Brief
The mechanism starts in the wholesale capacity market. Grid operators like PJM Interconnection run forward capacity auctions roughly three years ahead of a delivery year, buying enough generation and demand-response commitments to meet a projected reliability requirement. Every megawatt of forecast peak demand — including forecast data-center load — pushes that target up, and when available supply is tight relative to the target, the auction clears at a much higher marginal price that every buyer pays, not just the buyer whose demand caused the shortage. PJM's independent market monitor, Monitoring Analytics, has attributed roughly 38% of the $16.4 billion cleared in the region's most recent base residual auction, or about $6.3 billion, to data-center-related demand, and roughly 46% of $63.6 billion in the last four auctions combined. Monitoring Analytics itself flags that this attribution is an estimate rather than a published, PJM-endorsed method, since splitting a systemwide price increase across customer classes depends on load-forecast assumptions that aren't yet public.
That auction-cleared cost then has to be recovered from someone, and that's where retail rate design takes over. State utility commissions review a utility's total cost of service — power plants, transmission, distribution, operating expenses — and allocate it across customer classes (residential, commercial, industrial) using a cost-allocation study. A widely used method, coincident peak demand, assigns cost shares based on how much each class uses during the hour the whole system peaks. Because large data centers can potentially forecast and curtail load during system peak hours — a flexibility crypto-mining operations in Texas have already demonstrated — a facility that draws enormous power nearly around the clock can still show up as a small contributor to peak demand and be allocated a correspondingly small share of the capacity and transmission costs its own connection required.
Transmission cost allocation is a separate, ongoing fight layered on top of both of the above. In Virginia, Dominion Energy's Rider T-1 currently recovers roughly $1.5 billion in transmission costs by spreading them broadly across rate classes using a 12-coincident-peak method; the State Corporation Commission's own staff attorney has called this a 'glaring cross-class subsidization' benefiting new large-load customers, and the Commission faced an August 1, 2026 deadline to decide whether to adopt a stricter 'but for' cost-causation standard, require upfront contributions-in-aid-of-construction, or shift to a summer/winter peak-and-average method — with Dominion's own estimate of the residential bill impact swinging from $2.90 to $0.94 per month for a typical 1,000 kWh customer depending on the outcome.
The fourth layer is the special large-load tariff, which several states have now approved specifically to force data centers to internalize risk before the other three mechanisms even engage. Virginia's GS-5 tariff, taking effect January 1, 2027, requires any customer drawing 25 MW or more to sign a minimum 14-year contract, pay at least 85% of contracted transmission and distribution demand and 60% of contracted generation demand whether or not it uses the power, and post collateral of $1.5 million per MW of reserved capacity. AEP Ohio's comparable tariff, approved in July 2025, uses an 85% take-or-pay minimum, a load-ramp period plus at least eight years of commitment, and credit-based collateral, and the utility reports its speculative large-load pipeline fell from roughly 30 GW of preliminary interest to about 13 GW once real financial commitments were required.
What makes this a genuinely fragmented system rather than a single national policy is that FERC governs the wholesale capacity-market rules and interstate transmission cost allocation, while each state utility commission separately governs retail rate design and large-load tariffs — meaning a data center's actual cost exposure depends heavily on which utility territory and which state PUC it lands in, and the same MW of new demand can be treated completely differently in Ohio versus Virginia versus a state that hasn't yet adopted any large-load tariff at all.
Components (7)
PJM Interconnection base residual auction
Procures generation and demand-response capacity roughly three years ahead of a delivery year to meet a systemwide reliability requirement; the marginal clearing price is paid to all cleared resources and recovered from all load-serving entities.
Monitoring Analytics (PJM's independent market monitor)
Publishes post-hoc estimates attributing a share of auction cost increases to data-center load growth, though it acknowledges the attribution method is not yet public or PJM-endorsed.
State utility commissions (e.g., Virginia SCC, Public Utilities Commission of Ohio)
Conduct cost-of-service studies, approve rate-class allocations, and rule on utility-proposed large-load tariffs and transmission cost-recovery riders.
Coincident peak demand allocation methodology
Assigns each customer class's share of capacity and transmission costs based on usage during the hour(s) the whole system peaks, which can let a highly flexible large load minimize its allocated cost share even if its total energy use is very high.
Transmission cost-recovery riders (e.g., Dominion's Rider T-1)
Recover the cost of new transmission infrastructure from ratepayers; the allocation method (12CP versus summer/winter peak-and-average versus a 'but for' causation standard) determines how much of that cost data centers versus households absorb.
