Data Centers and Rising Electricity Costs

For over two years, data centers have been headline news within the energy industry. PJM’s July 2024 capacity auction clearing at $270 per megawatt day, up from $29 per megawatt day in the preceding auction, put data center impacts on energy prices squarely into public consciousness. The shock of that ninefold increase remained abstract to most consumers until the summer of 2025, when those costs began appearing on customer bills.

Since then, public sentiment has worsened considerably — while 28 percent of over 4,000 registered voters surveyed in August 2025 by Embold Research blamed data centers “a lot” for rising electricity prices, nine months later that figure exceeded 50 percent, with data centers topping the list of proposed culprits. From Forbes — “AI Data Centers Are Driving Up Electric Bills. Who Pays?” to Consumer Reports — “Inside the Energy Affordability Crisis: We Will All Be Affected by This,” media outlets reflect and amplify consumer anxiety.

Meanwhile, trade industry sources advance the opposing narrative, as illustrated by a NetChoice headline: “New Study Confirms What We’ve Been Saying: Data Centers are Lowering Your Electric Bill.” E3’s May 2026 report on electricity rate drivers, funded and reviewed prior to publication by the Data Center Coalition, found no evidence that data centers were historically subsidized and that states with the largest load increases saw the smallest rate increases.

Neither side in this debate has an exclusive claim on the truth. Clarum Advisors talked with an array of industry leaders — utilities, data centers, developers, and vendors — to tease apart fact from fiction. In this article, we separate whether data center costs reach other consumers from the question of how, addressing both to inform solutions.

Data Center Cost Taxonomy

Any meaningful discussion should be grounded in the mechanisms through which data centers affect electricity costs. There are three distinct routes by which those costs can reach other customers:

Molly Podolefsky: A significant number of large load tariffs today include minimum bills, collateral requirements, coverage of study costs, and contract terms extending ten or more years.

Tier One, Cost of Service. Directly assignable costs to serve data centers — dedicated substations, feeders, transformers — are recovered through tariffs. Because they are discrete and attributable, tariff design can largely ensure they are not borne by other ratepayers. This can also extend to definable network transmission costs.

Tier Two, Shared System Costs. Some transmission facilities identified through utility or regional wholesale-market planning are funded through broader cost-allocation formulas because they provide reliability, economic, or systemwide benefits. As a result, even when a data center contributes to the need for an upgrade, some costs may be spread across a transmission zone, utility system, or broader group of beneficiaries. Because these facilities serve multiple customers and purposes, determining the share attributable to an individual data center is difficult and often contested.

Tier Three, Market Prices. The costs of various inputs to electricity generation and pro­vision — wholesale electricity, capacity auctions, transformers, turbines — are set by market clearing prices, which data center demand drives up. These prices reach consumer bills, but they sit outside the traditional utility rate setting and revenue recovery process and are therefore harder to address.

Where data center-related costs sit in this taxonomy determines whether, and to what extent, they are ultimately borne by other customers. Data centers can also lower energy costs for other consumers, a counterweight we return to below.

Tier One: Examining the Cost-Shifting Claim

Lon Huber: The industry has largely ring-fenced the costs through the contractual side. What remains an open question is network transmission when upgrades are not individually triggered. That’s harder, and I think you’ll continue to see refinement there.

Some of the most vocal opposition to data center expansion hinges on a claim of cost shifting, that data centers pay less than their allocated cost of service, leaving other ratepayers to cover the shortfall. Three distinct mechanisms stand between new large loads and that outcome.

First, the utility conducts a cost-of-service study for each rate class — coincidence, load factor, demand, and energy consumption — that results in a revenue requirement determining rates.

Second, it completes interconnection, facility, and stranded-cost studies that itemize the new facilities required — substations, transformers, radial lines — determining how much the customer pays for construction and to cover risk. Contract terms — minimum billing demands, take-or-pay provisions, collateral, exit provisions — informed by years of observation and best practice development, are the third and final layer mitigating the potential for cost shifting.

Carefully constructed tariffs and contracts protect ratepayers from cross-subsidization. Xcel Chief Planning Officer Lon Huber observes, “Our contracts are solid and specific about what we charge and how we allocate costs and benefits. That part is now straightforward.”

Allan Schurr: If you don’t win this load, if you think your rates are high now, they’ll be higher still when you have no offsetting volume over which to amortize those investments. Utilities need to fight hard to win these customers, and to serve them below average cost. That drives rates down for everyone.

Because directly assignable costs are measurable, Tier One is the closest to solved. These costs are well documented and well understood. The real challenge is standardization and adoption of best practices, not building a better mousetrap.

