Data center energy demand is forcing utilities to rethink grid planning
The rapid rise of artificial intelligence is reshaping electricity markets at a pace few anticipated. Since the launch of ChatGPT in November 2022, U.S. electricity demand has increased by more than 4%, exceeding the total growth recorded during the previous 15 years. That sudden surge has thrust data center energy demand to the center of the nation’s energy debate, exposing the growing strain on an electric grid that was largely designed for decades of relatively stable consumption.
According to an in-depth analysis published in June 2026 in the Bulletin of the Atomic Scientists by Stephen Bessasparis, a senior consultant at energy consulting firm E3, the explosive expansion of digital infrastructure is forcing utilities and policymakers to move from reactive planning to proactive forecasting. Avoiding widespread power shortages, reliability issues, or costly financial mistakes will depend on the industry’s ability to produce far more accurate long-term demand forecasts.
The growing pressure on the electric grid
The scale of new AI infrastructure is unprecedented.
One planned facility, Meta’s Hyperion campus in Louisiana, is expected to consume roughly as much electricity as the entire state of Connecticut. In Texas, one of the country’s fastest-growing markets for digital infrastructure, grid operators project that annual electricity demand will more than double between 2025 and 2035, an increase equivalent to adding another California’s worth of electricity consumption in just a decade.
System-wide, estimates from the Electric Power Research Institute (EPRI) indicate that data center energy demand will become the single largest driver of U.S. electricity consumption growth over the next five to ten years. In highly concentrated markets such as Virginia, data centers are expected to account for nearly 60% of total electricity demand by 2030.
The result is already visible in utility investment plans. U.S. electric utilities increased their five-year capital spending projections by 20% in just one year, committing to billions of dollars in new generation, transmission, and grid infrastructure.
The forecasting dilemma: blackouts or stranded assets
Building energy infrastructure takes years.
A natural gas power plant typically requires about four years to develop. New high-voltage transmission lines often take more than a decade, while nuclear power plants can require close to two decades before entering operation. As a result, today’s investment decisions must anticipate electricity demand decades into the future.
Getting those forecasts wrong creates two equally serious risks.
Underestimating demand leaves the grid without enough capacity, increasing the likelihood of reliability problems and widespread blackouts. A clear example occurred in California in August 2020, when an extreme heat wave and higher-than-expected electricity demand forced grid operators to implement rolling blackouts affecting nearly half a million customers.
Overestimating demand, however, can be just as costly. Utilities may invest billions in power plants and transmission infrastructure that ultimately remain underutilized or are never completed. Those costs are typically passed on to consumers through higher electricity rates.
read also US energy storage hits record high with 18.9 GW installed in 2025
The lesson from the 1983 financial collapse
The Bulletin of the Atomic Scientists points to one of the largest financial failures in U.S. utility history as a cautionary example of poor demand forecasting.
In July 1983, the Washington Public Power Supply System (WPPSS) defaulted on $2.25 billion in municipal bonds, then the largest municipal bond default in U.S. history.
During the 1960s and 1970s, electricity demand had been projected to grow by roughly 7% annually, prompting the utility to begin construction on five nuclear power plants. But following the energy crises of the 1970s, actual electricity demand fell far below expectations.
Only one reactor was ultimately completed. Decades later, electricity customers across the Pacific Northwest are still paying for projects that never generated a single kilowatt-hour.
Why forecasting AI-driven electricity demand is so difficult
Predicting future data center energy demand has become increasingly challenging because of several sources of uncertainty.
- Limited data availability. The United States lacks a comprehensive national inventory of data centers, and many utilities publish only limited information about electricity consumption by customer category.
- Speculative interconnection requests. Many developers submit grid connection applications in multiple regions simultaneously, later withdrawing the projects that prove slower or more expensive to connect.
- Policy uncertainty. Tax incentives, zoning rules, and temporary moratoriums, such as the one considered in New York State, can quickly redirect investment from one region to another.
- The AI growth curve. Much of today’s projected electricity demand is based on expectations about future AI deployment rather than confirmed projects.
Building a more resilient planning strategy
According to Stephen Bessasparis, avoiding the costly mistakes of the past requires three fundamental changes.
Plan for multiple futures, not one forecast. Utilities should abandon reliance on a single “most likely” demand projection and instead evaluate a wide range of scenarios, from low to high growth. This approach makes it easier to identify “least-regrets” investments, projects that remain valuable across multiple possible futures.
Share financial risk more fairly. Today, consumers often bear most of the financial risk when utilities build infrastructure for projects that never materialize. Bessasparis argues that utilities should require upfront deposits and adopt take-or-pay contracts, obligating data center developers to pay for the electrical capacity they reserve even if projects are delayed, downsized, or canceled.
Improve transparency. Utilities should publicly disclose more detailed forecasting data, while major technology companies and data center operators should publish regular reports on project development timelines and energy efficiency, following the example set by companies such as Google.
As AI continues to reshape electricity demand, the challenge facing utilities is no longer simply generating more power. It is building an electric system capable of growing at AI speed without compromising affordability, reliability, or long-term financial sustainability.