Thanks to the exponential rise of AI, the rapid construction of large, gigawatt-gobbling data centers across the country has stoked concerns about how the power grid will hold up under this additional strain. 

Cornell researchers have proposed a novel way to manage the increased demand for electricity: a credit market that would essentially allow data centers facing required cutbacks to pay flexible energy users to take on their obligations when the grid is strained. 

The findings were published Sept. 22 in Applied Energy. The lead author is doctoral student Richard Mahuze. 

“The question is how to connect a data center’s obligation to cut back with flexibility elsewhere on the grid. Our proposed market would let eligible users trade that obligation while still delivering the reduction the grid needs,” said Max Zhang, the Irving Porter Church Professor of Engineering in the Cornell Duffield College of Engineering, who oversaw the project. 

The researchers got their start a year ago by talking with industry representatives – including David Nanus ’96, president of LS Power Equity Advisors – about ways to integrate data centers with energy grids, both in terms of possible incentives as well as network constraints. 

Their timing was fortuitous. In June, the Federal Energy Regulatory Commission (FERC) directed six regional grid operators to “justify or reform the rules that govern how data centers, manufacturing facilities and other large energy users connect to the electric grid.”

Meanwhile, proposed data centers can wait years to connect to the grid. To shorten the wait, utilities and grid operators are increasingly offering “non-firm” connections, which let a facility come online sooner if it agrees to cut back its power use when the grid is strained. The rules vary by region. 

Those cutbacks can be costly, Zhang said. A data center that curtails its usage may lose revenue or owe penalties to customers under service level agreements (SLAs) that promise nearly nonstop service. 

“Once a data center is connected to the grid, its agreement to cut back still has to be honored,” Zhang said. “Enforcing that commitment can become very tricky.” 

At the same time, there simply hasn’t been enough financial incentive to tap the grid’s potential for demand flexibility, said Zhang, whose lab has been working on similar flexibility in buildings. Energy users may be able to shift or reduce their electricity use, but they need a financial reason to do so, much like an airline passenger considering whether to give up a seat on an overbooked flight. 

“Passengers place different values on keeping their seats, so an airline can raise its offer until enough people volunteer,” Zhang said. “The proposed market mechanism could similarly compensate eligible energy users who can provide demand flexibility at a price they are willing to accept.” 

The researchers’ solution was to design the Curtailment Credit Market (CCM). It would work like this: As transmission lines become overwhelmed, the grid operator would issue a signal and data centers could enter the CCM and purchase a curtailment credit from a network of smaller, more flexible users, who would then reduce their energy consumption and profit from the exchange. The market would only approve trades that the grid’s power lines can handle. 

“A data center might be required to cut 100 megawatts but find that interruption especially costly because of its service commitments,” Zhang said. “It could purchase curtailment credits through the CCM from a more flexible energy user willing to take on part of that cutback, compensating that user for the flexibility it provides.” 

The sellers could range from factories and large businesses to virtual power plants (VPPs) and other aggregators, which bundle distributed generation, energy storage, electric vehicles and building control into a single resource. 

“These sellers don’t have to cut back so far that anyone loses power,” Mahuze said. “With this market, we’re advocating for a more distributed solution. I think today there’s no shortage of energy storage or solar panels, for example. So if we put more of those into the grid, this can be handled in a more distributed manner.” 

Sending required reductions to users who can most easily spare them would result in fuller use of existing power lines, which could reduce the need for costly upgrades and help keep energy bills down, according to the researchers. There could also be an environmental benefit, because data centers could pay others to cut back instead of running backup diesel generators, leading to cleaner air. 

In simulations on three test grids, including a simplified model of New York’s grid, the researchers found that trading could avoid a third or more of the economic damage that resulted from power reductions. The market could also generate new income for flexible energy users, who can voluntarily take on additional power reductions in exchange for payments from data centers that face higher costs from cutting back themselves. 

With support from the Cornell Atkinson Center for Sustainability, the researchers are now working with the Environmental Defense Fund to create a viable market-based mechanism that would ensure a fair and efficient compensation process through CCM. 

“Our main argument is that the market can find a lower-cost way to meet a required cutback,” Zhang said. “In the current paper we show it’s feasible, but we’re currently working on how to make it implementable.” 

Co-authors include doctoral student Ali Amadeh and Charlotte Gressel. 

The research was supported by the David and Patricia Atkinson Foundation and the Indonesia Endowment Fund for Education (LPDP).