An often-overlooked consequence of the artificial intelligence revolution is quietly forming beyond the realms of software and algorithms—it’s unfolding on the nation’s electrical grids. This emerging phenomenon, sometimes described as an AI-driven energy bubble, could soon have tangible repercussions for consumers, particularly through rising electricity bills. Across the United States, power providers that supply energy to massive data centers are petitioning regulators for authorization to embark on exceedingly expensive upgrades—spanning the construction of new power plants to extensive improvements in grid infrastructure. Their arguments hinge on electricity demand projections that, according to critics and independent experts, appear significantly inflated. If these overestimations persist unchecked, consumers may ultimately find themselves financing surplus generation capacity that may never actually be required.
A growing number of utilities are employing exceptionally high energy demand estimates as justification for an anticipated explosion in AI and cloud computing infrastructure. These projections stretch from the southeastern hub of Georgia through the energy-rich expanse of Texas and upward into the technologically oriented regions of Minnesota. The complication stems from the way data center developers request power—they frequently submit multiple applications to various utility providers in different states or districts, hoping that at least one will materialize. Constellation Energy’s CEO, Joseph Dominguez, likens this approach to the sport of fishing: many lines are cast into the water with the expectation that one might yield a catch. Similarly, developers cast numerous requests, each representing a potential project that may or may not come to fruition.
To illustrate the inherent inefficiencies in this process, imagine a developer planning a single data center submitting electricity requests to five separate utilities operating independently and unaware of each other. Each utility, perceiving the potential demand as genuine, faces the choice of expanding capacity or constructing new power plants. The consequence could be the construction of several redundant facilities—each financed by incremental increases to residential and commercial rates. Once the developer finalizes its choice of location, the remaining proposed projects—and their associated energy infrastructure—are often abandoned, creating surplus generation in regions where the anticipated consumption fails to materialize.
Regulators and utility commissions in some jurisdictions have started to scrutinize these practices, yet the tactic remains prevalent nationwide. These concerns are intensifying amidst already mounting electricity prices, prompting warnings from economists, environmental advocates, and energy analysts who see early signs of systemic strain.
The implications of these choices are enormous. Utilities must maintain a delicate equilibrium: undersupply risks grid instability and interruptions, whereas oversupply burdens consumers with decades of expenses tied to underutilized or idle power plants. As Mark Dyson, a managing director at the Rocky Mountain Institute—a nonprofit dedicated to accelerating sustainable and market-based energy transitions—explains, regulated utilities often operate under a structural incentive to forecast excessive demand growth. Such projections permit them to justify capital investments and subsequently recoup those costs plus profit through rate structures, an arrangement that subtly rewards expansion even when unwarranted by actual demand.
Experts like Jeremy Fisher, a principal advisor with the Sierra Club who has spent years assisting states and municipalities in energy resource planning, observe that few within the field truly believe utilities will build enough infrastructure to deliver on their most aggressive forecasts. Nonetheless, these speculative figures still wield influence: they shape narratives used to secure regulatory approvals and funding. Fisher warns that consumers may already be paying for speculative capacity embedded within current rates, or soon will be, as these forecasts continue to drive utility expenditure plans.
This scenario has reached global proportions. A McKinsey analysis projects that by 2030, data centers worldwide will demand roughly 219 gigawatts of power. Yet, 26 of the United States’ largest investor-owned utilities collectively claim planned or proposed projects requiring nearly 711 gigawatts—an astonishing figure that approaches the entire peak summer load of the continental U.S. According to the Sierra Club’s compilation of these estimates, the discrepancy reveals a pattern of overstatement that could distort national energy planning for years to come.
Underlying these forecasts is a complex web of incentives and misaligned interests. The major technology corporations—Amazon, Google, Microsoft, and Meta—rarely construct every facility themselves. They often rely on developers and intermediaries who specialize in acquiring land, obtaining permits, and initiating projects with the intent of later selling or leasing them to a larger client. In this frenzied environment, competition is fierce, and participants of all sizes—from small speculative builders to seasoned real estate firms—are eager to secure a foothold. Each player, in pursuit of opportunity, approaches multiple utilities seeking guaranteed future electricity access. As utilities evaluate these requests, many initiate feasibility studies, unaware that competing utilities are doing the same for the same developer and project.
Constellation Energy’s Dominguez confirms these practices, noting conversations with clients that demonstrate identical data center proposals being entertained concurrently in multiple jurisdictions. According to Astrid Atkinson, cofounder of Camus Energy and a former senior engineer at Google’s data centers, most of these applications never advance beyond preliminary stages. Developers customarily maintain several potential sites during early planning, assessing costs, permitting conditions, and grid capacity, before culling all but the most viable. Atkinson estimates that utilities might be receiving five to ten times more grid connection requests than the number of data centers that are ultimately constructed.