State-approved large-load tariffs (Virginia GS-5, AEP Ohio Data Center Tariff)
Impose minimum-term contracts, take-or-pay minimum billing, and upfront collateral on qualifying large loads specifically to shift stranded-cost and buildout risk away from other ratepayers before it ever reaches the capacity market or rate-of-service allocation.
FERC
Approves and oversees interstate wholesale capacity-market rules and transmission cost-allocation tariffs, setting the outer boundary within which state commissions and utilities design retail-level allocation.
How It Works (8 steps)
1Utility submits long-term load forecast to grid operator
A utility or load-serving entity forecasts future peak demand, incorporating known and speculative data-center interconnection requests, and submits this to the regional grid operator for capacity planning purposes.
UtilitiesData-center developers submitting interconnection requestsPJM Interconnection
Why this step: The capacity auction's reliability target is built directly from this forecast; if the forecast is wrong (as PJM's market monitor has flagged given the 'extreme uncertainty' in data-center load projections), the auction procures the wrong amount of capacity at the wrong price.
2Grid operator sets the reliability requirement
PJM translates the aggregated load forecast into a target reliability requirement (a reserve margin above expected peak demand) that the capacity auction must procure to avoid blackouts.
PJM Interconnection
Why this step: Without a binding target, there is no basis for the auction to determine how much capacity to buy.
3Base residual auction clears at a marginal price
Generators and demand-response resources submit competitive offers; PJM accepts offers in order of price until the reliability requirement is met (or the auction hits its price cap/collar), and every accepted resource is paid the last, highest accepted offer price — the marginal clearing price.
GeneratorsDemand-response providersPJM Interconnection
Why this step: This uniform clearing price is what allows the market monitor to observe that tighter supply-demand balance — driven partly by data-center load growth — produced a clearing price far above prior years' levels, and PJM's own reporting shows the 2028/2029 auction cleared at its $325/MW-day cap while still falling roughly 6,831 MW short of the reliability requirement.
4Capacity cost is billed to all load-serving entities
The aggregate capacity cost (roughly $16.4 billion in PJM's latest auction) is billed to every load-serving entity across the market footprint based on their share of system load, not isolated to the specific customers whose demand growth drove the price up.
PJM InterconnectionLoad-serving entities (utilities and retail suppliers)
Why this step: Because the auction is systemwide, a price increase attributable to demand growth concentrated in one utility's territory (like Dominion's Virginia service area) is still spread across all 13 states and the District of Columbia PJM serves.
5State commission allocates utility's total cost of service across rate classes
The state utility commission reviews the utility's full cost of service — including its share of the wholesale capacity bill, transmission investment, and operating expenses — and assigns cost shares to residential, commercial, industrial, and large-load rate classes using a cost-allocation study.
State utility commissionsUtilitiesIntervenors (consumer advocates, tech companies, environmental groups)
Why this step: This is the step where the 'who actually pays' question gets decided at the retail level, and different allocation methods (12 coincident peak versus summer/winter peak-and-average versus a stricter 'but for' causation standard) produce materially different bill outcomes, as the Virginia SCC's Rider T-1 proceeding illustrates with residential impact estimates ranging from $0.94 to $2.90 per month depending on method.
6Large-load tariff imposes contract terms on qualifying data centers
Data centers above a megawatt threshold (25 MW in Virginia and Ohio) are automatically placed into a special rate class requiring a long-term contract (14 years in Virginia, at least eight years past ramp-up in Ohio), minimum take-or-pay billing (85% of contracted transmission/distribution demand and 60% of generation demand in Virginia; 85% of contracted capacity in Ohio), and upfront collateral ($1.5 million per MW in Virginia).
State utility commissionsUtilities (Dominion Energy, AEP Ohio)Qualifying data-center customers
Why this step: This step exists specifically to prevent the 'stranded cost' scenario where a utility builds substations and transmission for a speculative data-center project that is later delayed, downsized, or canceled, leaving other ratepayers to cover the sunk investment; AEP Ohio reports its large-load pipeline fell from roughly 30 GW of preliminary interest to about 13 GW once the tariff required binding financial commitments.
7Transmission-specific costs are litigated separately
Because transmission investment (like Dominion's roughly $1.5 billion Rider T-1 recovery) is often reviewed in a separate proceeding from the base rate case, the transmission-specific allocation question — whether costs are assigned by a 'but for' causation standard tied to the customer that triggered the investment, or spread broadly via a peak-demand formula — is fought out on its own regulatory timeline.