Allan Schurr, Chief Commercial Officer for ERock concludes, “Almost every data center today pays a data center tariff designed to cover its cost to serve. The idea that rates have risen because data centers aren’t paying their way is, I think, old news.”

DELTa, a database maintained by SEPA and the NC Clean Energy Technology Center, suggests the gap between theory and practice is closing fast; 77 large load tariff filings had been logged as of March 2026, against 14 across the whole of 2018 through 2024.

Georgia Power’s proactive handling of data center engagement and contracting to protect customers from cost shifting sets a high bar. The company reassigned three of its senior leaders to manage data center engagement full time, a testament to the scale of the challenge and the company’s commitment to proactively addressing data center growth in Georgia.

Larry Bekkedahl: I told them I’d move anybody to the top of the queue if they could give me 20% flexibility. The immediate answer was a stiff arm; ‘No, we can’t.’ So, we said, ‘Fine, then you wait until the transmission line gets built.’ And they started coming back.

Erica McGill, formerly Vice President and Deputy General Counsel at Georgia Power, now serves as the regulatory and legal arm of the three-member executive team that oversees data center engagement, and her objective is not cost neutrality, it is downward rate pressure.

McGill identifies regulatory and business factors as critical to avoid cost-shifting, “We are proud and protective of our state-regulated, vertically integrated market structure. In a period of unprecedented growth, this economic model — where a single company is responsible for making, moving, and selling electricity — provides Southern Company with the visibility and accountability we need in order to grow responsibly and to do it in a way that does not shift costs to other customers.”

Contracting tools employed by Georgia Power include robust collateral, long-term obligations, stringent termination requirements, and minimum billing so that, “No matter what happens, we have a way to recover the costs.” A significant number of large load tariffs today include minimum bills, collateral requirements, coverage of study costs, and contract terms extending ten or more years.

To the extent that transmission costs can be attributed and allocated directly to an individual data center, impacts to customers in other rate classes can be avoided altogether. Customer-funded builds solve the problem of cost-shifting decisively but as we will discuss, direct allocation is not always feasible. We know how to solve this tier, and the remaining hurdle is implementation.

Tier Two: Unpacking Shared System Costs

“The industry has largely ring-fenced the costs through the contractual side. What remains an open question is network transmission when upgrades are not individually triggered. That’s harder, and I think you’ll continue to see refinement there.” Lon Huber’s observation moves us to the second and harder category — upgrades to network transmission and generation — facilities that are naturally shared.

New turbines, transformers, and lines cost far more than the depreciated book value of the fleet they join. Where only a portion of that cost is allocated to data centers, every customer in the class sees a rate reflecting the new blended average of older, cheaper assets and newer, costlier ones.

When upgrades cannot be cleanly mapped to a single customer, causation can be genuinely ambiguous. If an upgrade was already planned, or would have been needed eventually, then the data center accelerated that cost but did not cause it. “There has always been a cost-allocation question around local transmission impacts. What’s been overlooked historically are the system-wide impacts,” shares Scott Burger, Firma Power Co-Founder and President.

There is no universal solution, but several approaches address different parts of the problem. The first addresses allocation directly. A 2025 E3 paper proposes a straightforward rule: set data center tariffs at or above the marginal cost of new infrastructure, so other customers do not subsidize the new load.

NV Energy’s clean transition tariff provides an illustrative example: Google pays the difference between the cost of Fervo’s 115-megawatt Corsac Station geothermal project and the lower-cost resource the utility would otherwise have selected under least-cost planning, while Google also receives bill credits for energy and capacity delivered to the system by the new plant.

Other approaches allow utilities to avoid upgrades in the first place. Onsite generation grows more attractive as interconnection queues lengthen and upgrades add time and expense.

ERock’s Allan Schurr points to onsite generation as a bridge to grid connection, one that can also leave a utility with a flexible capacity asset once that connection is complete. Where the customer has already funded the asset, that resource can help the utility avoid duplicating investment elsewhere. “You may not need that very expensive transmission upgrade, or an incremental resource adequacy project.”

Alternately, Larry Bekkedahl, member of The Western Transmission Consortium and PJM Boards, and former SVP of Strategy and Advanced Energy Delivery at Portland General Electric, suggests a modular, distributed design, which is a network of smaller data centers sized to fit local interconnection and transmission constraints, as an alternative to forcing system upgrades. “Inference can run in smaller units that are chained together, with ten-to-20-megawatt units at each substation, built out as a network.”

Lon Cunninghis, EVP Strategic Accounts for DG Matrix says, “modular data centers and power supplies allow developers to deliver speed to power at lower cost.”

