When data centers bring their own power, who pays for unused capacity?

A data center can acquire a power bill before its first server earns revenue. The July expansion of the US Ratepayer Protection Pledge commits signatories to fund new supply and supporting infrastructure and pay negotiated costs even when consumption falls short. Power increasingly becomes a financed obligation ahead of operations.^[1]

The change affects siting, customer pre-leasing and construction phasing. It can also create financeable demand for equipment and generation. For cloud customers, the questions are the premium for earlier operation and who continues paying if demand arrives late. Actual economics remain in tariffs and contracts, rather than a uniform price set by the pledge.

Policy allocates cost; contracts set the bill

EPA’s July 23 release recalls that Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed in March, with participation subsequently expanded to states, developers and power providers. Commitments include new supply, delivery upgrades and negotiated payments even below planned consumption.^[1]

The release also describes more than $15 billion of new baseload generation arrangements in PJM. This is a policy program scale, not revenue awarded to an equipment vendor or proof that every project is approved, financed and operating.

Protecting existing customers does not remove the cost of expansion. Data centers can bear it through rent, capacity fees, tariffs or direct investment. The party paying—developer or ultimate tenant—determines the financial exposure.

The pledge should not be treated as one uniformly applicable statutory obligation for every US project. Local regulation, approved tariffs and executed agreements establish enforceable terms. Minimum demand, exit provisions and upgrade allocations need project-specific examination.

Low energy prices do not guarantee a low-cost site

An energy tariff excludes many potential costs: substations, network upgrades, backup capacity, deposits, construction interest and termination payments. A cheap site requiring a long grid wait can leave purchased servers and customer commitments exposed.

Developers need to compare grid delivery with the complete on-site generation schedule: equipment procurement, permits, fuel connection and commissioning. Faster generation may carry higher fuel, maintenance and capital costs. The comparison should cover its full period of use.

A power purchase agreement also does not itself deliver electricity to a campus. Contracted energy, generating assets, transmission and local connection have distinct conditions. Renewable nameplate capacity may require storage or complementary supply to support continuous load.

Phasing creates a trade-off. Reserving all future power ahead of signed customers brings fixed costs forward. Waiting too long can strand completed halls without power. Flexibility is valuable when supply commitments match a credible customer ramp.

Minimum payments change the utilization threshold

Consider a hypothetical 100MW reservation with a $20 million annual fixed capacity charge and energy at $60/MWh. Average use of 100MW produces 876,000MWh a year. The fixed charge adds approximately $22.8/MWh, for total energy cost of $82.8/MWh.

At only 60MW average use, annual consumption is 525,600MWh while the capacity charge remains. Fixed cost rises to $38.1/MWh and total cost to $98.1/MWh. Consumption falls 40%, but unit cost increases about 18.4%. The example illustrates the idle-capacity exposure that can sit with a new large customer.

For compute operators, this can compound idle servers: revenue falls while each productive GPU-hour carries more fixed electricity expense. Customer minimum payments can transfer some exposure. Long power commitments combined with purely usage-based customer billing leave the mismatch with the operator.

Capacity commitments can still be rational when they secure delivery, accelerate construction or reduce financing cost. Their value should be measured against those benefits. Supply reservations and binding customer demand need the same timeline.

A temporary customer delay calls for liquidity; a permanent reduction can require reletting, reduced expansion or termination payments. Annual average utilization alone conceals these different effects on equity cash flow.

On-site gas generation introduces another cost structure

Owning generation reduces dependence on a grid timetable but introduces fuel procurement, emissions controls, maintenance and redundancy. Nameplate output is not guaranteed net power for servers at every moment. Maintenance and auxiliary loads matter.

For illustration, 100MW operating at 80% produces 700,800MWh annually. At an assumed heat rate of 8MMBtu/MWh, a $1/MMBtu increase in gas price adds approximately $5.61 million of annual cost. This is neither a specific machine’s efficiency nor a fuel-price forecast.