State utility commission staffUtilitiesData-center customers and their counselConsumer and environmental advocates
Why this step: Transmission upgrades are large, lumpy, multi-year investments whose cost-causation is easier to trace to a specific large customer than diffuse generation capacity costs, making this the proceeding where 'cross-class subsidization' claims are most directly litigated — as in the Virginia SCC case where staff counsel argued a 'glaring cross-class subsidization' persists regardless of which allocation method is ultimately chosen.
8Bill impact lands on the residential ratepayer
After all four layers resolve, the net effect shows up as a line-item increase on a household's monthly bill, reflecting whatever share of capacity, retail, and transmission cost was not absorbed directly by the data-center rate class or tariff.
UtilitiesResidential and commercial ratepayers
Why this step: This is the observable end point analysts and journalists track — for example, one industry estimate has projected household bill impacts averaging around $70 per month above pre-boom levels in PJM's territory by 2028, though this figure comes from an advocacy analysis rather than a regulator-adopted forecast and depends heavily on which allocation rules are ultimately adopted in each state.
What Makes It Work
Marginal-price clearing in the capacity auction
Every accepted capacity resource is paid the same clearing price set by the last (most expensive) accepted offer, so a supply-demand imbalance driven by data-center growth in one part of the region can push the price paid by every load-serving entity in the whole 13-state market higher, not just the price paid for the specific new load.
Coincident peak demand cost allocation
Retail cost allocation based on usage during system-peak hours rewards demand flexibility; a data center that can shift or curtail load during the coincident peak can be assigned a small share of capacity and transmission costs even while consuming enormous amounts of power at other times.
Take-or-pay minimum billing with collateral
By requiring large-load customers to pay for a high minimum share of reserved capacity (regardless of actual use) and post substantial upfront collateral, utilities convert speculative interconnection requests into financially binding commitments, filtering out projects unlikely to be built and reducing the pool of stranded-cost risk that would otherwise fall on other ratepayers.
'But for' cost-causation standard
This proposed transmission cost-allocation principle assigns infrastructure costs directly to the specific customer whose demand triggered the investment, rather than spreading them broadly across a peak-demand formula — the central methodological fight in Virginia's Rider T-1 proceeding.
Where It Breaks (5)
Load forecast uncertainty distorts the capacity auction
Consequence: PJM's own market monitor has described the uncertainty in data-center load forecasts as unique and unprecedented, meaning the reliability requirement — and therefore the cleared price and total cost — may be based on demand that never fully materializes, or may understate real demand if forecasts lag actual buildout.
Safeguard: PJM has moved to more frequent forecast updates and accelerated a backstop reliability auction (planned for as early as September 2026) to procure the shortfall separately, but this remains an evolving, not fully tested, fix.
Flexible large loads game coincident-peak allocation
Consequence: A data center or crypto-mining operation that predicts and avoids system peak hours can be assigned little to no cost through coincident-peak-based allocation even while using large amounts of power at other times, effectively shifting cost recovery onto less flexible customer classes.
Safeguard: Some states are moving toward alternative allocation methods like summer/winter peak-and-average, though this shift is still contested and not yet adopted in most jurisdictions.
Stranded transmission and generation investment
Consequence: Utilities may build substations, transmission lines, or contract for generation based on speculative data-center plans that are later delayed, downsized, or canceled, leaving the sunk cost to be recovered from remaining ratepayers.
Safeguard: Large-load tariffs with minimum-term contracts, take-or-pay billing, and upfront collateral (Virginia's $1.5 million per MW, Ohio's creditworthiness and collateral requirements) are designed specifically to shift this risk onto the data-center customer, though these tariffs are new enough that their durability under real project cancellations is untested.
Jurisdictional fragmentation produces inconsistent outcomes
Consequence: Because FERC governs wholesale capacity markets and interstate transmission tariffs while state commissions separately govern retail rate design and large-load tariffs, an identical data-center project can face very different cost exposure depending on which state and utility territory it locates in, creating both regulatory arbitrage risk and public confusion about who is actually responsible for bill increases.
Safeguard: None systemic; some convergence is occurring as states observe each other's tariff designs (roughly 23 states had approved some form of large-load tariff as of mid-2026), but there is no unified national standard.
Auction price caps mask underlying scarcity
Consequence: When the capacity auction clears at an administratively imposed price cap or collar (as occurred in PJM's 2028/2029 delivery year auction), the reported clearing price may understate the true scarcity value of capacity, and PJM's own simulation suggested the uncapped price would have been substantially higher, meaning reported cost figures may not reflect the full economic signal of the supply-demand imbalance.
Safeguard: PJM's stakeholder process periodically reviews and adjusts caps and collars, but this is a market-design tradeoff between price stability and investment signal accuracy rather than a fix that eliminates the underlying scarcity.