Electric utilities and grid operators across the United States face a stark planning problem: data centers consumed about 4.4 percent of national electricity in 2023, and federal projections show that share could nearly triple within five years. The Department of Energy estimates data centers will use between 6.7 and 12 percent of all U.S. electricity by 2028, a range wide enough to reshape how and where new power plants get built. For ratepayers, developers, and state regulators, the question is no longer whether load growth is coming but how fast generation and transmission can keep up.
Why the DOE’s electricity range forces grid decisions now
The scale of the projected increase is what makes it urgent. Data centers used 58 terawatt-hours of electricity in 2014 and 176 TWh in 2023, according to a recent Department of Energy report. The agency projects that figure will climb to between 325 and 580 TWh by 2028. At the high end, that would mean data centers alone consume roughly as much electricity as the entire residential sector of several mid-sized states combined.
A gap that wide between the low and high scenarios reflects genuine uncertainty about how quickly artificial intelligence workloads will scale and how aggressively hyperscale operators will build. But even the lower bound, 325 TWh, represents an 85 percent jump from 2023 levels in just five years. Grid operators in regions already facing tight reserve margins, particularly in the PJM Interconnection territory stretching from the mid-Atlantic through the Midwest and in the ERCOT system covering most of Texas, cannot wait for the uncertainty to resolve before acting. New gas-fired generation, battery storage, and transmission lines all require years of permitting, environmental review, and construction. If utilities delay investment decisions until the load materializes, the result will be reliability shortfalls and price spikes for all customers on the same grid.
The hypothesis that the DOE range will trigger accelerated gas-plant and transmission permitting in at least three PJM and ERCOT zones within 18 months is testable. FERC filings, state siting dockets, and interconnection queue data will show whether utilities and independent power producers are responding to these projections with concrete applications. Early signs point in that direction: several large-scale generation projects have entered queue processes in northern Virginia, central Texas, and parts of Ohio, all regions where data-center clusters are expanding. Confirmation will come from tracking whether those applications move faster through review than historical averages and whether regulators explicitly cite data-center demand in their approval orders.
LBNL data and the Energy Act mandate behind the projections
The numbers driving this debate come from a specific study by Lawrence Berkeley National Laboratory, produced under a mandate from the Energy Act of 2020. That law requires the DOE to make an updated data center energy-use report publicly available, giving the findings statutory weight that distinguishes them from private-sector forecasts. The LBNL researchers built their estimates on facility-level surveys, utility billing data, and equipment efficiency trends, then modeled scenarios based on different rates of AI adoption and cloud migration.
The DOE has treated the LBNL findings as authoritative enough to propagate across multiple program offices. The agency’s geothermal energy division, for instance, cites the same 6.7 to 12 percent range in materials positioning geothermal resources as a potential supply source for data centers. That cross-cutting adoption signals the department views data-center load growth not as a niche technology issue but as a system-wide challenge affecting generation planning, transmission policy, and clean-energy deployment targets simultaneously.
S&P Global analysts have separately projected that data-center power demand will roughly triple by 2030, a timeline and magnitude broadly consistent with the DOE’s upper-end scenario. The convergence of a federal laboratory study and a major credit-rating agency’s independent analysis strengthens the case that the growth trajectory is real, even if the exact pace remains debatable. Utilities that dismiss these projections as speculative risk being caught flat-footed by load that arrives faster than their infrastructure plans assumed.
Gaps in regional data and what grid watchers should track next
The DOE and LBNL numbers describe national totals, and that aggregation obscures a critical detail: data-center load is not spread evenly across the country. Northern Virginia’s “Data Center Alley” in Loudoun County accounts for a disproportionate share of U.S. capacity, and new clusters are growing rapidly in Texas, Georgia, and the Phoenix metro area. The LBNL report does not break out consumption by utility service territory or regional transmission organization, leaving state regulators and grid planners without a federal baseline for their specific zones.
That gap matters because the strain on any given grid depends on where the load lands, not just how large it is nationally. A 50-megawatt data center connecting to a transmission node with ample headroom creates a different planning challenge than the same facility requesting service in a constrained corridor. Without facility-level or at least regional consumption data from the federal government, state commissions and regional transmission organizations must stitch together their own picture from interconnection queues, local land-use approvals, and utility integrated resource plans.
For grid watchers trying to understand how the DOE projections translate into real-world infrastructure, several indicators are worth monitoring. First, interconnection requests associated with data-center campuses or large AI clusters should show whether developers are favoring specific substations or transmission corridors, potentially exacerbating local bottlenecks. Second, capacity-market auctions in regions like PJM can reveal whether generators are betting on sustained load growth by locking in long-term commitments. Third, state-level resource plans and transmission expansion studies will indicate how quickly utilities are adjusting their assumptions about future demand.
Another key question is how much of the incremental data-center load will be met by on-site or dedicated clean-energy projects versus the broader grid mix. Hyperscale operators have signed gigawatts of power purchase agreements for wind and solar, but the timing and location of those projects do not always align with when and where new data centers connect. The DOE’s range implicitly assumes that, regardless of contract structures, the physical electrons must still flow through regional grids that may already be strained during peak hours.
As regulators and utilities digest the LBNL study and DOE projections, the lack of granular regional data should not become an excuse for inaction. Instead, it underscores the need for closer coordination between local land-use authorities that approve data-center campuses, state energy offices that set policy targets, and grid operators responsible for reliability. Even if the ultimate outcome lands near the lower end of the DOE range, the pace of change is fast enough that planning decisions made over the next two years will determine whether data-center growth is integrated smoothly or triggers avoidable reliability and cost problems for everyone connected to the same wires.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
