Who Pays for the Large Load?

Dominion's GS-5 class hits January 1, 2027 with 14-year commitments and minimum charges. APS wants collateral and exit fees. The fight over AI power is becoming a fight over other people's bills.

The loudest political attack on data centers is no longer “they use too much power.” It is “someone else will pay for it.” Utilities are answering with tariffs that make large loads own more of their cost. Developers who underwrote 2024-vintage rate assumptions need to read the new fine print.

Dominion Energy Virginia’s GS-5 large-load rate class takes effect January 1, 2027. It targets customers at or above 25 MW with a high load factor — the profile of most data centers in the Dominion footprint. Customers face 14-year service commitments and minimum demand charges: 85% of contracted distribution and transmission demand and 60% of generation demand, per commission materials and Dominion’s own GS-5 report (Dominion GS-5 report; Virginia SCC fact sheet).

That is a deliberate shift from tariffs that let a hyperscaler reserve capacity, ramp slowly and leave residential customers holding stranded generation and wires.

Collateral for loads that never arrive

Arizona Public Service said the quiet part in its IRP. Sustained interest from large customers raises the risk of “unrealized loads.” APS’s mitigation is minimum demand and energy-based contract terms, collateral against default, and early exit fees (Utility Dive).

Entergy’s amended Google package in Arkansas is sold partly as ratepayer protection: at least $500 million more savings than a prior agreement, with Google funding $600 million toward batteries. Meta-related generation in Louisiana will be judged in the same frame — does the hyperscaler pay for the CCGTs, or does the residential class (Utility Dive)?

Texas politics made the slogan electoral. Both Abbott and Hinojosa said Texans should not see higher bills because of data centers. Hinojosa added an attack on tax subsidies. The substantive fight in every commission docket is allocation: interconnection upgrades, new generation, stranded costs when a campus cancels, and whether economic-development riders socialize benefits that accrue to one fence line (Texas Tribune).

Leases are copying the tariff logic

CBRE’s H1 2026 work noted colocation operators experimenting with triple-net structures and “take or pay” power floors. Minimum utilization requirements in the 60%–85% range of allocated capacity mean a tenant can owe for reserved power whether or not servers draw it (Data Center Frontier / CBRE).

That is the private-market rhyme of GS-5 and APS collateral. Everyone who has been burned by phantom load — utilities, landlords, lenders — is writing minimums into the contract. Occupiers that built financial models on flexible ramp and easy contraction will find 2027 paper less forgiving.

Northern Virginia makes the stakes concrete. CBRE put NOVA inventory at 4,496.5 MW with vacancy around 0.24% and just 10.8 MW available. Asking rates for 10 MW-plus requirements ran roughly $160–$185 per kW-month. When the only capacity left is expensive and precommitted, a 14-year utility minimum and a colo take-or-pay floor are not theoretical. They are the price of staying in the market.

How to read a 2027 rate case

Three questions separate theater from tariff:

Does the large-load class require multi-year commitments and minimum billing determinants?
Who funds generation and network upgrades up front — customer contribution, rider, or general rates?
What happens on early exit or under-ramp — exit fees, collateral draw, remarketing?

If the answer to all three still socializes risk onto residential classes, expect a political fight. If the answer assigns risk to the large load, expect higher all-in $/kW for the campus and fewer LOIs that were never going to build.

Economic-development math is getting rewritten

For a decade, states and localities sold data centers with sales-tax exemptions, property-tax abatements and cheap interruptible-looking rate structures. Texas politics turned the subsidy question into a campaign line. Arizona paused new data-center sales-tax exemptions for three years. Illinois paused processing new agreements under its Data Center Investment Program while seeking a broader framework. Virginia’s SCC has published dedicated fact sheets on data-center rate initiatives because the residential bill fight arrived.

The direction of travel is higher direct contribution from the large load and fewer socialized free rides. That raises all-in cost for campuses and should reduce LOIs that were only ever hunting incentives. It is also how utilities and commissions keep political permission to connect the next gigawatt. Developers still pitching 2022-vintage incentive stacks into 2027 models will miss both the tariff and the election.

What to watch

First GS-5 customer agreements and whether NOVA deal velocity changes after January 1, 2027.
APS Corporation Commission treatment of large-load collateral language.
Arkansas and Louisiana dockets on Entergy customer-funded packages.
More state legislatures copying “data centers pay their own way” into statute, not just rhetoric.

Why it matters for builders, buyers and investors

All-in power cost is becoming a function of tariff design, not just energy price. Builders should model GS-5-style minima before bidding land. Buyers should negotiate ramp schedules that match utility minimums so they are not paying twice. For $D and peer utilities, large-load tariffs that stick are a credit positive for other ratepayers and a volume risk for developers who over-reserved. Underwrite the allocation fight; it is not going away.

Not investment advice.

Sources

Report a problem with this page ↗