Some established operators acknowledge the problem and seek to distinguish themselves from less deliberate competitors. Ryan Mallory, CEO of Flexential—a data center operator backed by Morgan Stanley and running more than forty facilities—emphasizes that indiscriminately submitting load requests across dozens of markets is poor practice. However, he admits that newer entrants often adopt precisely that approach, “carpet-bombing” the market with speculative inquiries. Such behavior, he warns, generates widespread uncertainty that heightens tension among utilities, regulators, and legitimate developers alike.
Even as demand projections soar, another layer of irony persists: the hardware necessary to use such vast amounts of power remains scarce. Semiconductors, specifically GPUs manufactured by Nvidia and AMD—the computational backbone of contemporary AI—represent more than half of an AI data center’s overall energy consumption. Market research by Enverus suggests that global GPU production will only support approximately five gigawatts of computing power in the current year and may grow to about 9.5 gigawatts by 2028. Accounting for additional energy requirements such as cooling, lighting, and non-computational operations, the total electricity demand likely peaks near 19 gigawatts, far below utility-level forecasts. As analyst Carson Kearl of Enverus notes, these discrepancies trace back to forecasting methods that appear increasingly divorced from tangible technological supply constraints.
Data centers, nevertheless, represent a rare growth engine for utilities whose traditional power sales have stagnated due to decades of efficiency improvements in appliances and industrial systems. Research from Lawrence Berkeley National Laboratory reveals that data centers consumed roughly 4.4% of all U.S. electricity in 2023. That figure could rise substantially, reaching between 6.7% and 12% by the end of this decade. In parallel, the expanding market for electric vehicles is also pushing demand upward. A Business Insider investigation found over 1,200 data centers planned or operational nationwide as of 2024, with combined energy demands equivalent to that of entire states such as Ohio or Florida.
While analysts generally agree that at least some new generation capacity will be necessary to accommodate the AI surge, the critical question remains: how much is enough? Grid Strategies, a firm that monitors utility load forecasting, observed late in 2024 that both the Electric Reliability Council of Texas (ERCOT) and the PJM Interconnection—covering major portions of the eastern U.S.—have each dramatically increased their five-year demand outlooks. ERCOT raised its projection by nearly 37 gigawatts, and PJM by over 15 gigawatts. Historically, such forecasts have tended toward overestimation. Data from the Rocky Mountain Institute indicate that between 2012 and 2023, utilities overshot demand by an average of about 23%.
Illustrating these challenges, Kentucky-based utilities recently justified $3 billion in new gas-fired plants based on expectations of six gigawatts of forthcoming data center load, despite evidence that only a fraction of those projects were likely to materialize. Regulators nonetheless approved the expansion, underscoring the persistence of optimistic forecasting even in the face of contradictory data. Similarly, Dominion Energy, serving Virginia’s so-called “Data Center Alley,” has acknowledged that many incoming connection requests stem not from major tenants but from speculative developers with limited credentials. A Dominion spokesperson noted that actual contracted demand continues to track closely with the forecasts used by PJM, offering some validation, yet conceded that competition among developers is fierce—“everybody competing for a growing slice of the pie.”
Industry experts, including John Wilson of Grid Strategies, explain that forecasting inaccuracies also partially reflect methodological obsolescence. Until a few years ago, utilities modeled demand according to population and economic growth. That approach worked when electricity usage primarily mirrored residential and industrial expansion, but it falls short amid the unprecedented scale of AI data center loads. Today, more than 80% of new demand forecasts involve large, uncertain projects, underscoring a pressing need for updated predictive tools and frameworks.
Policymakers are beginning to intervene. One method to curb speculative requests involves imposing financial commitments upfront. Ohio recently introduced one such regulation, requiring data center developers to pay for at least 85% of the capacity they reserve, irrespective of actual usage. After this rule’s implementation, requests to the state’s largest utility dropped by more than half—a clear sign that many prior submissions had likely been exploratory rather than genuine. Similar “large load tariffs” are now under consideration in roughly 20 states, marking the first coordinated national response to curtail runaway speculation.
Nevertheless, costs for consumers continue to surge. In the PJM region, prices in the capacity market—where utilities secure future power supplies—reached record highs, with overall commitments totaling over $16 billion, an increase partly attributed to the burgeoning data center industry. Analysts warn that if unchecked, this dynamic may translate into billions of dollars in annual additional costs borne by ordinary ratepayers.
According to Sierra Club advisor Jeremy Fisher, awareness is finally broadening among regulators and public advocates. He describes a shifting momentum as commissions, consumer groups, and even some utilities now recognize the financial risks inherent in subsidizing speculative projects. Measures like Ohio’s are hailed as early successes—an attempt to ensure fairness, promote grid reliability, and align investment with reality rather than hype. In essence, the emerging debate is not merely about electricity—it is about foresight, accountability, and the immense economic forces shaping the infrastructure of the AI age.
Sourse: https://www.businessinsider.com/ai-boom-bubble-power-utilities-forecasting-demand-2025-11
