The American power sector spent two decades in near-stagnation. Electricity demand barely grew between 2005 and 2022. Efficiency gains from LED lighting, more efficient appliances, and industrial process improvements offset growth in population and economic activity. Utilities planned around flat load curves. Transmission investment slowed. Regulatory frameworks calcified around assumptions of modest, predictable demand growth.
AI infrastructure has ended that era. The scale, speed, and geographic concentration of AI-driven electricity demand is not a temporary surge. The power sector cannot absorb it through incremental adjustment. It is a structural shock. It is forcing utilities, regulatory frameworks, capital markets, and rate structures to adapt faster than any of these institutions were designed to move. The restructuring is already underway. Its full consequences for how American electricity gets built, priced, and distributed will define the power sector for the next thirty years.
What Two Decades of Flat Demand Did to the Power Sector
The interconnection queue process manages how new large loads connect to the transmission grid. Utility engineers designed it around historical project volumes. It assumed that interconnection requests would arrive at a pace their teams could process without significant backlog accumulation. That assumption produced a process architecture that worked adequately when it matched reality. It breaks down completely when it does not.
The AI infrastructure buildout arrived with a volume, speed, and geographic concentration of large load requests that the process simply could not handle, a dynamic explored in depth in our analysis of grid interconnection strategy as competitive infrastructure. The resulting backlog is not a temporary administrative inconvenience. It is a symptom of a regulatory and institutional infrastructure optimized for a world that no longer exists. Utilities that spent two decades investing minimally in new large-load connection capacity now face a demand signal that requires them to rebuild that capability at speed. However, rebuilding institutional capacity takes years, not months.
The Workforce Gap Nobody Planned For
Two decades of flat demand contracted the workforce with expertise in large-scale grid construction, major substation development, and high-voltage transmission engineering. When the work contracted, the training pipelines contracted with it. Universities produced fewer power systems engineers. Utilities hired fewer specialists. As a result, the institutional knowledge required to plan, permit, and build transmission infrastructure at scale became concentrated in a shrinking cohort of experienced professionals.
AI infrastructure demand now competes for that workforce across dozens of simultaneous large-scale projects in the same geographic markets. The shortage is not just of steel and transformers. It is also of the engineers and project managers who know how to deploy them. Utilities cannot hire their way out of this constraint quickly. Training a qualified power systems engineer takes years. Consequently, rebuilding the institutional knowledge base that flat demand eroded will take longer than the current infrastructure build cycle allows.
How Planning Assumptions Became Liabilities
Utility integrated resource plans project future demand and plan the generation and transmission capacity to meet it. Most utilities updated their plans on three to five year cycles. Those cycles assumed that load forecasts would shift gradually enough that planning processes could absorb the changes without fundamental restructuring. That assumption no longer holds.
AI infrastructure compressed that assumption into irrelevance. A single large AI campus can add hundreds of megawatts of sustained load to a utility’s service territory on a timeline that existing planning processes were never designed to accommodate. Furthermore, utilities whose integrated resource plans projected flat or modest demand growth now find those plans structurally inadequate before the ink is dry. The planning frameworks themselves require redesign, and redesigning regulatory frameworks takes time that the market does not have.
Capacity Markets Are Pricing a New Reality
PJM Interconnection manages the grid serving more than 65 million people across thirteen states and the District of Columbia. It runs annual capacity auctions that set the price utilities and large customers pay for guaranteed power availability. Capacity prices in PJM rose from roughly twenty-nine dollars per megawatt-day for the 2024 to 2025 delivery year to more than three hundred dollars per megawatt-day for the 2026 to 2027 delivery year. That is a tenfold increase in two years. AI-driven load growth in PJM’s service territory directly drove that increase by outpacing the capacity that generators committed to the market, a pattern covered in our analysis of power economics and the US data center grid bottleneck.
