Four jurisdictions, four approaches: an unapproved Ohio settlement, a TVA board-set rate, a contested FERC proceeding in Wisconsin, and a Texas audit of its large-load queue.
Microsoft is not disputing that it should pay for the transmission lines being built to serve its data-center campus in Mount Pleasant, Wisconsin. It is disputing the instrument that decides how much.
On August 21, the company filed a protest at the Federal Energy Regulatory Commission against the cost-allocation agreements American Transmission Company submitted on July 24: a Minimum Transmission Charge Agreement plus four amended large-load project agreements covering roughly $600 million of approved lines and substations. The agreements were "negotiated by and between ATC and WEPCo (two affiliated entities) without any opportunity for participation or input from Microsoft," the protest says, and their shortcomings are "systemic." Microsoft asks FERC to set "the entirety of the Agreements" for hearing and settlement-judge procedures or, failing that, to reject them without prejudice for renegotiation and refiling.
The customer being billed for the wires says the bill is built wrong. It isn't alone. The Citizens Utility Board of Wisconsin filed a limited protest the same day. Wisconsin's utility regulator warned a week earlier that the cost-allocation method fails "to prevent cost shifting to other wholesale transmission customers," and the state's industrial energy group filed against the "piecemeal," project-specific approach. In FERC docket ER26-3265, the customer, the consumer advocate, the industrial group, and the state regulator have all told the commission the pricing instrument mismeasures the cost of serving what CUB calls "Wisconsin's first hyperscale data center customer." They disagree about which direction.
For most of the past two years, the energy story around the AI and supercomputing buildout has been a supply question: whether generation and interconnection could be built fast enough. That is the story of a 2031 delivery slot for a gas turbine, of acquirers valuing grid-connected megawatts as the scarce asset, of contracted power emerging as the buildout's real moat, and of FERC's fast lane for large-load interconnection arriving with a curtailment clause attached.
Between August 14 and August 26, a different question came to a head in four jurisdictions: once the wires are built, who is bound to pay for them, on what measuring stick, and for how long. The four proceedings use four different approaches — a negotiated settlement awaiting approval in Ohio, a board-set administrative rate in the Tennessee Valley, a contested federal docket in Wisconsin, and a statewide audit in Texas — and they work through different mechanisms: transmission cost allocation, wholesale rate classification, interconnection-queue validation. What unifies them is one question: who bears the infrastructure and nonperformance risk when a wire's cost is certain, and the load behind it is not.
Ohio is closest to resolution, though the proposed settlement leaves transmission cost-allocation details to a follow-on proceeding. In PUCO Case 25-0958-EL-AIR, AES Ohio's three-year rate plan, the state's manufacturers spent the spring fighting a staff-proposed data-center tariff modeled on AEP Ohio's template — "ratemaking by speculation," the Ohio Manufacturers' Association Energy Group called it. Then the parties negotiated. The stipulation filed July 21 is unanimous: every party, including the manufacturers, the consumer advocate, and commission staff, either signed or agreed not to oppose it.
The stipulated Data Center Tariff is stronger than the template the industrials rejected. It establishes data centers as a separate customer class, and requires them to pay "the actual cost of transmission upgrades necessary to serve the Data Center plus the Data Center's reasonable share of existing transmission costs," including supplemental projects that "would not have been planned by the Company but for the service request." Exit fees and forfeited collateral get credited back to retail ratepayers, and speculative data-center load is barred from long-term forecasts.
Nothing is in effect, and the hardest question is deferred: the stipulation requires AES Ohio to open a follow-on proceeding by November 1 to set cost allocation and rate design for its transmission cost recovery rider, and the parties reserve their positions there. First-year rates take effect January 1, 2027 only if PUCO approves the package, which puts the commission's decision on a year-end clock.
The Tennessee Valley Authority reached a data-center rate by a different route: its board set one administratively. On August 20, the board approved changes to TVA's wholesale rate structure that its release says will "protect residential and manufacturing customers from subsidizing the expenses associated with the significant growth of data center load in the Valley."
