Every quarter, the analysts who cover the AI infrastructure market produce detailed models of hyperscaler capex, GPU shipment rates, data center construction pipelines, and neocloud operator economics. Those models are sophisticated, well-sourced, and widely read by the institutional investors who allocate capital to AI infrastructure. They share a common blind spot. They do not systematically track utility rate cases, the regulatory proceedings filed with state Public Utilities Commissions that determine the power costs, interconnection timelines, cost allocation frameworks, and generation investment requirements that govern every AI data center’s operational economics for the next ten to fifteen years.
A utility rate case is not a press release or an earnings call. It is a formal regulatory filing, typically hundreds of pages, that lays out a utility’s proposed rate structure, its capital investment plans, its load forecasts, and its requests for regulatory approval of cost recovery mechanisms. It is processed through a public docket, subject to intervention by consumer advocates and commercial customers, and ultimately resolved through a PUC order that has the force of regulatory law in the utility’s service territory. The AI infrastructure investor who reads Moody’s capex forecasts and Nvidia earnings transcripts but does not read Dominion’s GS-5 rate case, AEP Ohio’s data center tariff filing, and Xcel Energy’s large load tariff proposal is modelling AI infrastructure economics without the documents that most directly determine what AI infrastructure will cost to operate over its useful life.
The Rate Cases That Are Already Reshaping Infrastructure Economics
The volume of utility rate case activity specifically addressing large data center loads has accelerated dramatically over the past 18 months. In 2025, state regulators approved 29 large-load tariffs and 77 are pending in 36 states, according to the Smart Electric Power Alliance tariff database. Each of those tariffs represents a regulatory determination about how the cost of serving large data center loads will be allocated between the data center customer and the broader ratepayer base. The aggregate commercial implications of 77 pending large-load tariff proceedings, each determining cost allocation for data centers in a specific utility service territory, are larger than any individual infrastructure deal that the market covers closely.
Dominion Energy’s GS-5 rate class, approved by Virginia’s State Corporation Commission in November 2025, is the most comprehensive AI-specific utility tariff enacted to date. It establishes a 14-year contract term for large loads above 25 megawatts, includes demand guarantees covering generation and transmission costs, and creates a separate cost allocation mechanism for data center infrastructure that protects residential and small commercial customers from absorbing costs attributable to data center load additions. The 14-year contract term is the most commercially significant feature. It means that a data center developer who signs a GS-5 agreement with Dominion is committing to a specific cost structure and interconnection arrangement for the period from 2026 to 2040.
The long-term cost certainty that provides has real value for financial models that extend over multi-year infrastructure investment periods. The obligation it creates, paying demand charges regardless of operational status for 14 years, has real downside risk for developers who over-commit relative to their actual deployment timelines.
The Colorado Tariff That Could Redefine Cost Allocation
Xcel Energy’s large load tariff proposal, filed April 2, 2026 with the Colorado Public Utilities Commission, follows the AEP Ohio model in requiring data centers consuming 50 megawatts or more to cover the full cost of transmission lines, substations, interconnection upgrades, and new electric generation capacity needed to serve them. The proposal includes 15-year contracts with minimum monthly bill requirements and exit penalties equal to the sum of remaining minimum monthly bills, creating the same long-term commitment structure as Dominion’s GS-5 but applied to the Colorado market where Xcel is projecting that large-load customers will represent two-thirds of its new electricity demand over the next five years.
The Colorado PUC must approve the proposed tariff, which will be the subject of contested regulatory proceedings involving Xcel, consumer advocates, and data center developers, and the final tariff language will determine the economics of AI data center development in Colorado for the next generation.
The AEP Ohio Model That Others Are Replicating
AEP Ohio’s large load rate class, created in July 2025, was the first utility tariff to require data centers to post financial assurance covering the cost of grid upgrades before their interconnection requests are processed. The AEP Ohio model requires that data centers cover 85% of their request demand in minimum monthly charges regardless of actual usage, eliminating the risk that a data center developer secures grid capacity and then fails to deploy load at the committed level, leaving the utility with stranded infrastructure investment. The model has become the reference point that utilities in other states are adapting as they develop their own large-load tariff proposals.
The adoption trajectory matters for AI infrastructure investment because it means the cost allocation principle, data centers pay for the infrastructure they require rather than socialising those costs across ratepayers, is becoming the regulatory standard across multiple utility service territories simultaneously. A data center operator who modelled infrastructure economics in 2023 on the assumption that grid upgrade costs would be distributed across the ratepayer base is facing a regulatory environment in 2026 where that assumption is incorrect in Ohio, Virginia, Colorado, and an expanding list of states as PUCs adopt the AEP Ohio framework or variants of it.
