American households and businesses face a stark new reality: the electricity needed to power artificial intelligence could push total U.S. power demand to levels not seen in decades. The Electric Power Research Institute has projected that data centers alone could consume up to 17 percent of all U.S. electricity by 2030, roughly triple the share they claimed just a few years ago. That surge, driven almost entirely by AI workloads, is forcing utilities, regulators, and grid operators to reckon with whether new generation and transmission can scale fast enough to keep the lights on without raising costs or carbon emissions for everyone else.
Why a tripling of data-center load changes the calculus for every ratepayer
The speed of this demand spike is what separates it from past cycles of industrial growth. Traditional load forecasting assumed slow, steady increases tied to population and economic output. AI-driven data centers break that pattern. A single hyperscale campus can draw as much power as a small city, and dozens of these facilities are under construction or in advanced planning across Virginia, Texas, Georgia, and the Midwest. When that concentrated demand lands on local grids, it competes directly with residential and commercial customers for the same generation and transmission capacity.
The International Energy Agency has identified data centers as a primary driver of electricity demand growth in advanced economies through the end of the decade, highlighting this trend in its energy and AI summary. That assessment tracks with what U.S. utilities are reporting in their integrated resource plans: load growth projections that would have seemed implausible five years ago are now baseline assumptions. If generation additions lag behind, wholesale power prices rise, and those costs flow through to monthly bills for households and small businesses that have no direct connection to the AI boom.
One hypothesis circulating among energy analysts holds that if utilities accelerate procurement of 24/7 carbon-free energy contracts at the pace set by major hyperscale deals struck during 2024, the share of data-center load met by new renewables could exceed 60 percent by 2028. That outcome would materially lower the emissions intensity projected in current forecasts. The logic is straightforward: tech companies with net-zero commitments are willing to sign long-term power purchase agreements that guarantee revenue for new solar, wind, and battery projects. Those contracts, in turn, pull forward clean generation that benefits the broader grid. The catch is that contract announcements and actual megawatts on the wire are not the same thing. Permitting delays, interconnection queues, and supply-chain bottlenecks for transformers and high-voltage cable all threaten to widen the gap between pledged capacity and delivered electrons.
EPRI’s 17 percent projection and the data behind it
The most widely cited number anchoring this debate comes from the Electric Power Research Institute, which stated that data centers could consume up to 17 percent of U.S. electricity by 2030 in a scenario outlined on its public announcement. EPRI is a nonprofit research organization funded by more than 450 utilities and energy companies worldwide, and its estimates carry significant weight in regulatory proceedings and utility planning cycles. The 17 percent figure represents an upper-bound scenario, not a single-point forecast, which means it accounts for aggressive AI adoption timelines and assumes that planned data-center construction proceeds without major delays.
Within that scenario, AI inference and training workloads are the dominant contributors to incremental demand. Compared with traditional cloud computing, AI models require far more computing power per unit of work, and they run continuously once embedded into consumer products and enterprise software. That shift from episodic to always-on computing helps explain why EPRI’s upper-bound estimate rises so sharply relative to earlier projections of data-center electricity use.
The U.S. Department of Energy has adopted EPRI’s framing as part of its own policy response. The department’s Office of Electricity published a resource page on clean energy options to meet data-center electricity demand, citing the EPRI estimates and linking to a funding opportunity through the Office of Clean Energy Demonstrations. That federal attention signals that Washington views the data-center power crunch not as a narrow industry problem but as a grid reliability and climate policy challenge that affects the entire energy system.
In parallel, international modeling reinforces these concerns. The IEA’s technical work on energy demand from AI frames the United States as the single largest growth region for AI-related electricity consumption, driven by the concentration of hyperscale operators, chip fabrication facilities, and venture capital flowing into AI startups. While the global analysis does not dictate U.S. outcomes, it supports EPRI’s conclusion from a different angle: even under moderate adoption scenarios, data-center electricity use in the U.S. grows faster than total generation capacity additions in the near term, tightening reserve margins unless new resources arrive on schedule.
Grid bottlenecks and unanswered questions heading into 2027
Several critical unknowns will determine whether the 17 percent scenario materializes or whether practical constraints slow the buildout. First, the full EPRI technical report with state-level load forecasts and detailed methodology has not been publicly released beyond the headline figure. Without those granular projections, state regulators and independent system operators are working with incomplete information as they approve new generation and transmission projects. That opacity makes it harder to distinguish between regions facing manageable growth and those at risk of severe congestion or reliability shortfalls.
Second, the specific funding criteria and award data from the Department of Energy’s Office of Clean Energy Demonstrations remain limited to what is posted on the initial listing. Until project selections, locations, and timelines are disclosed, grid planners cannot reliably incorporate those federal investments into their capacity outlooks. The result is a planning environment where utilities must prepare for rapid load growth while only loosely estimating when federally backed clean resources will come online.
Third, the physical constraints of the grid loom as a potential brake on both AI expansion and clean-energy deployment. Long interconnection queues already delay new solar, wind, and storage projects by years in some regions. Large data centers, which often require new substations and high-voltage lines, face similar backlogs. Even when projects clear regulatory review, shortages of critical equipment such as large power transformers can push in-service dates well beyond initial expectations. These bottlenecks raise the possibility that data-center developers will secure interim solutions, such as on-site gas generation, that meet reliability needs but undercut decarbonization goals.
Finally, the pace and shape of AI adoption itself remain uncertain. If hardware efficiency improves faster than expected or if companies shift more workloads to off-peak hours, actual electricity use could fall below today’s aggressive scenarios. Conversely, if AI becomes embedded in a wider array of consumer devices, industrial processes, and public services, demand could overshoot even EPRI’s upper bound. For now, policymakers and utilities must navigate that uncertainty while making long-lived investment decisions.
What it means for consumers and climate policy
For ratepayers, the stakes are straightforward. If utilities can align rapid data-center growth with equally rapid deployment of new, low-cost clean generation and grid upgrades, the AI boom could help finance infrastructure that ultimately benefits all customers. Long-term contracts with creditworthy tech companies can lower financing costs for renewables and storage, spreading fixed costs across a larger base and potentially moderating bills.
If that alignment fails, however, households and small businesses could face higher electricity prices and greater exposure to reliability risks without sharing in the economic upside of AI. In that scenario, regulators may come under pressure to revisit how grid costs are allocated, whether large data centers should pay higher demand charges, and how to ensure that communities hosting these facilities see tangible benefits beyond construction jobs.
Climate policy is equally entangled in these choices. A surge in fossil-fueled generation to serve AI demand would complicate efforts to decarbonize the power sector, forcing other industries and consumers to cut emissions more deeply to hit national targets. By contrast, if the AI buildout is paired with accelerated clean energy deployment, it could reinforce the case that digital innovation and climate goals can advance together. The difference between those paths will be determined less by the abstract promise of AI than by the concrete details of grid planning, permitting reform, and investment decisions made over the next few years.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