Bekkedahl also recommends load flexibility as an alternative to upgrades, suggesting data centers may be more willing and able to provide flexibility than utilities assume. “I told them I’d move anybody to the top of the queue if they could give me 20 percent flexibility. The immediate answer was a stiff arm; ‘No, we can’t.’ So, we said, ‘Fine, then you wait until the transmission line gets built.’ And they started coming back.” In other words, through patience, persistence, and cooperation, utilities may be able to negotiate contracts that connect data centers faster while avoiding costly buildout.

A final approach treats the problem as systemic. Utility planning processes are slow — often six months or more — and typically siloed and sequential, a series of handoffs between utilities rather than a collaborative effort.

At the core is information sharing; utilities are often unwilling or unable to give one another the proprietary data that regional coordination and large load planning require. The result is latency and lost opportunity for joint optimization.

Larry Bekkedahl would consolidate the process at the regional level, with entities that can hold utilities’ sensitive data in confidence. “I believe the regional entities under NERC — WECC in the West for example — ought to take this on. They should be able to help run these studies quickly and bring the data together in one place.”

Tier Three: Addressing the Invisible Hand

Market-wide price effects lie outside the reach of utility tariffs. Inelastic data center demand meeting short-run inelastic supply pushes up market-clearing prices for everyone. Directly, data center consumption intensifies competition in wholesale energy and capacity markets while new generation takes years to arrive.

Scott Burger observes, “It’s hard to look at it and say it isn’t a real issue. In PJM, a lot of new load was added to the grid and not much new generation. That wasn’t entirely the data center industry’s fault, there’s load growth outside of data centers. But fundamentally, demand outstripped new supply, and that drove up rates.”

Data center loads also threaten reliability, a point driven home by Larry Bekkedahl. “As we’ve added these data centers, we’ve taken the spare room out of the system. We don’t have the safety margins any longer.”

Indirectly, data centers’ massive appetites for energy create scarcity in the market for inputs to energy production. Lon Cunninghis shares, “Prices are going up because of supply chain issues. Take transformers; depending on the size and type you need, it takes years to get one. The demand for this equipment and legacy transformers’ scarcity are happening all at once.” Those higher input prices, in turn, lead to increases in the downstream electricity prices paid by consumers.

Allan Schurr summarizes, “If the unit cost of a transformer rises because global demand has risen, then yes, that affects energy prices. Data centers will pay their own way for their own transformers, but the rest of the rate base has to buy transformers too, and those rates can go up.” In other words, a data center can pay 100 percent of its cost of service and still raise everyone’s energy bills via multiple channels.

Two channels exist to address market-induced price increases: unblock the supply response and add supply faster than demand grows. Addition, not allocation, is the lever here.

Lon Cunninghis sees that response as happening in real time, though it has not yet caught up to demand. “All of a sudden, every one of the partners we work with is building new factories, expanding plants, and trying to meet demand but no one was prepared for how fast demand is coming at them.” Onshoring and supportive policy have a role, but the generation response also depends on clearing slow regulatory, interconnection, and permitting processes, and the years-long queues they produce.

Adding supply faster than demand growth will require utilities, regulators, and data centers to work collaboratively toward creative solutions. Allan Schurr states this plainly. “You can’t just be neutral on the grid. You have to be an asset for the grid.”

Larry Bekkedahl would have data centers add more generation to the grid than their own load requires. “Imagine you’re a data center, and you say: ‘I’m not only going to build generation to supply this facility, I’m going to add another 200 megawatts to help the system. What am I doing that’s above and beyond — that helps society? I can still make money on those 200 megawatts, but I’ve done more than the bare minimum.’ I think that has been the missing element here.”

In this sense, bring-your-own-generation, as in the case of NV Energy, can serve that broader purpose. Scott Burger predicts, “People are going to want to point at resources on the grid and say, ‘We are bringing new generation to support our data center. We’re not driving up energy prices here.’”

A Positive Counterweight With Guardrails

Data center load also offers a legitimate counterweight to the largely negative account so far. A utility’s fixed costs are recovered through rates that are largely volumetric as bills rise with energy use. So, where a large load pays above its incremental cost of service, adding it shrinks or “dilutes” the residual left for everyone else to cover.

Allan Schurr drives this point home. “If you don’t win this load, if you think your rates are high now, they’ll be higher still when you have no offsetting volume over which to amortize those investments. Utilities need to fight hard to win these customers, and to serve them below average cost. That drives rates down for everyone.”

The benefit rests on utilities serving data centers efficiently enough that the rate paid sits above the incremental cost caused. Stated plainly: the counterweight argument holds only if the Tier One tariff is well designed.