Those price signals are not abstractions. They translate directly into higher electricity bills for the households and businesses sharing the grid with AI data centers. Utility commissions in Virginia, Ohio, and other states with high concentrations of AI infrastructure are already receiving complaints from residential and commercial customers about rising capacity costs. As a result, the political economy of electricity pricing is reshaping itself in ways that regulators, utilities, and data center operators have not yet fully worked through together.
What Rate Cases Will Look Like in Five Years
Utilities recover the cost of grid upgrades through rate cases. These are the regulatory proceedings through which utilities propose new electricity tariffs and commissions approve or modify them. The cost allocation methodology that determines how upgrade costs spread across the customer base has historically been contested territory. Data center operators argue that grid upgrades produce broad system benefits that justify socializing the costs. In contrast, residential and commercial customers argue that they should not subsidize infrastructure built primarily to serve large industrial loads.
This dispute is becoming the central regulatory battleground of the AI infrastructure era. The outcome will determine whether data center development economics in the most constrained markets remain viable. It will also determine whether communities hosting AI infrastructure absorb cost increases that offset any local economic benefits. State public utility commissions that have not yet updated their cost allocation frameworks for AI-scale loads will find themselves adjudicating these disputes without adequate precedent or analytical tools.
How Merchant Generators Are Repositioning
The capacity price signals coming out of PJM and other organized markets are triggering a repositioning among merchant power generators. These are companies that own and operate generation assets without the regulated rate of return that investor-owned utilities receive. Merchant generators whose assets were marginal under the flat demand environment of the previous decade are suddenly operating in markets where their capacity commands prices that make previously uneconomic assets financially attractive.
Nuclear plants that faced retirement pressure a decade ago are now getting relicensed and upgraded. Natural gas peakers that operated infrequently under low capacity price regimes are running more hours and at better margins. Meanwhile, the financial revival of these assets is directly attributable to the demand signal that AI infrastructure is sending through capacity markets. Merchant generators who positioned themselves early in markets where AI infrastructure concentrates are capturing value that competitors who did not make those positioning decisions cannot easily replicate.
How Hyperscalers Are Changing the Utility Business Model
For most of American electricity history, regulated utilities or merchant generators built generation, transmission, and distribution infrastructure within frameworks established by state and federal regulators. Large customers bought power at regulated tariffs and had limited ability to influence how or where generation capacity got built. AI infrastructure operators have broken that model entirely.
The scale of their power requirements, the urgency of their timelines, and their willingness to commit capital that dwarfs what most utilities deploy in a year have given hyperscalers leverage to restructure the terms of their relationship with the power sector. No previous category of large customer has achieved this. The hyperscaler is no longer a passive electricity consumer. It is an active participant in shaping how American power infrastructure gets built.
Behind the Meter Generation as Structural Bypass
The clearest expression of hyperscaler power in the utility relationship is the proliferation of behind-the-meter generation. By co-locating their own generation capacity at or adjacent to their data center campuses, hyperscalers and their infrastructure partners bypass the interconnection queue entirely. They avoid the rate structures that utilities use to recover grid upgrade costs. Additionally, they gain control over their power delivery timelines in ways that grid-dependent development cannot achieve.
The Federal Energy Regulatory Commission’s December 2025 ruling approved co-location agreements between data centers and power plants. This formalized what the market had already been pursuing informally. Projects representing dozens of gigawatts of behind-the-meter generation capacity are now in various stages of development across Texas, the Southeast, and the Mountain West. Crusoe’s 900 MW campus in Abilene, Texas, built on a behind-the-meter model to support Microsoft’s AI infrastructure, represents one of the most visible examples of how this approach reshapes the geography of AI infrastructure development.
What Co-Location Means for Utility Revenue
Behind-the-meter generation does more than give AI operators control over their power supply. It removes load from the utility’s rate base. That reduction in the rate base reduces the revenue utilities collect from large customers. It also places a larger share of fixed grid costs on the remaining customer base. As more AI infrastructure moves behind the meter, utilities lose their largest and most credit-worthy potential customers to a model that bypasses the grid relationship entirely.
