Google, Nvidia, Emerald AI and other companies have launched a coalition promoting data centers that can reduce their demand for grid power when electricity supplies are strained. The proposal could make it easier to connect new AI infrastructure to the grid—but whether it protects households from higher costs depends on when data centers are required to cut back and who pays for the infrastructure needed to serve them.

The proposal comes as Americans are increasingly concerned about the effect of data centers on their electricity bills. A new AP-NORC poll found that 84% of Americans are at least somewhat concerned about data centers’ impact on local electricity prices. Americans have also raised concerns about the water and other resources required to support the expansion of AI infrastructure.

Those concerns are growing alongside electricity demand. Lawrence Berkeley National Laboratory estimated that all U.S. data centers—not just those used for AI—accounted for about 4.4% of U.S. electricity consumption in 2023. By 2030, data centers could account for nearly 12% of U.S. electricity use.

The coalition’s proposed solution is flexibility. Instead of drawing the same amount of electricity from the grid around the clock, data centers could shift or pause some AI computing jobs, use batteries or rely on nearby power generation when the grid is strained. Coalition members argue that facilities able to reliably reduce their demand at critical times should be able to connect faster and potentially avoid some costly grid upgrades.

But flexible computing does not automatically mean households come first when power is scarce, nor does it eliminate the costs of serving large data centers. The rules governing each facility would have to determine when and how much power it must give up, while regulators still have to decide who is responsible for the remaining infrastructure costs.

That is the decision now facing regulators and utilities: whether faster grid access for flexible data centers can be structured so that the companies benefiting from that access also take responsibility for reducing demand when necessary and for the costs their projects create.

Flexible data centers are facilities that can temporarily reduce how much electricity they take from the grid when power is in high demand, such as by shifting some computing jobs to another time, using batteries or relying on nearby power generation.

There is evidence that growing data center demand is increasing some electricity costs, but the impact is not the same everywhere.

Lawrence Berkeley National Laboratory found that states with growing electricity demand have historically often seen average electricity prices fall rather than rise. Utilities already have costs for power plants, transmission lines and other parts of the electric system. When the system can accommodate more demand without requiring major new investments, those existing costs can be spread across a larger amount of electricity use.

The calculation changes when serving new data centers requires the grid to add expensive new infrastructure or secure significantly more power—and those additional costs are passed along to other customers.

PJM, the regional grid operator serving all or parts of 13 states and Washington, D.C., including Northern Virginia’s massive concentration of data centers, provides an example. Its independent market monitor says actual and forecast data center demand has been the primary driver of higher costs in the region’s capacity market.

The capacity market pays power suppliers to make sure enough electricity will be available when the grid needs it. Berkeley Lab found that higher PJM capacity prices added about 0.9 cents per kilowatt-hour to average wholesale electricity costs in 2025, with another 0.6 cents per kilowatt-hour expected from the capacity-price increase in 2026. That does not translate directly into the same increase on every household’s bill, but it shows one way rapidly growing demand can increase electricity costs.

Then there is the cost of expanding the grid itself. According to PJM’s independent market monitor, PJM approved $1.4 billion in transmission upgrades specifically related to Northern Virginia’s “Data Center Alley.” Instead of assigning those costs only to the data centers seeking the new connections, the costs were spread across retail electricity customers throughout the PJM region.

That cost-sharing system is exactly what Maryland’s Office of People’s Counsel, which represents residential utility customers, is challenging. Maryland has relatively modest projected electricity-demand growth compared with parts of the PJM region experiencing significant data center growth. But under PJM’s current rules for dividing the cost of regional transmission projects, Maryland customers can still be required to help pay for infrastructure built largely to support data centers in other states. The Office of People’s Counsel argues those costs should instead be assigned to the areas where the data centers are located or directly to the data centers themselves.

The office estimates that the transmission costs it is challenging could add about $1.6 billion to Maryland customers’ electricity bills over ten years, or approximately $345 for the average residential customer during that period. That does not mean Maryland households have already been charged an extra $345. It is an estimate of what the average household could pay over the next decade if the disputed costs remain in place. These are also different costs from the $1.4 billion in Northern Virginia upgrades discussed above.