Even then a timing mismatch persists, cost effects arrive within a year or two, while dilution accrues across a contract that often runs a decade or more. As a result, a data center contract can genuinely yield positive net present value for ratepayers, and yet still raise bills in the near-term. With guardrails — thoughtful tariff construction and amortization treatment — data centers can drive meaningful cost reductions.

An EPRI working paper analyzing FERC data for the decade spanning 2015 through 2024 found just this; average retail electricity prices fell 3.5 percent for every doubling of data center capacity, about six percent at the state level. The authors caveat it; for these benefits to hold, the demand must materialize as predicted. Hence the need to insulate customers from the risk of short duration, transitory large loads.

Georgia Power provides an architecture for other utilities to follow. According to McGill, it is not enough to ensure data centers merely cover their own costs as they can and should be contributing positively to energy affordability for all customers. “It’s not just that we’re recovering the cost to serve data centers over the life of the contract. We are charging them a premium so that we can put downward pressure on rates.”

This premium, in turn, has enabled Georgia Power to reduce the typical residential customer’s energy bill by $180 per year, roughly $15 a month between 2029 and 2031, in addition to a previously approved three-year base rate freeze for over two million customers.

In terms of willingness to pay, the utility has found data centers agreeable partners who are committed to making a positive difference in Georgia’s communities and, critically, paying their full share of the costs to serve them. “I have zero concern that they are going to say, ‘Oh that’s too costly. I’m not interested in doing business here.’ They remain focused on speed to power, not on whether they’re paying a premium to help address affordability for residential customers — that’s not the issue.”

Acknowledging Externalities

Externalities sit outside the price-versus-load-growth equation by definition but warrant discussion because the two are often conflated. They are the effects of producing and consuming a product — negative and positive — that prices do not capture: producers and consumers create them without paying for them.

On the positive side of the externalities ledger, Lon Cunninghis points to a “gold rush around skilled labor,” and the national security implications of advancing AI. “We’re in a race for superintelligence with our competitors, and it’s critically important that we win that race. If we’re curtailed — no pun intended — that is at risk.”

Data centers present significant local and regional economic development opportunities, as well. Quoting McGill, “When you see the investment and jobs statistics, you realize that these sites are creating entire ecosystems of economic development, from the people who are making the roads to the construction workers onsite.”

Conversely, data centers have been shown to cause various negative externalities, such as reducing the water supply in drought-prone regions and contributing to noise pollution in cities. Perhaps the most relevant negative externality to this discussion, and the most within reach, is energy system reliability.

Data center-induced load growth poses significant challenges in terms of resource adequacy. Larry Bekkedahl observes, “We are running on the margin. There isn’t much capacity in anybody’s reserve margins. When the peak hits — as it did this summer — West Coast, East Coast, everybody’s running right at the edge of their systems.”

Most externalities fall largely outside the energy cost discussion. Canonical economic remedies — taxes, and subsidies where the externality is positive — belong in the wider discussion of data center effects, but not in the narrower one about energy costs borne by ratepayers. Reliability is the partial exception, as capacity markets and resource adequacy requirements exist precisely to put a price on it, making it accessible to regulators.

The Bad News and the Good News

First, the bad news: there is no one-size-fits-all solution for how data centers impact electricity prices, because there isn’t a single problem to solve; there are many. Data centers reach consumers through multiple channels, each with its own mechanism and, more importantly, its own set of potential solution owners. Who solves what, is the question. The good news is that the answer rests on no single set of shoulders.

Direct cost shifting is solvable; comprehensive tariffs and sound contract terms ensure data centers pay their way. By ensuring data center rates cover marginal cost, commissions and utilities can solve Tier One, reinforcing volumetric dilution of fixed costs to the benefit of all ratepayers, and begin to address Tier Two shared system costs.

Data centers and regional entities each own part of the solution to shared transmission and generation costs. Through onsite generation, load flexibility, and modularity, data centers can help avoid new infrastructure upgrades in the first place, while reliability organizations can improve upgrade efficiency through integrated regional planning and resource coordination.

Universal price increases determined through supply and demand are beyond the reach of tariffs; yet regulators, utilities, and data centers can relieve pressure. By clearing interconnection delays and other barriers to supply, regulators and utilities can bring market-clearing prices down; hyperscalers can position themselves as grid assets, bringing on generation faster than they add demand.

The energy system is under strain. Ratepayers are feeling the stress of rate increases at a time when the system itself is struggling to keep up with demand, putting reliability at risk and leaving customers exposed.

Data centers, utilities, regulators, and system operators each have a role to play in changing that dynamic. The illusion of a single solution distracts from the real task of digging in and working together to claw back energy affordability using all the tools available to us.