# Cost Allocation The division of a shared infrastructure system's costs among the customers who use it. In electricity this is the question underneath every rate fight, and it is genuinely hard rather than merely contested: a grid is a jointly-used network whose costs are overwhelmingly fixed and shared, and there is no uniquely correct way to attribute a shared cost to individual users. **Scope note.** This page treats cost allocation as it operates in regulated utility ratemaking. The term has separate technical meanings in managerial accounting and in transportation and telecommunications regulation, which follow similar logic but different rules. ## The doctrine: cost causation The governing principle in American utility law is **cost causation** — costs should be borne by those who cause them, roughly in proportion to the burden they impose. It is the standard [[Federal Energy Regulatory Commission]] applies to transmission cost allocation and the standard state commissions apply to retail class allocation. The doctrine is sound and almost never dispositive, because "cause" is underdetermined for a shared network. If a transmission line is upgraded when a new customer arrives, but the line also improves reliability for everyone and would have needed replacement in eleven years anyway, what fraction did the new customer cause? Every method available — peak responsibility, average-and-peak, twelve-monthly-coincident-peak, marginal cost — produces a different defensible answer. Choosing among them is a policy decision performed in the register of accounting. > [!important] The reliable tell > When an argument about fairness is conducted entirely in the vocabulary of methodology, the methodology is doing political work. This is the same structure as [[Metric Capture]]: the contested judgment does not disappear, it relocates into a technical parameter where objecting requires credentials. ## The two directions Load growth can push other customers' rates either way, and both directions are real. **Fixed-cost dilution — rates fall.** A large share of utility cost is fixed and does not rise with sales. Spread over more kilowatt-hours, the fixed cost per unit falls, and rates for everyone can decline. This is not utility propaganda; it is standard economics, and it has empirical support. An EPRI working paper using an instrumental-variables design found that data centers **caused average US retail rates to fall modestly between 2015 and 2024**, with a residential customer in a state whose data-center capacity grew 160% seeing rates roughly 6% lower than the no-growth counterfactual. An LBNL/Brattle review found that states with the largest data-center growth generally saw price decreases from 2019 to 2025. **Incremental cost exceeding average — rates rise.** If the new load requires capacity that would not otherwise be built, at a cost above the existing average, and if the new customer does not pay that increment, the difference lands on everyone else. This is what PJM's capacity market has been doing: the market monitor attributed **$29.4 billion of $63.6 billion (46%)** in capacity charges across four base auctions to data-center load. See [[Capacity Market]]. Both findings can be true because they describe different periods and different mechanisms. The pre-2024 growth was gradual and largely absorbed into existing headroom; the 2025–2028 buildout is fast enough to force new capacity at a premium. Whether this is a regime change or a supply lag that new generation corrects is the live empirical dispute, not a partisan one. ## What actually determines the sign Economists converge on a short list. Whether a large load raises or lowers other customers' rates depends on: 1. **Whether the load pays its true incremental cost of service**, or receives a discounted rate to attract it. Utilities have a structural reason to offer discounts — they earn on capital, and load growth justifies capital — so this risk is not hypothetical. 2. **Whether serving it requires capacity that would not otherwise be built**, or fits within existing headroom. 3. **Whether the load can curtail during the small number of hours that drive most capacity cost.** A flexible load imposes far less system cost than an inflexible one of the same size. See [[Demand Response]]. 4. **Whether a long-term take-or-pay commitment protects against the load disappearing** after the infrastructure is built. Every 2025–2026 regulatory instrument in this space — minimum-take tariffs, capacity commitment charges, credit support requirements, curtailable service, "bring your own capacity" — is an attempt to force one of those four conditions to hold. That is the coherent way to read what otherwise looks like a scatter of unrelated state actions. ## Stranded cost The failure mode that the commitment requirements exist to prevent. If a utility builds for a customer that never arrives or later leaves, the sunk cost remains in the rate base and is recovered from whoever is left. The mechanism is not speculative: Wisconsin ratepayers owe close to a billion dollars for already-shuttered plants, with charges running until 2039 — decades after generation stopped. Applied to the AI buildout, the exposure is a function of how much announced capacity is real. Credit analysts have flagged rising state opposition and moratoriums as a factor that could increase cancellation rates *after* infrastructure commitments are made. See [[Interconnection Queue]] for the gap between announced and built. ## The rhetorical asymmetry One further point worth holding. The benefits of a data center are concentrated, nameable and announceable: a county gets an investment figure, a jobs number, a ribbon-cutting. The costs, if they materialize, are diffuse, delayed, and arrive as a slightly larger bill with no attribution on it. That asymmetry is the ordinary shape of a distributive fight in which one side is organized and the other is not, and it predicts the outcome better than any argument about the merits. ## See Also - [[Rate Case]] — the proceeding where allocation is decided - [[Capacity Market]] — the wholesale mechanism that moves cost the fastest - [[Large Load Tariff]] — the instrument built to make large loads pay incremental cost - [[Public Utility Commission]] — the venue