwhile, “states with greater load growth generally saw smaller price increases, or even price declines” when adjusted for inflation. Texas is the largest among those states that allowed power-generation capacity to expand alongside new customers.

There is no one correct level of demand for electricity, so long as it is matched by equivalent supply. The problem in many states is that capacity is severely constrained by regulation or deliberately made more costly through mandates for certain types of energy over others. California, for example, requires utilities to procure 60 percent of power from renewable sources by 2060, and New York mandates 70 percent, forcing companies to retire traditional plants. Red tape makes it arduous to onboard even renewable sources. In California, it takes around nine years to connect a new power station to the grid, compared with 3.5 years in Texas.

To ensure that data centers and the public have adequate access to electricity, the first thing policymakers must do is ease unnecessary restrictions so that supply may grow again. National generation capacity previously expanded in every decade since 1950, with growth halting only in the last 20 years. Time to turn the switch back on.

Progressive states will have to lift their hold on fossil-fuel plants. Economic trends may continue to phase out coal, but natural gas is abundant and historically cheap in real terms. Conservatives may scoff at renewables, but wind and solar are increasingly economical without subsidies. The principal barrier to these sources is permitting bottlenecks for transmission lines, which is why the relatively laissez-faire Texas dwarfs restrictive California in renewable energy.

Utility regulators also need to adapt. Traditionally, they have instructed utilities to socialize the costs of new generation and transmission across their entire customer base to avoid discriminatory rates. Since data-center developers are mobile, however, several states are creating new rate categories for large-load customers to pay for all the infrastructure costs they incur. Others are requiring data centers to bring their own power to the grid and have existing customers absorb any excess generation.

Another idea is to separate data centers’ power use from the grid altogether. State and federal proposals would exempt power stations from regulations if they are islanded to supply only private, contractual customers. Data centers could fuel themselves without sitting in interconnection lines, with no costs shifted to normal ratepayers.

Artificial intelligence will require many more power plants and transmission lines than America currently has — there’s no way around that. But electricity abundance would be a worthy aim even without the needs of data centers, and intelligent policy reforms can meet the demand.