The difference between PJM and other regions helps explain why the debate is not simply about how much electricity data centers use. What matters is whether the existing grid can handle the additional demand, what new infrastructure has to be built to serve data centers and who is required to pay for it.

If data centers could reduce their demand during the hours when the grid is under the most pressure, could that flexibility reduce the need for new infrastructure and some of the additional costs passed on to consumers?

The Flexible Data Center Coalition is proposing that data centers adjust how much electricity they take from the grid when power is in high demand. Instead of requiring enough grid capacity to meet their maximum electricity needs at all times, participating data centers could temporarily reduce their grid use when the system is strained.

There are several ways they could do that. Some AI computing jobs could be shifted to another time or paused temporarily. Data centers could also draw electricity from batteries or nearby power generation instead of the grid. The coalition has not shown that every AI workload can be paused, but workloads that are not time-sensitive could give operators more room to adjust when they use grid power.

In exchange for proving they can reliably reduce their demand when needed, coalition members want flexible data centers to receive quicker access to electricity from the grid. They could also be allowed to access more grid power than would otherwise be available to them, with the understanding that they would reduce their use during periods when the grid is strained.

The idea is that if a data center does not need its full amount of grid power during the most demanding hours, utilities may not need to build as much additional infrastructure solely to meet those peaks. That could mean fewer infrastructure costs ultimately passed on to consumers.

That could address part of the problem we saw in PJM. Reducing data center demand at the times when the grid is under the most pressure could ease some of the need for additional power capacity and infrastructure. But those potential savings have not yet been demonstrated at full scale, and the coalition’s proposal does not mean data centers would necessarily use less electricity overall.

A data center could, for example, postpone an AI computing job during a period of high electricity demand and run it later when more power is available. That changes when the electricity is needed without necessarily changing how much electricity the job ultimately uses.

Flexibility, then, could help the grid handle growing data center demand without preparing for every facility to use its maximum amount of grid power at exactly the same time. But it raises an important question: If electricity becomes scarce and both households and data centers need power, who is actually required to cut back?

Flexible data centers would not automatically get priority over households when electricity is scarce—but there is also no blanket rule guaranteeing that households always come first.

What happens depends partly on the electricity service a data center has agreed to. Under “firm” service, a data center contracts for a specified amount of grid power that is intended to be reliably available whenever it needs it. Under a flexible arrangement, the data center could instead agree that its access to grid power can be limited under certain conditions, such as when demand across the grid is especially high.

FERC is now examining how those firm and flexible arrangements should work for large electricity users such as data centers. It has directed PJM to justify its current rules or propose changes, including rules governing how much power these facilities can take from the grid and when that access can be limited. FERC has not yet approved the coalition’s proposed faster path to grid access.

A sudden increase in demand for AI services does not automatically give a data center access to more electricity from the grid. If millions of people begin using an AI service at the same time, the data center still has to operate within the amount of grid power available under its electricity-service agreement. How much power it can draw—and when it must reduce that use—depends on the terms of that agreement. Under a flexible arrangement, the data center could be required to reduce its grid use when specified conditions are met.

But that still leaves the biggest question: If the grid becomes so strained that somebody has to use less electricity, who cuts back first?

PJM’s independent market monitor says the region still needs clear rules establishing who cuts back first when the grid is strained. It has proposed that certain new data centers should add new electricity supply to help support the additional demand they bring to the grid. That could involve contracting with a power company to build or provide a new source of electricity that is added to the PJM system specifically to help meet the data center’s demand.

For certain new data centers that connect without adding that additional power supply, the market monitor has proposed requiring them to reduce their electricity use first when the grid is strained. They would cut back before customers already enrolled in programs in which they have voluntarily agreed to reduce their power use during periods of grid stress.

For now, that remains a proposal rather than a requirement.

And even if regulators establish that requirement, it only works if data centers actually follow it. That raises the next question: How would regulators know that a flexible data center really can cut its demand when the grid needs it to?

A data center promising to reduce its electricity use when the grid is strained would need to show that it can actually do so—and that the commitment can be enforced.

