Last updated 2 months ago. Our resources are updated regularly but please keep in mind that links, programs, policies, and contact information do change.
Microsoft’s stock plummeted in early 2026 after the company revealed an $80 billion backlog of customer orders it cannot fulfill. The problem isn’t chips or engineering talent—America’s power grid cannot provide enough electricity.
This came weeks after five of the world’s largest technology firms announced plans to spend between $660 billion and $690 billion on data center infrastructure in a single year. But the market’s enthusiasm quickly curdled into skepticism.
This exposed what the massive spending announcements obscured: the limiting factor in the AI infrastructure race is not capital or ambition but electrical capacity.
The authority to allocate that capacity rests with a scattered network of federal regulators, state utility commissions, regional grid operators, and local governments. These officials control the allocation of electricity through legal authorities that pit corporate expansion against residential ratepayers, industrial manufacturers, and existing consumers.
The Multi-Layered Regulatory System
The United States electricity system distributes decision-making authority across multiple levels of government. Between federal and state authorities sit Regional Transmission Organizations and Independent System Operators—technically private organizations that manage the physical grid but operate under rate schedules approved by federal and state regulators. Municipal governments and local authorities control the franchises and permissions utilities need to build infrastructure within their jurisdictions.
For a data center developer, the pathway to energization requires moving through all of these layers simultaneously.
The process typically begins with an interconnection request filed with the transmission firm that controls the relevant portion of the grid. That request triggers engineering studies to determine whether the new load can be connected without overloading the existing system—studies that can take years and cost millions of dollars.
These studies are governed by interconnection procedures that FERC established in 2023, which restructured how new generation and storage facilities connect to the grid, but which until recently did not clearly apply to large loads like data centers.
Historically, FERC treated generation interconnections as its responsibility but left large load interconnections to state and local authorities, assuming these would be small and scattered. When data centers began consuming the equivalent energy of entire cities in concentrated locations, the fragmentation of authority created chaos.
A developer in Virginia must satisfy the Virginia State Corporation Commission, Dominion Energy, and PJM Interconnection, each with different procedures, timelines, and legal standards. A developer in Texas faces Oncor, the Electric Reliability Council of Texas (ERCOT), and the Public Utility Commission of Texas, which operate under a different framework entirely. A developer in California goes through the California Public Utilities Commission, Pacific Gas and Electric, and CAISO with yet another set of rules.
This fragmentation reflects America’s federalist structure and the historical development of utility regulation as primarily a state matter. But it creates a practical problem when faced with surging demand.
In October 2025, the U.S. Department of Energy issued a letter directing FERC to assert federal jurisdiction over large load interconnections and establish standardized procedures. The deadline for FERC to propose rules was April 30, 2026.
The outcome of this rulemaking will determine whether the patchwork of state and local authorities can be circumvented by federal standardization or whether allocation remains trapped in a fragmented system not designed for massive industrial demands.
The practical effect of this regulatory fragmentation is profound delays. These reflect the genuine engineering complexity of integrating enormous demands (gigawatts) into grids designed around smaller, dispersed demand, combined with the deliberate procedural protections built into regulatory systems to ensure that big projects support reliable service without harming existing consumers.
Interconnection Queues and Regional Bottlenecks
The specific point of congestion in the approval process is the waiting list for grid connections. These include renewable energy projects, battery storage systems, and new generation facilities, all competing for the same queue. In some regions, new data center interconnection requests have been added at such a pace that queues have become backlogs measured in years.
In PJM, which operates the grid serving thirteen states from New Jersey to Illinois and includes the nation’s largest data center hub in Northern Virginia, the queue has become a political flashpoint. FERC has initiated proceedings demanding that PJM explain and potentially reform its interconnection procedures for large loads.
Lawrence Berkeley National Laboratory research shows that the median duration from initial interconnection request to commercial operation has doubled from under two years for projects built in 2000-2007 to over four years for those built in 2018-2024. For some data center projects in constrained regions, timelines of seven years or longer are now the norm.