The magnitude of the cost differential between the 2023 assumption and the 2026 reality depends on the specific transmission and distribution upgrade requirements of each data center’s interconnection, which vary significantly by location and load profile. But the direction is consistent: AI data center operating costs are higher under the reformed tariff frameworks than under the frameworks they are replacing, and the developers who have not updated their economic models for the reformed frameworks are operating with materially incorrect cost assumptions.
The Resource Plans That Determine Infrastructure Availability
The utility rate case is the mechanism for approving specific rate structures. The utility resource plan is the document that determines how much generation capacity, transmission infrastructure, and distribution capacity the utility will build over the next ten to twenty years, and therefore how much data center load the utility can actually serve within its service territory. Resource plans are formal regulatory filings that are subject to PUC review and approval, and the approved resource plan establishes the binding commitment that the utility has made to its service territory regarding future infrastructure investment.
Dominion Energy’s 2024 resource plan projects nearly 27 gigawatts of new generation capacity by 2039, including 21 gigawatts of renewable energy and 5.9 gigawatts of gas, supported by transmission expansion investment that is the largest in the utility’s history. The Dominion resource plan is the binding commitment that underlies every data center development in Northern Virginia that depends on Dominion power service. If the resource plan’s generation and transmission timelines are met, the data centers planned for Northern Virginia have the power supply they need to come online on schedule. If the resource plan’s timelines slip, for regulatory reasons, construction delays, or permitting challenges, the data centers depending on Dominion power service face delays that no amount of developer capital can resolve, because the constraint is the utility’s regulatory build schedule rather than the developer’s capital access or construction capability.
The Information Gap Most Investors Ignore
The investor community tracks Dominion’s stock and earnings. It does not, as a general matter, track Dominion’s resource plan docket at the Virginia State Corporation Commission with the same rigour. But the resource plan docket is where the supply-side constraints on Northern Virginia data center development are first visible, because the resource plan reveals the generation and transmission investment timeline that determines when new capacity will be available. An investor who reads every Nvidia earnings transcript and every data center developer earnings call but does not track the Dominion resource plan docket is modelling AI infrastructure capacity with a significant information gap about the single most important binding constraint on Northern Virginia data center delivery.
The Xcel Rate Case That Sets the Colorado AI Infrastructure Market
Xcel Energy filed a general electric rate case in November 2025 requesting approval to recover costs of its large-scale capital investments through rate increases, with a procedural schedule extending through 2026. The rate case covers Xcel’s overall rate structure, including but not limited to data center-related investments. At the same time, Xcel filed the separate large load tariff proposal in April 2026 that addresses cost allocation specifically for large data center customers. The interaction between the general rate case and the large load tariff in the Colorado PUC proceedings will determine both the absolute level of electricity rates that all Colorado customers pay and the specific cost allocation between data center customers and other ratepayers.
The Colorado proceedings matter for the AI infrastructure investment community because Colorado is a secondary data center market that multiple hyperscalers and neocloud operators have been evaluating as an alternative to the capacity-constrained Northern Virginia and Silicon Valley markets. If the Colorado PUC resolves Xcel’s general rate case with rate increases that make Colorado electricity costs materially more expensive for all customers, and simultaneously approves the large load tariff with full-cost recovery requirements for data centers, Colorado becomes less competitive as a data center market relative to states with more favourable utility rate environments. The site selection decisions that AI infrastructure developers are making today for capacity needed in 2027 and 2028 should be informed by the regulatory proceedings that will determine Colorado’s long-term utility rate environment, and those proceedings are available in the Colorado PUC’s e-filing system for any investor who chooses to read them.
The FERC Proceedings That Supersede State Rate Cases
The utility rate case at the state PUC level is the most directly accessible regulatory document for AI infrastructure investment analysis, but it operates within a federal regulatory framework set by FERC that can supersede state-level cost allocation decisions in areas of federal jurisdiction. The relationship between FERC proceedings and state PUC proceedings on AI data center cost allocation is one of the most technically complex and commercially significant areas of AI infrastructure regulation, and the investment community’s failure to track it systematically is one of the most significant analytical gaps in the AI infrastructure market.
FERC has jurisdiction over wholesale electricity markets, including the capacity markets and energy markets that determine how much data centers pay for grid power at the wholesale level. State PUCs have jurisdiction over retail electricity rates, including the distribution charges and utility service costs that residential and commercial customers pay. AI data center costs at the grid level sit at the intersection of both: the capacity market costs that flow from data center load additions are federally regulated, but the retail rate design that passes those costs to customers is governed by state regulators. FERC’s December 2025 order directing PJM to develop co-location tariff rules and its ongoing proceeding RM26-4-000 on large-load interconnection are both federal proceedings whose outcomes will supersede state-level cost allocation approaches in the federally regulated portions of the grid cost structure.