Even the Flexible Data Center Coalition says faster access to the grid should depend on a facility’s ability to reliably reduce its demand. Emerald AI CEO Varun Sivaram, who is leading the coalition, has said that flexibility should be verifiable and enforceable before a data center receives faster or greater access to grid power.

FERC is examining some of the same questions as it reviews how PJM connects large electricity users to the grid. It has directed PJM and transmission owners to address three basic issues:

Are the electricity forecasts accurate? Data centers need to provide reliable information about how much electricity they expect to use.
Can the facility actually cut its grid use? FERC is asking what equipment and operating requirements would allow a facility to limit the amount of electricity it takes from the grid.
Can the agreement be enforced? The terms need to establish what the data center is required to do when its grid use must be reduced.

FERC is also examining who bears the financial risk if a large project requests significant grid capacity but ultimately uses less electricity than expected—or never gets built. If utilities expand the grid based on a data center’s projected demand and that demand never materializes, other customers could otherwise be left paying for infrastructure that was built for a project that did not use it as planned.

For consumers, there are two risks. If a data center asks for more electricity than it ultimately needs, consumers could end up paying for infrastructure that was not necessary. If a data center promises to cut its electricity use when the grid is strained but cannot—or does not—the grid could be left with more demand than it can handle.

The coalition argues that this kind of flexibility could reduce grid costs, but those savings have not yet been demonstrated at full scale. And even a data center that successfully reduces its demand during critical hours may still require new power generation, transmission lines or other infrastructure.

That leaves one final cost question: Who should pay for the infrastructure that still has to be built?

Even if flexible data centers reduce some pressure on the grid, utilities may still need to build new power generation, transmission lines or other infrastructure to serve them. That leaves another question: Should those costs be paid by the data centers creating the additional demand or spread among other electricity customers?

Some of the largest technology companies have already pledged to pay those costs. In March, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed the White House’s Ratepayer Protection Pledge. The companies agreed to build, bring or buy the additional electricity supply their data centers require and pay for the power-delivery infrastructure upgrades needed to serve them. They also agreed to pay for the power and infrastructure brought online for their projects even if they ultimately use less electricity than expected.

Individual companies have made similar commitments. OpenAI says every Stargate site will pay the additional electricity-generation and grid-upgrade costs it creates. Meta says it pays the full cost of electricity used by its data centers, while Microsoft has committed to paying its way so its data centers do not increase consumers’ electricity prices.

But those commitments raise another question: If major technology companies have already agreed to cover the costs created by their data centers, why are regulators and lawmakers still debating who pays?

The reason is that corporate pledges do not by themselves determine how utilities divide infrastructure costs among their customers. As the dispute in PJM shows, the rules used to allocate transmission and other grid costs can still result in some of those costs being passed on to consumers.

Congress is now considering legislation aimed at addressing that gap. The bipartisan Ratepayer Protection Act would apply a proposed federal standard to certain large electricity users, including qualifying data centers with peak demand of at least 100 megawatts at a single site or campus.

Under the proposed standard, the rates or agreements used to provide electricity to those large customers would be designed to recover the full additional cost of power generation, transmission and distribution upgrades necessary to serve them. The data center would also have to provide financial assurances to cover those costs before the utility undertakes the upgrades. Those protections are intended to keep other customers from being left with the bill if a project ultimately uses less electricity than expected or is never completed.

Because utility rates are generally regulated at the state level, the bill would require state utility regulators and certain utilities to consider the standard and decide whether to adopt it, rather than automatically imposing the same rate structure nationwide. States that have already adopted comparable protections could be exempt from that process.

The bill has not become law. The House debated the amended Ratepayer Protection Act on Sept. 15 but postponed the recorded vote.

Whether through voluntary commitments or new regulations, the unresolved question is how to make sure the costs created by data center growth stay with the companies creating them rather than being passed on to other electricity customers.

The Flexible Data Center Coalition’s proposal could change how new data centers gain access to large amounts of electricity. Instead of waiting for utilities to build enough infrastructure to guarantee their maximum electricity needs at all times, flexible data centers could potentially begin receiving the power they need sooner by agreeing to use less grid electricity during periods when the system is strained.