Northern Virginia, where Amazon operates massive AWS data center campuses and other tech companies are expanding rapidly, has become so congested that new projects are being turned away. Dominion Energy has exhausted available transmission capacity in some areas. This geographic constraint creates winners and losers among competing tech companies.
Amazon, which operates substantial infrastructure in Virginia and benefits from existing relationships with Dominion, has more negotiating leverage than a new entrant. Oracle and Meta, lacking such established infrastructure, are competing for remaining available capacity by spreading their facilities across less constrained regions like Iowa, Louisiana, and Ohio.
Iowa, which now hosts significant Google and Meta facilities, has taken a relatively permissive approach to data center interconnection, viewing them as economic development assets. Virginia’s State Corporation Commission has approved special rate classes for large loads, recognizing that data centers should bear costs they directly cause but also that these costs must not be allocated to residential customers.
Cost Allocation and Rate Design
The most contentious regulatory issue is the allocation of costs for infrastructure upgrades that must be built to serve new data centers. The fundamental question is: if a utility must build expensive new transmission lines, substations, or generation capacity to serve a data center, who should pay?
The traditional utility regulatory model spreads infrastructure costs across all customers. But data centers are upending this assumption. When a single new customer drives the need for hundreds of millions of dollars in new infrastructure, the question of fairness becomes urgent—particularly when that customer is a multinational technology corporation with $100+ billion annual revenues and the other customers are households and small businesses with limited bargaining leverage.
In Virginia alone, data center-driven transmission projects approved in 2024 are projected to cost $4.4 billion, with nearly half the costs concentrated in Virginia.
Researchers examining utility rate cases found evidence that utilities are using special deals to shift Big Tech’s energy costs to the broader customer base. When utilities secure firm long-term contracts from data centers at below-market rates—rates justified by spreading grid expansion costs across existing ratepayers—everyone except the data center customer subsidizes the infrastructure upgrade.
Virginia’s State Corporation Commission held an extended hearing in April 2025 where Dominion Energy proposed a special rate class for “high-load customers”—data centers consuming 25 megawatts or more. Dominion’s proposal required such customers to sign fourteen-year contracts guaranteeing payment of 60 percent of generation costs and 85 percent of transmission and distribution costs, providing revenue certainty for the utility’s $22 billion in infrastructure investments projected for the next fifteen years.
Consumer advocacy groups and environmental organizations testified that this arrangement locked in decades of cost obligations while providing no guarantee that Dominion would connect the facilities on reasonable timelines. Tech companies pushed back with a joint filing from Amazon, Google, Microsoft, and industry groups suggesting lower cost-sharing percentages—50 percent of generation costs and 75 percent of transmission costs—with shorter contract terms and more favorable exit provisions.
California’s Public Utilities Commission offered a middle path. In a 2025 case involving PG&E and STACK Infrastructure’s 90-megawatt data center, the CPUC approved infrastructure costs but modified the refund structure to account for risk. Rather than allowing STACK to receive a refund based on expected revenues that could amount to nearly full cost recovery in the first year, the CPUC limited refunds to 75 percent of transmission-related revenues collected, holding back 25 percent for maintenance and broader grid upgrades. Under this formula, STACK would receive its full refund across approximately six years rather than one.
Pennsylvania’s Public Utility Commission, by contrast, was still developing its approach as of November 2025, issuing a tentative order proposing a model tariff for large loads that would require demonstrated financial readiness, site control, and minimum contract terms.
These varied regulatory approaches create a patchwork where the economics of a data center project vary dramatically by geography. Companies are increasingly choosing locations based on regulatory friendliness and cost structure, not simply on technical factors like fiber connectivity or climate.
This competitive federalism creates pressure on states to offer favorable terms to secure data center investment, but it also creates the risk that states racing to the bottom on cost allocation protections will impose hidden subsidies on ratepayers.
Tech Company Influence and Strategic Positioning
Tech companies have become sophisticated participants in regulatory proceedings, filing detailed comments on proposed rule changes, retaining expert witnesses for rate case testimony, and maintaining government affairs operations focused on access. Some of this influence is transparent—recorded testimony at public hearings, docket comments available on regulatory websites. But much occurs behind the scenes through negotiations with utilities, meetings with state economic development officials, and private communications with regulators.