The Three Regulatory Layers Investors Must Track
An AI infrastructure developer operating in the PJM footprint faces cost allocation decisions at three separate regulatory layers: the FERC wholesale capacity and energy market layer, the PJM tariff layer governing transmission and interconnection services, and the state PUC retail rate layer governing distribution service and local generation costs. The rate cases that the investment community does not read, the state PUC proceedings, address only the state retail layer. The FERC and PJM proceedings that address the two federal layers are even less systematically tracked by the AI infrastructure investment community, despite being the layers where the largest cost allocation decisions are being made.
The Monitoring Analytics Q1 2026 state of the market report documented $13 billion in added costs distributed across PJM ratepayers from data center load additions, a figure that was determined through PJM capacity auction mechanics governed by FERC-approved tariff rules. The state PUC proceedings are the downstream mechanism for allocating some of those costs to data center customers rather than residential ratepayers. The FERC proceedings are where the size of the cost pool is determined.
The Intervention Rights That Data Centers Are Not Using
One of the most practically significant aspects of utility rate case proceedings that the AI infrastructure investment community consistently underutilises is the intervention right that commercial customers have in proceedings that affect their rates. A data center operator whose electricity costs will be materially affected by the outcome of a state PUC rate case has the right to intervene in that proceeding, file comments on the proposed tariff, present evidence on the cost allocation methodology, and advocate for a regulatory outcome that reflects the data center operator’s commercial interests. Utility lawyers know this. AI infrastructure investment managers typically do not.
The intervention right creates a direct path from regulatory engagement to competitive advantage. A data center developer who intervenes in Dominion’s GS-5 rate case and advocates successfully for cost allocation methodologies that more accurately reflect the cost causation of different load types is not just engaging in regulatory policy. It is reducing its future operating costs relative to competitors who did not intervene and who will be subject to the same tariff at the same rates. The cumulative value of regulatory intervention across multiple state PUC proceedings, multiplied by the volume of data center capacity being developed, is a multi-billion dollar opportunity that is available to every major AI infrastructure developer and that is being systematically underutilised because the AI infrastructure industry does not have the same regulatory engagement infrastructure that incumbent utilities, consumer advocacy groups, and industrial manufacturers have built over decades.
The Load Forecast Battle That Determines Infrastructure Capacity
The most consequential section of any utility resource plan from an AI infrastructure perspective is the load forecast, the utility’s projection of how much new demand will be added to its service territory over the planning period, broken down by customer class, geographic area, and timing. The load forecast determines how much new generation and transmission the utility plans to build, which in turn determines whether the capacity will be available when data center developers expect it to be.
AI data center load is unique among utility customer classes in its ability to create large, rapid changes in the load forecast that utility planning models were not designed to accommodate. A hyperscaler that announces a 500-megawatt data center campus in a utility’s service territory has, with a single announcement, added more load to the utility’s planning model than the entire commercial and industrial sector of a mid-sized US city. The utility’s resource plan, approved by the PUC two years ago based on load forecasts that did not include the hyperscaler announcement, is now fundamentally inadequate for the load it will actually be required to serve.
Why Utilities Need New Tariff Structures
The utility must file a revised resource plan with the PUC, obtain approval for the additional generation and transmission investment, procure the equipment, and complete the construction, all on a timeline that is compressed relative to the normal regulatory and construction cycle by the commercial urgency of the data center developer who wants to be operational as soon as possible.
At least 36 utilities have adopted large-load tariffs specifically to manage the load forecast uncertainty that AI data center announcements create. The tariffs require data centers to post financial commitments before grid capacity is reserved, which gives utilities the revenue certainty needed to finance resource plan revisions without exposing other ratepayers to the cost of infrastructure that may never be needed if the data center project fails to proceed. The tariff mechanism is the utility’s tool for managing load forecast uncertainty in the AI era. The PUC proceedings where those tariffs are approved are the regulatory infrastructure for the entire AI data center capacity pipeline, and the investment community that tracks them will understand the capacity pipeline with materially greater fidelity than the community that does not.
The Information Advantage Utilities Already Possess
The load forecast battle is also where the most consequential information asymmetry in the AI infrastructure market exists. Utility load forecasters who are receiving interconnection requests, power purchase inquiries, and tariff applications from data center developers have a view of the forward pipeline of AI infrastructure development that is more current and more complete than what any analyst or investor can compile from public announcements and earnings calls. The utilities are receiving this pipeline information months or years before it becomes public, because data center developers must approach utilities to understand interconnection feasibility and cost before they can make public commitments.