This creates a financial dynamic that regulators have not yet addressed systematically. Utilities investing in grid upgrades to accommodate AI-scale loads may find that a significant portion of those loads ultimately bypass the grid rather than connect to it. Therefore, the capital investment utilities make in anticipation of AI load growth may not generate the rate base recovery they projected. The market continues to move toward behind-the-meter models faster than utility planning processes anticipated.
The Transmission Investment Gap
American transmission infrastructure is old. Builders constructed a significant portion of the high-voltage lines that move electricity across the country in the 1950s, 1960s, and 1970s. Transmission investment declined as demand growth slowed. The regulatory frameworks governing transmission cost recovery created limited incentives for utilities to invest ahead of demonstrated need. The result is a transmission system that entered the AI infrastructure era without the capacity headroom required to absorb the demand concentrations that AI campuses create.
Closing the transmission investment gap requires capital at a scale and speed that the existing regulatory framework was not designed to mobilize. Transmission projects in the United States take a decade or more from planning to operation. They navigate permitting processes involving multiple federal and state agencies, environmental review requirements, and landowner negotiations across routes spanning hundreds of miles. AI infrastructure demand needs power now, not in ten years.
Where Federal Policy Is Trying to Move
The Federal Energy Regulatory Commission has pursued reforms to the interconnection queue process and transmission planning requirements. However, implementation across regional transmission organizations has varied considerably. Some regions have made meaningful progress on queue reform. Others have encountered implementation challenges that have delayed the anticipated improvements.
The Trump administration directed major technology companies to build their own power generation rather than rely on the aging grid. That direction reflects a pragmatic acknowledgment that federal policy cannot close the transmission investment gap on the timeline that AI infrastructure demand requires. Encouraging behind-the-meter development is not a solution to the transmission problem. Rather, it is a workaround that defers the problem while the underlying infrastructure gap continues to widen.
The Role of Long Duration Energy Storage
One technology that could meaningfully change the economics of transmission investment for AI infrastructure is long duration energy storage. If AI data centers can store electricity during periods of low demand and high renewable generation, they reduce the peak transmission capacity required to serve their loads. That reduction in peak demand translates into smaller, less expensive transmission upgrades and potentially faster interconnection timelines.
The technology is advancing but has not yet reached the cost and reliability thresholds required for widespread deployment at AI infrastructure scale. Iron-air batteries, advanced compressed air systems, and other long duration storage technologies are moving through demonstration and early commercial deployment phases. The pace of that development relative to the pace of AI infrastructure buildout will determine how much relief long duration storage can provide to a transmission system under pressure from demand it was not built to serve.
How Ratepayer Politics Will Shape AI Infrastructure Geography
The political economy of electricity pricing is becoming a constraint on AI infrastructure development that the industry has not yet fully internalized. Rising capacity prices in PJM and other organized markets are producing rate increases for residential and commercial customers. Those increases are large enough to attract political attention. State legislators in Virginia, Ohio, and Pennsylvania are introducing bills that would restrict data center development, require data centers to demonstrate local economic benefits before receiving expedited permitting, or mandate that data centers source their own power without burdening the existing rate base.
These legislative efforts reflect genuine constituent pressure from households and businesses experiencing electricity cost increases that they attribute to AI infrastructure development. The political logic is straightforward. If data centers drive up electricity bills for residents who receive limited direct economic benefit from their presence, elected officials face pressure to limit that impact. The industry response of pointing to job creation and tax revenue is adequate in markets where those benefits are broadly distributed. It is less persuasive in markets where the jobs go to specialized workers recruited from outside the region and the tax revenue flows to jurisdictions that already have strong fiscal positions.
Virginia as the Defining Test Case
Northern Virginia hosts the largest concentration of data center capacity in the world. The region’s dominance in the data center industry reflects decades of investment in fiber infrastructure, a skilled technical workforce, favorable tax treatment, and proximity to major East Coast population centers. Those advantages have not disappeared. However, rising electricity costs and grid congestion are creating real friction for new development.
