Large-load tariffs are becoming the data center credit screen

Utility Dive's Sept. 23 report on the widening state push to curb data center speculation puts a number on the pipeline's newest constraint: more than $170 billion in AI data center capacity has been blocked, withdrawn, or stalled by community opposition since January 2024, a figure the energy advisory firm Relae, formerly Carbon Direct, published in June. Set against the roughly $581 billion Goldman Sachs expects hyperscalers to spend in the U.S. on AI infrastructure this year, that is enough damage to matter and not yet enough to break the boom. What utilities and state regulators are doing about it, though, is less the slowdown the backlash has been demanding than a screen on who gets to finance what remains.

Put the two figures on the same clock and the picture sharpens: the blocked capacity accumulates across more than two and a half years while the spend estimate covers one, so annualized, the obstruction works out nearer 11% of hyperscaler spend than the 29% the raw ratio implies. Moratoriums command the coverage — the Utility Dive piece notes they have taken hold in some places — but the measurable policy shift is happening one tariff docket at a time, and tariffs are where the underwriting consequences live.

The backdrop is political as much as technical: community opposition has emerged as a major barrier to data center development, and the other constraints named in the report compound it—construction labor shortages, long lead times for critical electrical equipment, limited power availability in key markets, and uncertainty about consumer demand for AI tools. Utilities spent the second quarter tying project execution to ratepayer protections on their earnings calls, and a large-load tariff is what that promise looks like once it reaches a docket.

The instruments are specific—regulators and utilities have been layering on minimum contract durations, minimum billing demand, collateral requirements, upfront payments for impact studies, exit fees and defined ramp schedules, with some states adding incentives for developers that bring their own generating capacity, accept flexibility in their energy use, or commit to clean energy and economic development goals. Read as a package, the schedule asks one question of every project in the queue: if the load never materializes, who pays? A developer with a signed lease from an investment-grade tenant can answer it; a land assembler holding a site option and a rendering cannot.

That is the useful way to read these tariffs, and it is where rate design and project finance are converging on the same test from opposite directions: the regulator wants to know who pays if the load never shows, the lender who pays if the tenant never shows. Minimum durations and minimum billing demand nudge developers toward anchor-contracted projects of precisely the shape that clears an infrastructure credit committee—the same shape this publication has argued is the only one that earns infrastructure pricing. Utilities are protecting ratepayers, not setting out to shape the capital stack; the effect on the stack is real anyway.

As this publication has argued, grid permission is the underwriting asset; queue positions and interconnection contracts price before electrons do. A tariff that charges for the permission does something the interconnection queue never managed: it puts a carrying cost on an option that previously cost nothing to hold. Speculative requests that once ran to a filing fee and a study deposit now run to a balance sheet, and the developers who cannot post collateral will find that out before they find out whether the site has water.

The second-order consequence lands on the capital hierarchy in digital infrastructure: projects that clear a utility's collateral and duration tests are the same projects that attract anchor-tenant financing, which quietly narrows the pool competing for infrastructure pricing to the ones that deserve it. That is a healthier market than one in which queue positions trade as free options, and it puts pressure on the developers whose business model was assembling land and flipping queue priority rather than energizing load.

None of this resolves the supply side, where the constraints stack: Goldman Sachs said in May that it expects U.S. data center power demand to more than double from 2025 levels to 66 GW in 2027, while only about half the capacity scheduled for the next one to two years comes online on time amid delays and cancellations. The Electric Power Research Institute said in February that data centers will represent 9% to 17% of U.S. electricity demand in 2030 and as much as 20% by 2035. The tariff screen is operating on a pipeline already being thinned by steel, staffing and supplier slots—the delivery constraints this publication has tracked moving out of contractor backlogs and into equipment and raw materials, with staffing the final bottleneck before energization. A developer that clears the tariff may still miss its energization date; one that fails it never gets the chance.

The incentive half of the policy may prove more consequential than the penalties. States are dangling favorable treatment for developers that bring their own capacity, and the behind-the-meter generation market is being priced by data center demand rather than by utility resource plans—hyperscaler contracts, not utility rate cases, set SMR prices today, as this publication has argued. A tariff regime that rewards self-supply accelerates that and sharpens the split between markets that hold firm capacity and those that do not: our own reporting on Pennsylvania's modeling put a 2030 date on PJM's capacity shortfall and made firm capacity and grid position the scarce assets in the interim.

Watch the tariff dockets rather than the moratorium headlines. They are public, dated and priced, and they will identify which projects have a credit behind them long before any financing does. A project that can post collateral, hold a minimum billing demand and pay an exit fee is one a lender can underwrite; the rest are queue positions holding a bill.