APS's 45% Rate Hike Signals the End of Cheap Grid Access for AI Operators

The Bill Has Arrived

Arizona Public Service filed a request in June 2025 with the Arizona Corporation Commission proposing a rate increase of more than 45% for extra-large energy users—a category that explicitly covers data centers. This is not a surcharge under study or a think-tank proposal. It is a formal regulatory filing, and the mechanism it establishes is designed to be permanent and self-adjusting.

The utility's argument is straightforward: APS has proposed over a 45% rate increase for extra-large energy users like data centers to help ensure that they continue to pay for their costs of service. APS is not asking data centers to subsidize residential customers. It is asking them to stop being subsidized by them. APS developed a high load factor rate in 2017 to ensure rates paid by these customers reflected the cost to serve them; now, as circumstances and the magnitude of growth have changed, APS is proposing to update this rate so that high load factor users will cover their increasing costs and to ensure residential customers are not subsidizing data centers.

The regulatory logic holds. Between 2023 and 2025, data centers were projected to account for 94% of all growth in energy demand on the APS system. Meanwhile, APS expects its total peak load to jump 40% by 2031, driven almost entirely by data centers along with electric vehicles, advanced manufacturing, and mining operations. Spreading those costs across the entire rate base amounts to taxing residents to fund hyperscaler infrastructure. APS decided to stop.

The Mechanism Matters More Than the Headline Number

The 45% figure makes headlines, but the formula rate proposal embedded in the filing is what should concern operators. With formula rates, APS would evaluate annually whether current rates recover the costs to serve customers, with retail rates adjusted to more appropriately reflect the costs to serve the growth in each customer class; the annual process would verify that residential and other business customers are not receiving cost increases associated with data center growth.

This converts a one-time rate case into an annual repricing cycle. The 45% is the opening bid; the formula rate is the lock. If extra-large customers like data centers cause costs to rise, those costs will be assigned specifically to them. Operators building 10-year depreciation models on today's energy pricing are working with outdated inputs.

The cost drivers are real, not manufactured. APS reports that from 2021 to 2025, overhead wires became 87% more expensive and some transformers are almost 90% more expensive. Today's rates were set based on costs incurred in 2021 and 2022, and costs have risen sharply since then. A utility requesting cost recovery on actual, auditable infrastructure spend is far more defensible before regulators than one chasing speculative markups.

A National Wave

APS is not alone. Utility rate activity targeting large loads has reached levels that make the Arizona case a data point, not an outlier. In 2025, utilities requested states to approve a record $31 billion in rate increases across the country.

Several utilities have already formalized separate cost recovery structures for data centers. Dominion Energy's GS-5 rate class, approved by Virginia's State Corporation Commission in November 2025, is the most comprehensive AI-specific utility tariff enacted to date; it establishes a 14-year contract term for large loads above 25 megawatts, includes demand guarantees covering generation and transmission costs, and creates a separate cost allocation mechanism for data center infrastructure that protects residential and small commercial customers. American Electric Power established a data center tariff in Ohio in July 2025 that requires financial commitments from developers before grid connections are approved, a model that other utilities are now studying.

Texas moved legislatively. In June 2025, the Texas State Senate enacted Senate Bill 6, a package of planning, interconnection, cost-sharing, transparency, and emergency operations reforms, which formalizes ERCOT's Large Load Interconnection Study process and directs the Public Utility Commission of Texas to determine a reasonable share of upgrade costs for new large loads.

The aggregate commercial implications of 77 pending large-load tariff proceedings, each determining cost allocation for data centers in a specific utility service territory, are larger than any individual infrastructure deal that the market covers closely. Operators tracking Nvidia earnings but missing state PUC dockets are ignoring the primary document class determining what their infrastructure will cost.

Winners and Losers

Hyperscalers face real cost increases but have real tools to manage them. Amazon, Google, and Microsoft have publicly committed to covering direct infrastructure costs rather than passing them to ratepayers. Dominion's spokesperson stated that infrastructure costs are allocated based on how much it costs to serve each group of customers, with data centers paying an increasingly large percentage of transmission costs. Firms with long-term power purchase agreements, captive generation, or renewable contracts locked before this repricing cycle are partially insulated. Those without are not.

Mid-tier colocation operators and inference startups are more exposed. A 45% electricity increase on a business where power represents 30 to 40% of operating expense is roughly a 15-point margin compression. That is not absorbed without repricing contracts or cutting elsewhere. Operators who built unit economics on legacy energy rates in markets that have not yet filed rate cases are holding an impaired position.

The Phoenix metropolitan area now ranks second in North America for proposed data center development, making Arizona difficult to abandon. Load is already committed. The question is whether operators locked into Arizona capacity accounted for the right energy cost.

Political Opposition Won't Reverse the Trend

The APS filing faces opposition. Arizona Attorney General Kris Mayes has said her office will oppose the 14% rate hike proposed for residential customers and could ask state regulators to decrease rates. Mayes contends APS could manage with a 3% residential increase. The data center rate itself faces less principled resistance: APS proposed a more than 45% rate increase for these users to help ensure that data centers continue to pay for their costs of service and avoid shifting those costs to families or small businesses. That framing has broad political support, particularly from residential customer advocates.

Judge Charles Hains will review testimonies from more than 30 third parties before presenting a recommendation to the commission for a final vote. The commission is unlikely to approve the filing as written. Utilities rarely get everything they request. But the floor has shifted. A partial approval still reprices data center electricity in Arizona materially and gives every other utility in the country a precedent to cite.

What to Watch

1. Arizona Corporation Commission ruling, expected late 2026. Watch whether the commission approves the formula rate mechanism specifically—that clause has the longest operational life.

2. Virginia SCC implementation of Dominion GS-5. The Dominion GS-5 rate case is in its implementation phase, with the Virginia SCC monitoring compliance and the tariff terms already set for the 14-year contract period. Northern Virginia operators should track every compliance filing.

3. Xcel Energy's large-load tariff in Colorado. The Xcel large load tariff proposal is in active contested proceedings at the Colorado PUC, with evidentiary hearings scheduled through 2026 and a final order expected before year-end. Colorado's outcome will shape Mountain West site selection.

4. FERC RM26-4-000 rulemaking. FERC's RM26-4-000 large-load interconnection rulemaking is in the comment and response phase, with a final rule expected in 2026 or 2027. A federal framework supersedes individual state proceedings.

5. Operator repricing. Watch for colocation providers to revise long-term contracts with energy escalation clauses indexed to utility rate cases rather than CPI. That language shift signals the industry has internalized the new cost structure.