A new legal framework proposes breaking from the utility industry’s sacred duty to serve all customers at all times—and the changes could reshape how we power the digital economy
The math is stark: A single hyperscale data center can consume as much electricity as 80,000 American homes. With artificial intelligence driving an unprecedented construction boom, these power-hungry facilities are forcing utilities to confront a question that strikes at the heart of how America has managed electricity for over a century.
Should the grid’s foundational promise—that utilities must serve all customers, all the time—survive the age of AI?
A groundbreaking new analysis from legal scholars at the University of Michigan and UC Law San Francisco argues it shouldn’t, at least not in its current form. Their proposed alternative, called “demand-side connect-and-manage,” would allow major new electricity users like data centers to connect to the grid with the understanding they could be curtailed during shortages—a radical departure from traditional utility regulation.
The Old Rules Meet New Realities
For decades, U.S. electricity demand remained essentially flat, growing just 0.1 percent annually between 2005 and 2020. Utilities built their business models around predictable load growth and the legal obligation to maintain sufficient generating capacity to meet peak demand without curtailments.
That world is ending fast. Recent projections suggest electricity demand could jump 25 percent by 2030, with data centers accounting for 12 percent of nationwide consumption—up from just 4 percent in 2023. The surge is so rapid that some utilities have seen their interconnection queues balloon from a few gigawatts to over 100 gigawatts in just two years.
“We have an obligation to serve,” one utility executive explained to researchers. “If they have the land and the ability to build and ramp, they can do that, and we have to find the assets to serve them.”
That obligation is creating dangerous imbalances. Utilities eager to attract major tech customers are proposing massive buildouts of new power plants—often natural gas facilities that can be constructed faster than renewables. But data center development is notoriously speculative, with industry insiders noting that even seasoned companies like Microsoft and Google propose “several times more projects than they’re likely to need.”
The result: Either consumers bear the cost of expensive infrastructure that may never be fully utilized, or the grid struggles to keep pace with genuine demand growth.
Lessons From Water and Gas
The Michigan-UCSF analysis draws on two other resource management systems that explicitly plan for scarcity rather than trying to eliminate it.
In natural gas markets, “interruptible” contracts have long allowed pipeline companies to serve additional customers at lower rates with the understanding that service could be curtailed during shortages. When natural gas shortages struck in the 1970s, this system—combined with secondary markets where unused capacity could be traded—proved more resilient than rigid supply-demand matching.
Western water law offers an even starker example. Under the “prior appropriation” doctrine, water rights are allocated by seniority, with newer users explicitly accepting that they may receive nothing during dry years. This system has not only functioned for over a century but has spurred significant innovation, particularly among holders of junior rights who face the greatest curtailment risk.
The researchers point to Southern California as a prime example. Urban water agencies there, operating under relatively insecure water contracts, have become global leaders in water recycling, desalination, and demand management—innovations largely absent in Northern California, where more secure senior water rights reduce pressure to innovate.
The AI Opportunity
Data centers may actually represent ideal candidates for this new approach. Unlike residential customers, they’re operated by sophisticated companies with resources to manage supply risks. Many already maintain backup generation and can shift computational loads across multiple facilities.
More importantly, tech companies’ willingness to pay premium rates for reliable power could accelerate the deployment of emerging clean energy technologies. Google and Meta are already contracting with geothermal developers, while Amazon has struck deals to help finance new nuclear capacity.
Research suggests that allowing large loads to connect with limited curtailment risk could dramatically speed interconnections. A 2025 Duke University study found that adding a new load equivalent to 10 percent of current U.S. electricity demand would result in curtailment rates of just 0.25 percent—manageable for sophisticated users but far preferable to waiting years for traditional grid connections.
Political and Practical Hurdles
The transition won’t be smooth. Both natural gas and water allocation systems have faced decades of litigation and political conflict, particularly when powerful users face curtailment. Data centers backed by some of the world’s largest corporations are unlikely to accept service interruptions without a fight.
There are also technical challenges. Determining curtailment priorities based on contract terms and connection dates may seem straightforward, but both gas and water systems have struggled to implement even simple allocation rules consistently.
The researchers acknowledge these difficulties while arguing that some form of demand-side management is inevitable. “At least some periods of electricity scarcity are probably unavoidable,” they conclude, regardless of how aggressively the U.S. pursues new generation.
State Experiments Begin
Several states are already testing elements of this approach. Texas legislation now allows mandatory curtailment for electric loads over 75 megawatts during grid emergencies. Minnesota has prohibited utilities from passing data center infrastructure costs through to other consumers. Ohio regulators require large data centers to pay for at least 85 percent of their contracted capacity, whether they use it or not.
These early experiments suggest growing recognition that the traditional utility compact—unlimited service in exchange for cost-of-service regulation—may not survive the collision between AI ambitions and grid realities.
The question is no longer whether electricity allocation will change, but how quickly and how equitably those changes will unfold. As the researchers note, the alternative to managed scarcity may not be abundance, but rather a chaotic scramble that leaves both consumers and the climate transition worse off.
In the end, teaching the grid to say “no”—or at least “not now”—may be the key to keeping the lights on for everyone else.