The pattern is clearest in Texas, where state lawmakers and ERCOT have become explicitly accommodating to data center requests. In June 2025, Texas Governor Greg Abbott signed Senate Bill 6, legislation that fundamentally reshaped the regulatory framework for large loads in ERCOT. The bill requires utilities to establish interconnection standards for loads of 75 megawatts or larger, allows utilities to recover interconnection costs from the data centers themselves rather than spreading them to general ratepayers, and establishes a process for arrangements where data centers share electricity with nearby generators. The bill also allows data centers to provide information about backup generation only if they choose to do so, limits ERCOT’s ability to curtail them outside of emergency conditions, and provides them with programs that pay data centers to use less electricity when needed.
This accommodation has created political backlash. As data center demand has driven up wholesale prices in Texas—with some regions experiencing price spikes 15 times higher than other regions—consumers have begun demanding accountability. In February 2026, the Texas State Republican Executive Committee passed a near-unanimous resolution calling for “rigorous independent assessments” of proposed data center projects and their impact on grid reliability before final approval, and urging a pause on water-intensive cooling systems.
Virginia presents a more complex political economy. As the nation’s largest data center hub, Virginia has been simultaneously the most welcoming to data center expansion and the most active in trying to protect ratepayers. The state’s tax exemptions for data center equipment cost Virginia roughly $1 billion in tax revenue in 2025 alone. When Dominion Energy sought rate increases in 2025 to fund data center-driven infrastructure upgrades, consumer advocates and environmental groups forcefully challenged the company’s cost allocation methodology, arguing that ratepayers should not subsidize infrastructure built to serve companies claiming significant tax breaks.
Amazon’s stock has remained strong despite higher capex spending because the company has established positions in Virginia and other regions with abundant capacity. Microsoft, lacking equivalent existing infrastructure, has been more aggressive in securing long-term commitments through nuclear partnerships and direct generation investment. Google has pursued a dual strategy, both negotiating favorable interconnection terms with utilities and acquiring Intersect, an energy development firm, for $4.75 billion in December 2025 to accelerate its own generation and speed the approval process by reducing dependence on third-party utilities. Meta negotiated an arrangement with Entergy to match its Louisiana facility’s consumption with 100 percent renewable energy, reducing regulatory controversy by incorporating environmental goals into its sourcing strategy.
Google’s acquisition signals the company’s assessment that regulatory approval and utility coordination have become critical bottlenecks—more valuable to solve through direct investment than through negotiation.
Competitive Dynamics Among Tech Giants
The fragmented, capacity-constrained environment is creating winner-take-most dynamics among the five tech giants themselves. When electricity is unlimited, the five companies compete primarily on technical capability, execution speed, and customer relationships. When electricity is scarce and allocated through regulatory approval processes, the companies also compete on political positioning, established infrastructure presence, and relationships with utilities and regulators.
Amazon’s position is strongest in this context. The company has operated data centers in Northern Virginia since the early days of AWS, giving it established relationships with Dominion Energy and regional grid operators, demonstrated track records of managing large commitments, and exemptions from new rules because they started before the rules under some regulatory frameworks. Amazon’s 2026 capex budget of $200 billion, the highest of the five, gives it capacity to invest in infrastructure itself, including its $20 billion commitment to convert the Susquehanna nuclear plant site in Pennsylvania into an AI-ready data center campus powered by carbon-free nuclear energy. By securing its own supply, Amazon reduces dependence on utilities and regulators who might otherwise limit its growth.
Microsoft lacked significant existing data center presence in the most constrained regions when the AI boom accelerated. Rather than compete for scarce Virginia capacity, Microsoft pivoted to securing long-term commitments outside the grid. This is a bet that regulatory approval for new connections will remain so constrained and slow that securing dedicated generation is more cost-effective than waiting for utility interconnections.
Google, lacking established data center presence in the most valuable markets, has spread geographically. The company announced data center expansion in Iowa in May 2025, in a region with less regulatory resistance and more available capacity than Virginia. Google has also pursued partnerships with renewable energy developers, signing long-term contracts to buy renewable energy with Clearway Energy in 2025.