A pass-through contract shifts that increase to the customer and supports operator margins. Fixed electricity pricing leaves the operator exposed. Long-term gas procurement can reduce price variability, but contracted volume, transport and counterparties still need to match load.

Equipment suppliers may book sales before the campus earns revenue. Continuous generation also requires dependable parts and field service. Shipments reveal demand early; service income and warranty expense reveal more of the lifecycle return.

The islanded-generation clarification has limits

EPA’s July 16 memorandum addresses the Acid Rain Program treatment of certain grid-isolated generators serving private consumers. It expressly does not make a final determination for a specific source or address other Clean Air Act programs. It is not a blanket permitting exemption.^[3]

Later grid connection can change the analysis, so a project moving from islanded operation to utility connection needs to consider both phases. State requirements may differ and the relevant permitting authority remains important.

Turbines and engines can face different emissions standards, operating conditions and permitting paths. Modeling, controls, operating hours and monitoring can affect cost and timing. Emergency operation under a specific reliability order should not be generalized into authorization for year-round commercial generation.^[2][4]

These conditions belong in project budgets and revenue dates. Missing permits can postpone service; additional controls can raise investment and maintenance. A generic risk premium cannot substitute for examining a project’s actual constraints.

Allocating cost does not eliminate demand risk

Charging large users for dedicated infrastructure can protect other ratepayers from some stranded costs. Someone must still carry demand risk. Minimum payments and deposits place more exposure with the campus and its customers; any remaining utility exposure needs to be explained through contractual and regulatory arrangements.

For a developer, a reservation payment buys access to future supply. It can improve capacity certainty while consuming cash before load arrives. Customer leases aligned with power commitments reduce that pressure. Reserving large amounts first and finding tenants later can turn apparently inexpensive electricity into a long fixed obligation.

Financing amplifies timing differences. If minimum power charges start before data halls are delivered while construction loans accrue interest, two cash obligations can run before any operating revenue. Liquidity reserves must cover that overlap as well as physical cost overruns.

Policy and connection agreements therefore belong in a single cash schedule: payment dates, usable power, delay compensation and rent commencement. Developers that align these dates can convert access into returns. A larger connection request alone does not improve project economics.

Valuation should start with deliverable service rather than requested electrical capacity. Even an approved connection is not equivalent to continuous billable output. Maintenance, reserve requirements, conversion losses and the data hall’s own ramp-up can keep useful service below the headline rating. Multiplying requested MW by every hour of the year can therefore overstate receipts while understating the fixed charge allocated to each saleable unit.

A practical cash model begins with the customer’s load schedule and works backward to the supply and redundancy required. The start-up period should be modeled separately from steady operation. A project can be attractive once fully occupied while still requiring substantial cash to survive the interval between connection payments, equipment acceptance and customer billing.

The same distinction informs the countercase. Faster approval creates value only when it shortens the path to usable supply. If gas delivery, equipment or the customer’s construction becomes the new constraint, the nominal regulatory improvement may not advance revenue. An investor should ask which activity is actually setting the completion date, and whether paying for an alternative power route changes that activity. Otherwise, a new power solution can add cost without bringing forward the first bill.

Converting policy opportunity into returns

Potential beneficiaries include suppliers that can deliver and maintain generation and distribution equipment on time. Growth also requires factory investment and working capital, and higher pricing can be absorbed by components, labor and warranty expense. Demand support does not guarantee margin expansion.

Developers with power, permits and customers aligned may avoid expensive waiting. Their acquisition price still matters. Without transaction valuation or sourced market expectations, power scarcity alone cannot be described as an overlooked investment opportunity.

The strongest counter-case is faster grid expansion, slower customer load growth or on-site costs above the premium customers will pay. Minimum commitments then become fixed burdens and early construction can reduce returns.

Over six to twelve months, minimum-payment contracts, actual committed load, permits and energization dates matter more than national investment totals. Delivering them together makes the revenue start more dependable. Locking in payment before workload arrives increases the capital burden on shareholders.