Virginia’s State Corporation Commission has been managing an increasingly complex set of competing interests among Dominion Energy, data center operators, residential customers, and environmental advocates. The cost allocation disputes playing out in Virginia rate cases attract close attention from regulators in every other state with significant AI infrastructure development. The decisions made in Virginia over the next two years will establish precedents that shape how the ratepayer politics of AI infrastructure get resolved nationally.
How the Industry Needs to Respond
The data center industry’s traditional approach to regulatory engagement relied heavily on economic impact arguments and behind-the-scenes lobbying. That approach is no longer sufficient in markets where electricity cost impacts on residential customers are large enough to generate organized political opposition. A more sophisticated engagement model is required. It must address the cost allocation question directly rather than deflecting it.
Operators who make credible commitments to behind-the-meter generation, who structure their grid relationships to minimize cost shifting onto residential customers, and who invest in transmission upgrades that provide broad grid benefits will find the regulatory environment more navigable. In contrast, developers who treat ratepayer impact as a political problem to be managed rather than an economic problem to be solved will encounter resistance that grows more organized and effective over time. The social license to build AI infrastructure in American communities is not guaranteed. It has to be earned, and the terms on which it gets earned are changing faster than most of the industry has recognized.
The Long Term Restructuring Nobody Has Fully Modeled
The power sector transformation driven by AI infrastructure is not a transition between two stable states. It is an ongoing restructuring whose endpoint nobody has fully modeled. The pace of AI infrastructure development and the policy responses it triggers are both moving too fast for conventional long-range planning to track reliably. Utilities, regulators, and capital markets need to make multi-decade investment decisions. They are doing so in an environment where the assumptions underlying those decisions are shifting faster than the planning cycles through which they get made.
The operators and developers best positioned as this restructuring plays out are those who engage with it as a structural reality rather than a regulatory obstacle. That means investing in utility relationships that go beyond transactional procurement. It means participating in transmission planning processes that shape the grid investments utilities will make over the next decade. It also means making power infrastructure commitments that provide utilities with the demand visibility they need to justify capacity investment.
What the Next Decade Looks Like for Utilities
American utilities face a strategic choice with no clean precedent in their regulated history. They can invest aggressively in the grid infrastructure required to serve AI-scale loads. However, doing so means accepting the regulatory risk that cost recovery mechanisms may not fully compensate them if load growth projections prove optimistic. Alternatively, they can invest conservatively and accept the risk that AI infrastructure operators build around them through behind-the-meter generation. That path leaves utilities with aging assets, shrinking rate bases, and a customer mix increasingly weighted toward the residential and commercial customers least able to absorb cost increases.
Neither path is without risk. The utilities that navigate this choice most successfully will be those that build genuine partnerships with AI infrastructure operators. Those partnerships require developing the mutual understanding and shared planning frameworks needed to make large capital commitments on timelines the market demands. Both parties must accept transparency about their plans and constraints. That transparency does not yet exist in most markets. Building it will take time and trust that neither side has fully committed to establishing.
Why This Restructuring Is Irreversible
The restructuring that AI infrastructure drives in the US power sector is not a cycle that will reverse when AI investment slows or interest rates rise. The physical infrastructure being built now, including the transmission lines, substations, generation assets, and interconnection agreements, will shape the grid for decades. That is true regardless of what happens to AI investment levels in any particular year. Additionally, the regulatory precedents established in rate cases, interconnection proceedings, and legislative chambers will define the frameworks within which the power sector operates long after the current AI infrastructure buildout runs its course.
The operators, utilities, and regulators who recognize the permanence of this restructuring and invest accordingly will shape the outcome. Those who treat it as a temporary disruption to be managed until conditions normalize will find themselves operating within a power sector whose rules, economics, and competitive dynamics others have already written for them. The AI infrastructure era is not a moment in the power sector’s history. It is the beginning of a new chapter, and the investment decisions, regulatory proceedings, and commercial negotiations playing out across the American electricity system today are setting its terms right now.
