American households and businesses face a collision between surging data-center power demand and a grid that was never designed to absorb it. The U.S. Department of Energy reported that data centers consumed about 4.4% of total national electricity in 2023 and projected that share will climb to between 6.7% and 12% by 2028. At the same time, the Federal Energy Regulatory Commission has launched a formal review of deals that let data centers plug directly into power plants inside the PJM regional grid, raising pointed questions about who pays for the grid upgrades those arrangements require.
Why a tripling of data-center load threatens household power bills
The gap between 4.4% and 12% is not an abstraction. If data centers triple their share of U.S. electricity consumption within five years, every other customer on the same wires, from suburban homes to steel mills, competes for generation and transmission capacity that was planned years ago for slower, steadier growth. The tension is sharpest in the PJM Interconnection territory, which covers all or part of 13 Eastern states and the District of Columbia. There, large technology companies have been negotiating co-location agreements that let data centers draw power directly from generating stations, bypassing the open grid.
FERC responded by ordering a review of co-location arrangements involving AI-enabled data centers at generating facilities in PJM. The commission’s stated focus is reliability and cost allocation, two issues that translate directly into rate cases. When a power plant that once served the broader grid redirects megawatts to a single data-center tenant, the remaining customers may need new transmission lines or replacement generation. The central question is whether utilities that approve these deals will pass the cost of grid reinforcements to residential ratepayers rather than to the data-center operators who triggered the spending.
That hypothesis, that PJM utilities approving co-location deals will file rate cases within 18 months shifting at least 15% of incremental grid-upgrade costs onto households, cannot yet be confirmed or rejected. No publicly available cost-allocation modeling from PJM or individual utilities quantifies the residential share of those upgrades. But the structural incentive is clear: utilities earn regulated returns on capital investment, and spreading costs across a broad residential base is far simpler than negotiating bespoke contracts with a handful of hyperscale tenants. The FERC proceeding is the first federal checkpoint that could force transparency on this question.
DOE and LBNL numbers behind the 6.7-to-12% forecast
The headline range comes from a U.S. Department of Energy analysis that draws on Lawrence Berkeley National Laboratory’s tracking of data-center electricity use. In that work, summarized in a recent federal assessment, researchers estimated that data centers accounted for roughly 4.4% of U.S. electricity in 2023 and could rise to between 6.7% and 12% by 2028. The wide spread between the low and high scenarios reflects uncertainty about how fast AI training clusters, cloud computing expansions, and new facility construction will actually come online.
A separate explainer from the Congressional Research Service adds an important definitional wrinkle. Standard data-center energy counts often exclude cryptocurrency mining operations, which run their own power-hungry facilities. Including crypto loads could push the true electricity share higher than even the DOE’s upper bound. The CRS brief cites the same LBNL work but flags these scope gaps, warning that policymakers and grid planners may be working with an incomplete picture of total demand.
The DOE’s broader effort to track and manage demand growth is organized through an electricity-demand hub that highlights AI workloads and cloud expansion as primary drivers. Federal agencies are treating data-center load growth as a grid-planning emergency, not a distant trend. Yet neither the DOE report nor the LBNL dataset has released state-level or hourly load projections that would let independent analysts verify where and when the strain will hit hardest. Without that granularity, grid operators and state regulators are making billion-dollar infrastructure bets on national averages that may mask severe regional pinch points.
Gaps in the evidence and what ratepayers should watch next
Three significant holes remain in the public record. First, the FERC show-cause action on co-location does not include any attached cost-allocation model or consumer bill-impact analysis. Regulators have identified the problem but have not yet quantified it. Second, the CRS report references the LBNL methodology without reproducing its sensitivity cases, so outside researchers cannot independently stress-test the 6.7% to 12% range or evaluate how different assumptions about AI adoption or chip efficiency change the outcome. Third, the DOE hub lists related programs and policy tools but contains no primary records of utility integrated resource plans that would show exactly how much transmission and generation spending data centers are driving in specific service territories.
For households in PJM states, the next concrete signal will come from utility rate filings. When a utility that has approved a co-location deal submits its next rate case to a state public utility commission, the filing will reveal how grid-upgrade costs are allocated among customer classes. Ratepayers and consumer advocates should read those documents for three details: which projects are explicitly tied to new data-center load; how much of each project’s cost is assigned to residential versus large commercial customers; and whether utilities are proposing special tariffs or direct-contribution agreements that require data-center operators to shoulder a larger share of the bill.
Consumer advocates can also press for more disclosure at the state level. Public utility commissions have authority to demand detailed justifications for capital projects and to require utilities to show how alternative scenarios-such as slower data-center buildouts or higher efficiency standards-would change the need for new infrastructure. If utilities cannot demonstrate that co-location deals benefit the broader customer base, regulators could limit cost recovery or mandate separate contracts that keep those expenses off household bills.
At the federal level, the FERC review could compel PJM and its member utilities to standardize how they treat co-location in interconnection and transmission-planning processes. That might include requiring data centers connected behind the plant fence to participate in regional planning, contribute to shared network upgrades, or face limitations on how much capacity they can reserve. Any such rules would ripple quickly into rate design, influencing whether the next wave of AI infrastructure is effectively subsidized by families and small businesses.
None of these steps will slow the underlying demand growth that DOE and LBNL describe. But they could determine who pays for the concrete, steel, and copper needed to keep the lights on as that growth arrives. For now, the core facts are stark: data centers already consume a mid-single-digit share of U.S. electricity; that share could plausibly double or even nearly triple within a few years; and the main mechanisms for allocating the resulting costs are being negotiated in regulatory dockets that most customers never see.
Households in PJM and beyond do not need to master the technicalities of grid planning to protect their interests. They do, however, need to know where the fault lines are: co-location deals that quietly reroute power plant output; planning models that may underestimate total digital demand; and rate cases that can spread the costs of a highly concentrated industry across millions of captive customers. As the data-center boom accelerates, those who pay monthly electric bills will have to decide whether to remain passive bill-payers or to become active participants in the arguments that will decide how much of this new digital infrastructure they are ultimately asked to finance.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
