Ask whether data centers raise your electricity bill, and the argument you will hear is mostly about fairness: if a data center pays for what it uses, everyone else is left whole. This premise of accountability underlies every large-load tariff, the wave of bills in Congress, and the ratepayer pledge signed at the White House. It is right for the costs a data center can be billed for. Some of what a data center costs can be traced to it and charged to it, but a data center also moves prices, and those costs appear on every customer's bill at the same time, including customers with no data center in their state. As data center load climbs from about 4.5% of US power toward a projected 9 to 17% by 2030, the gap between what can be billed and what cannot only widens.
Regulators are starting to assign the costs that have an owner, such as transmission and capacity. The costs without an owner are larger, and current ratepayer protections do not reach them. A higher price is not a cost on anyone's account that can be reassigned. Every customer pays it at once.
Transmission, distribution and capacity
At the utility level, the focus is transmission and distribution. A hyperscale campus needs substations, feeders and sometimes a new transmission tap, and the fight has been over whether those costs sit in general rates or with the customer who caused them. Virginia's State Corporation Commission approved a large-load tariff for Dominion that puts customers above 25 MW on 14-year minimum contracts, obligates them for 85% of contracted transmission demand, and requires collateral of $1.5 million per MW. Oregon's POWER Act directed the PUC to create a separate rate class for facilities above 20 MW. New York opened a proceeding in February on how large loads connect and what share of the resulting costs they carry. Wisconsin has proposed direct assignment of distribution, transmission and generation costs to large loads.
At the market level, the focus is capacity prices. PJM's 2027/2028 Base Residual Auction cleared at the maximum price for the second consecutive year while failing to procure enough capacity to meet its own one-event-in-ten-years standard: a 14.8% reserve margin against a 20% requirement. PJM's market monitor found that data centers accounted for $6.5 billion, or 40%, of the auction's $16.4 billion in cost, roughly $6.2 billion of it attributable to facilities that have not been built. The monitor has been direct that large loads are the primary reason for high capacity prices.
The remedies
What both fronts have in common is a workable remedy. The instruments are different but the logic is the same wherever we can identify the cost: assign it to the load that caused it.
- Large-load tariffs and separate rate classes put minimum-term commitments, minimum-demand obligations and collateral behind an interconnection, so the utility is not left with a stranded asset and other ratepayers are not left with the bill. Dozens of these have moved through the states, tracked in the DELTa database and catalogued in Sierra Club's 2026 review and ArentFox Schiff's regulatory survey.
- Direct assignment and contribution-in-aid-of-construction transfer the known/traceable capital costs out of shared rates entirely.
- Capacity market design addresses the capacity price. PJM has proposed to make new large loads bring their own power, keeping them out of the capacity procurement so they no longer set the price everyone else pays. Price caps and floors, load forecasting rules, and faster interconnection of new supply are further levers, each set by a market operator through a tariff filing.
Whether the commitments behind all this prove enforceable is a fair question. But the cost is understood, the remedy is understood, and there is a proceeding in which to argue about it.
Three costs that cannot be assigned
The costs below are more difficult to assign directly to these new large loads. They are broader and indirect, and they are larger. There are three: equipment inflation, natural gas prices, and the electricity prices that follow from gas.
The first, higher equipment prices, is dealt with by Andy Lubershane, who gives a thorough and compelling account in Steel For Fuel of how equipment price inflation for key new infrastructure will elevate the costs of electricity for consumers. We address the next two.
Gas bills
Serving that load with gas means producing more gas. Increased demand for gas, especially rapid increases, can put upward pressure on prices, because drilling and pipelines take years to catch up. Every gas buyer pays that price, not just the new one.
Our model represents this. Oil and gas production is endogenous, on a framework derived from the EIA's Hydrocarbon Supply Module, so the gas price responds to how much is demanded and how fast it arrives. Pipelines and storage are represented physically, and gas is balanced on sample days, not annually.
Data center demand growth is taken from EPRI's Powering Intelligence. Across four trajectories, the national average gas price in 2030 rises from $4.09/MMBtu with no data center growth to $4.33 in the low case, $4.49 in the medium case and $4.74 in the high case — a premium of up to $0.65/MMBtu, about 16%.