For data center developers, that could shorten one of the biggest delays to building new AI infrastructure: waiting for enough grid capacity to become available. For utilities, it could reduce the need to build the system around the assumption that every data center will use its maximum amount of electricity at the same time. And for consumers, that could mean less pressure to pay for infrastructure needed only to meet those peak demands.

Whether consumers actually benefit depends on the rules attached to that flexibility. A flexible connection only reduces risk for other customers if the data center can reliably cut its grid use when required and if the costs of the additional electricity supply and infrastructure it still needs remain with the companies creating that demand instead of consumers.

That makes the decisions now facing regulators more consequential than simply whether data centers should receive electricity faster. They are establishing the conditions under which the next wave of AI infrastructure can draw power from a grid that was not built for this level of demand.

The question now is what happens when there is not enough electricity for everyone who wants it. Data center construction is moving quickly; the rules deciding who cuts back and who pays are not.

Q: What is a flexible data center?
A: A flexible data center is a facility that can temporarily reduce how much electricity it takes from the grid when power is in high demand. It could do this by shifting some computing jobs to another time, using batteries or relying on nearby power generation. The Flexible Data Center Coalition argues that facilities able to reliably reduce their grid demand should be able to gain access to electricity sooner.

Q: Why do AI data centers want faster access to grid power?
A: New data centers can require enormous amounts of electricity, and utilities may need to build new power generation, transmission lines or other infrastructure before they can reliably serve that demand. Flexible data centers could potentially begin receiving electricity sooner by agreeing that they will reduce their grid use during periods when the system is strained rather than requiring their maximum amount of grid power to be available at all times.

Q: Are data centers causing electricity prices to rise?
A: Data centers can increase electricity costs when their additional demand requires new power supplies or grid infrastructure and those costs are passed on to other customers, but the effect varies by region. In PJM, the region's independent market monitor has identified actual and forecast data center demand as the primary driver of higher capacity-market costs. Data-center-related transmission costs have also been allocated to electricity customers outside the areas where some of the new demand is located.

Q: Do households get priority over AI data centers when electricity is scarce?
A: Households are not automatically guaranteed priority over AI data centers when electricity is scarce. How much electricity a data center can take from the grid and when it must reduce that use depend on its electricity-service agreement and applicable grid rules. PJM's independent market monitor has proposed requiring certain new data centers to cut their electricity use first during periods of grid stress if they connect without adding new electricity supply, but that is not currently a requirement.

Q: Does flexible computing mean AI data centers will use less electricity?
A: Not necessarily. Flexible computing can change when a data center uses grid electricity without reducing how much electricity it ultimately consumes. An AI computing job could be postponed when the grid is strained and completed later when more electricity is available. Flexibility is intended primarily to reduce demand during critical periods rather than guarantee lower overall electricity consumption.

Q: How can data centers prove they will reduce electricity use when the grid is strained?
A: Flexible data centers would need operating capabilities and enforceable agreements that allow their grid use to be reduced when required. FERC is examining whether electricity-demand forecasts are accurate, what equipment and operating requirements would allow large facilities to limit their grid use, and how those commitments should be enforced.

Q: Who pays for the electricity infrastructure required by AI data centers?
A: Who ultimately pays depends on the utility and regulatory rules governing how infrastructure costs are allocated. Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI have pledged to cover additional electricity supply and infrastructure required by their data centers, but corporate commitments do not themselves determine how utilities divide costs among customers. Regulators and lawmakers are considering additional protections intended to prevent costs created by large new electricity users from being passed on to other customers.

Q: What is the Ratepayer Protection Act?
A: The Ratepayer Protection Act is proposed federal legislation intended to protect other electricity customers from costs created by certain large new electricity users, including qualifying data centers with peak demand of at least 100 megawatts at a single site or campus. The proposed standard would seek to recover the full additional cost of necessary power generation, transmission and distribution upgrades from the large customer and require financial assurances for those costs. The bill has not become law.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.