The utility resource plans and load forecasts that are filed with state PUCs are the mechanism through which some of that pipeline information becomes publicly available, and reading them systematically is the most direct way to access the intelligence that utilities have accumulated from the AI infrastructure development community before that intelligence is reflected in public markets.
What Systematic Rate Case Tracking Would Reveal
The AI infrastructure investment community that builds a systematic practice of tracking utility rate cases, large load tariff proceedings, and resource plan dockets across the primary and secondary data center markets would have earlier and more reliable intelligence on the following questions than the community that relies exclusively on the public announcements, press releases, and earnings calls that currently form the primary information set. First, which service territories are developing cost allocation frameworks that will make AI data center development more expensive over the medium term, and which are developing frameworks that provide cost certainty that supports long-term infrastructure investment decisions. Second, which utilities are committing to generation and transmission investment timelines that support the AI infrastructure build schedules that developers have committed to, and which have resource plans that are more aggressive than their regulatory approval timelines can support.
Third, where the interconnection queue positions that data center developers have secured are supported by approved transmission investment that will actually be built, and where the interconnection agreements are contingent on transmission investments that have not yet received regulatory approval and may not on the timelines the queue position assumes. The gap between the 12 gigawatts of US data center capacity slated to come online in 2026 and the roughly 5 gigawatts actually under active construction reflects, in part, the outcome of regulatory proceedings that the developers did not fully account for in their original project timelines. Utility rate cases and resource plan approvals are the regulatory proceedings most directly responsible for that gap, and systematic tracking of those proceedings would have provided earlier warning of the delay risk that is now materialising in the 2026 delivery pipeline.
How to Read a Rate Case for AI Infrastructure Intelligence
The practical challenge that prevents systematic rate case tracking from being standard practice in the AI infrastructure investment community is not the availability of the information. Every PUC docket is publicly available, often in searchable electronic databases like the Colorado PUC’s e-filing system or Virginia’s SCC e-cases portal. The challenge is the volume and complexity of the filings, which are written in regulatory language designed for utility lawyers and PUC staff rather than for infrastructure analysts and investment managers.
The sections of a rate case filing that are most directly relevant to AI infrastructure investment are the large load tariff proposals and their cost allocation methodologies, the revenue decoupling provisions that determine how the utility recovers costs when large customers reduce consumption, the capital recovery schedules that reveal how quickly the utility’s new infrastructure investment will be reflected in customer rates, and the load forecast assumptions that determine whether the utility’s resource plan is sized adequately for the data center load projections that developers are counting on. Those sections typically represent a fraction of the total filing volume, are written in relatively accessible regulatory terminology, and contain the specific intelligence that infrastructure investors need to understand the regulatory environment they are investing into.
The rate case is not a supplementary document that adds colour to the standard investment analysis. For AI infrastructure, it is the primary document that determines whether the standard investment analysis is using the right cost assumptions. The infrastructure investment community that treats it as such will have a material analytical advantage over the community that continues to treat it as background.
The Regulatory Calendar That Infrastructure Investors Should Be Tracking
The specific regulatory proceedings that have the most direct near-term implications for AI infrastructure economics are identifiable and their calendars are publicly available. The Dominion GS-5 rate case is in its implementation phase, with the Virginia SCC monitoring compliance and the tariff terms already set for the 14-year contract period. The Xcel large load tariff proposal is in active contested proceedings at the Colorado PUC, with evidentiary hearings scheduled through 2026 and a final order expected before year-end. PJM’s backstop reliability auction framework is under development with stakeholder proceedings active in 2026. FERC’s RM26-4-000 large-load interconnection rulemaking is in the comment and response phase, with a final rule expected in 2026 or 2027.
The 56 million Americans already affected by utility rate hike approvals from 2025 are bracing for additional increases as utilities file new rate cases to recover AI-related infrastructure investments. Those rate increases are the residential consequence of the same regulatory proceedings that determine AI data center cost structures. The AI infrastructure investor who tracks the rate case calendar is simultaneously tracking the ratepayer impact story that is driving the political momentum behind 78 data center moratoriums and the community opposition that is the most significant non-technical risk to the AI infrastructure buildout.
Regulatory Filings Are Strategic Intelligence
The regulatory proceedings are not just financial intelligence. They are the primary mechanism through which the infrastructure buildout and the community opposition to it will be mediated, and understanding that mediation process from the inside, through the actual docket filings rather than the press coverage of their outcomes, is the analytical capability that will most differentiate the infrastructure investors who navigate the next phase of the buildout successfully from those who continue to be surprised by regulatory outcomes they did not see coming.
The US electricity market not being built for AI documented, the regulatory frameworks are changing faster than the market currently prices them. The utility rate case is where that change is documented, contested, and ultimately determined. Reading it is not optional for serious AI infrastructure investors. It is the most important habit they are not yet developing at scale.