Loading map…
The premium also appears in regions where we do not expect significant data center demand by 2030. Zonal premiums in the high case run from $0.61 to $0.80/MMBtu across the lower 48. A transmission upgrade is local and a capacity price is regional. A gas premium is close to national.
Customers pay this directly, through delivered gas rates. Gas bills rise $14.4 billion above the no-growth case in 2030, concentrated in Texas, the Gulf South and the gas-heavy Midwest, which is a function of their use profile, not the relative price impacts in those regions. Even customers with no data center in their state pay part of it.
Loading map…
These premiums assume the pipeline expansions planned for 2030 are built on schedule. Pipelines have been delayed before, by siting, permitting and cost-allocation disputes, and any delay makes these premiums the optimistic case for the regions affected.
Electricity bills
Elevated gas prices show up on electric bills through two mechanisms.
In RTO and ISO regions, through the market price. In most hours gas sets the marginal price of electricity, and the wholesale energy price is approximately the market heat rate times the marginal price of gas. A gas premium therefore raises the power price for every megawatt-hour served, not just the increment the data centers consume. At a 7,500 Btu/kWh market heat rate, a $0.65/MMBtu premium is roughly $4.88/MWh on wholesale power alone.
In non-RTO/ISO regions, as a fuel-cost pass-through. Vertically integrated utilities across the Southeast, much of the West and elsewhere have no market price to clear. They recover fuel and purchased-power costs from customers through fuel adjustment clauses, so a higher gas price reaches retail bills directly.
In both, higher load raises grid utilization, which brings less efficient plants into more hours. That raises the heat rate at the margin and compounds the gas price increase.
Applied to the electricity that everyone other than the data centers was already buying, the higher price adds $31.7 billion to electricity bills in 2030. With the $14.4 billion added to gas bills on the same basis, the two together account for $46.1 billion of additional cost above the no-growth case. This is the cost of elevated prices for loads other than the new data centers.
Loading map…
The zones with the largest electricity cost increases are the data center zones, so the cost looks local at first glance. But the gas premium driving part of it is national, and in an interconnected market the energy price effect spreads well beyond the zone that hosts the load.
Off-grid data centers
Some data centers have begun skipping the grid entirely, generating on site rather than waiting in the interconnection queue. That does not relieve this pressure.
For transmission and capacity, this works. No interconnection means no shared network costs to allocate and no capacity obligation to argue about, which is why islanding is presented as the answer to the ratepayer question. It does not work for the gas price, and it makes that pressure worse.
The technologies of choice for island mode are natural gas turbines and reciprocating engines, so an off-grid campus adds gas burn directly. It also burns more gas per megawatt-hour than the same load served from the grid would have, and that follows from the way an islanded system has to be built.
A grid-connected plant can run near its most efficient point because the rest of the system provides reserves, ramping and backup. An islanded campus has to provide those itself, so it runs a mix of units, some of them less efficient and often at part load. The result is a higher average heat rate than the grid would have had. We did not model off-grid campuses, so we do not know by how much. Either way, the price effect in Figure 1 does not depend on which side of the meter the gas was burned.
From there, the pass-through is unchanged. A higher marginal gas price reaches every gas customer through delivered rates, and reaches every grid electricity customer through the market heat rate, including customers of a utility that never signed an interconnection agreement with the facility. The campus buys no electricity from the grid and still raises the price other customers pay. And because island-mode projects bid for the same turbines as everyone else, they add to the equipment inflation described above.
What current protections reach
Transmission and capacity costs can be assigned to the customer who caused them. Higher gas and electricity prices cannot be assigned to anyone, because no single interconnection caused them.
That does not make them small. Our gas and electricity numbers add $46 billion a year by 2030, paid by customers everywhere, including in states with no data centers. If the goal of the last two years of ratepayer-protection work is that data center growth should not raise other people's bills, the instruments built so far reach only the smaller half of the problem.
Data centers will pay for what they can be billed for. The larger cost cannot be billed to them, so it is paid by everyone else.
Modeling results are from Evolved Energy Research's RIO model of the US energy system, 2030, relative to a no-data-center-growth counterfactual. Bill impacts are the change in the market price of electricity and of pipeline gas in each zone, allocated to the consumers of those products in the counterfactual, excluding new data centers.