These varied strategies reflect different assessments of regulatory constraint and different competitive positions in the race for available capacity. Companies with established infrastructure, regulatory relationships, or direct generation investment have clearer paths to scaling. Companies without these advantages face significant delays and cost premiums for access. A smaller, more innovative AI company cannot compete with Meta or Google simply by having better technology if it cannot secure electricity for its data centers. The constraint advantages the largest, most established companies over potential disruptors.
Execution Risk and Investor Skepticism
The most consequential question, which drives Microsoft’s stock decline and investor skepticism, is whether the $690 billion in announced capex will be deployed.
The gap between announced spending and approved capacity is substantial and widening. Microsoft’s $80 billion unfulfilled Azure backlog explicitly exists because of constraints. Amazon’s $200 billion capex guidance for 2026 exceeded analyst expectations by roughly $50 billion, causing Amazon’s stock to decline 8-10 percent amid concerns about whether the company can make money from infrastructure investments if electricity remains constrained. Google’s chief sustainability officer warned that transmission barriers have become the number one challenge the company faces.
Even accounting for projects that will be withdrawn and requests that are speculative, only a fraction of the $690 billion in announced capex can plausibly be energized within the 2026-2027 planning horizon.
This mathematical reality creates significant investor risk. Companies are committing capital on the assumption that electricity will be available, but deliveries are running three to four years behind demand growth. If constraints prove as intractable as current trends suggest, companies will have deployed substantial capex that generates insufficient returns, either because it cannot be used without electricity or because it must be powered through expensive alternatives like co-located generation or dedicated nuclear plants.
Microsoft’s investor presentation strategy has shifted toward emphasizing the durability of its backlog and the productivity gains that will justify the capex, but the underlying market skepticism reflects rational concern about execution risk in a constrained system.
The Federal Energy Regulatory Commission’s pending rulemaking on large load interconnection, due to complete by April 30, 2026, could substantially change these dynamics. If FERC successfully asserts federal jurisdiction and establishes streamlined interconnection procedures for data centers, approval timelines could potentially be reduced from seven years to twelve to eighteen months. If FERC’s assertion of jurisdiction is successfully challenged in court or if the resulting rules prove ineffective, the fragmented state-by-state approval regime will persist, and constraints will continue to limit deployment.
This regulatory outcome will determine which companies successfully execute their capex plans and which end up with wasted infrastructure investments.
Who Pays for AI Infrastructure
The $690 billion in announced capital expenditure by the five largest tech companies represents a conviction that artificial intelligence will consume unlimited amounts of computing capacity. But that conviction runs directly into a physical and regulatory reality: electrical capacity is not unlimited, and the authority to allocate it is fragmented across federal regulators, state utility commissions, regional grid operators, and local authorities, each operating under different legal authorities and responding to different constituencies.
Federal FERC action could standardize and streamline the approval process, potentially unlocking capacity and accelerating timelines. State Public Utility Commissions are making decisions about cost allocation that determine whether infrastructure is subsidized or borne by the companies directly benefiting. Regional grid operators are establishing operational rules that determine how much flexibility data centers must provide during grid stress. Local governments retain authority over utility franchises and siting decisions that can enable or block projects.
Tech companies are lobbying and negotiating at each level, gaining advantage through established relationships, direct investment in generation, and strategic geographic positioning in less constrained regions.
For ordinary Americans, the outcome of these regulatory and political battles determines whether data center infrastructure is subsidized through higher bills or whether costs are borne directly by companies. It determines whether electricity is allocated to support new industrial growth or protected for existing consumers and critical industries. It determines whether the AI infrastructure boom accelerates or is constrained by grid limits.
The answer emerging from current regulatory practice is that some costs are being spread to everyone through utility rates, while some are being borne by companies themselves. Where the line falls, and how individual states, regions, and regulators draw it, is becoming the hidden architecture of AI competition.
The officials making these decisions—state utility commissioners, FERC board members, regional grid operators—are determining not who gets electricity but who gets to build the future.
Our articles make government information more accessible. Please consult a qualified professional for financial, legal, or health advice specific to your circumstances.