The release did not include mechanism details. Per CFO remarks reported by the Associated Press, the rate amounts to roughly a 10% average increase phased over three years, effective October 1. Axios reports it applies to new and existing data-center customers across four demand-based rate classes, and that demand above 5 MW triggers an additional capacity-cost provision for incremental costs the standard rate does not cover. TVA is a federal corporation whose board sets rates without a state commission, so there was no contested case and no docket.
After the rate vote, the board approved a new directly served TVA customer for the Colossus 2 campus: MZX Tech LLC, a wholly owned subsidiary of SpaceXAI (formerly xAI) that holds the Tulane Road and Southaven assets. Under the reported terms, the company funds all dedicated transmission and interconnection facilities itself. TVA's release lists the action only as "Approved Load Greater than 100 megawatts," without naming anyone. That closes a loop SCN examined in May, when xAI bet its Memphis buildout could outrun regulators: that company now buys power from the federal utility directly.
Wisconsin is where the measuring-stick argument is sharpest. ATC calls its structure a customer-protection mechanism following "cost causer, cost payer," and nobody in the docket disputes the principle. The protests are about the instrument.
The Minimum Transmission Charge bills We Energies, and through it Microsoft, at ATC's overall network rate for any shortfall between requested transmission capacity and service billed, capped at ATC's estimate of the facilities' lifetime incremental revenue requirement. CUB contends that rate is the wrong one: incremental upgrades built at today's costs will run higher per megawatt than the embedded historical average the network rate reflects, so the network-rate charge "would be too low." Microsoft argues the misalignment cuts both ways: because the charge is pegged to projected in-service dates and capacity rather than the incremental facilities' actual costs, with no true-up, it "will correspondingly lead to overcharging or undercharging, with no opportunity for correction." The state commission's August 14 comments, cited in Microsoft's protest, modeled one large customer overpaying by roughly $451 million while another underpays by $441 million.
Microsoft's protest goes further down the contract, arguing an early-termination fee is "an unjustified windfall payment to ATC." The company also cites the White House's Ratepayer Protection Pledge as something the settlement process should memorialize in tariff records. A pledge designed to reassure households about hyperscaler cost-shifting is now being invoked by the hyperscaler against a utility's pricing plan. FERC has not yet ruled on the protests.
Texas is not arguing about rate design. It is testing whether the loads are real. On August 3, Governor Greg Abbott directed the state's utility commission and ERCOT to run a "comprehensive verification and audit of all data centers advancing through ERCOT's interconnection process" before approving any new ones. The queue holds roughly 474 GW of requests, about 90% of them data centers, more than five times ERCOT's record peak demand. Projects must disclose public financial assistance, power sourcing, water use, and ownership; noncompliant ones "must be denied connection to the Texas grid."
The audit is in execution. ERCOT has targeted December 10 for its report, with roughly 300 facilities of 75 MW or more to verify. As of late August, it was still developing the request-for-information template for its Batch Zero large-load queue, and had separately sent utilities RFIs covering medium-sized loads. Deadlines have already slipped: ERCOT's regulatory chief says the pending Batch Zero study will not be done by its April 2027 deadline.
None of this started in August. AEP Ohio's data-center tariff, the template the Ohio industrials pushed past, was approved in summer 2025 and had 5,642 MW under contract by February. The DELTa database of large-load tariff filings counts 53 proposed or approved for 2025 alone. The disputes now run in every direction. In Illinois, developer PowerHouse Hillwood is fighting Commonwealth Edison at FERC over a canceled transmission security agreement for a planned 1.8 GW campus in Joliet, warning that "utilities like ComEd will use their monopoly power as a bludgeon." Industrials versus utility, customer versus utility, developer versus utility, state versus its own queue. The hyperscaler posture shifted early this year: Microsoft president Brad Smith wrote in January that "it's both unfair and politically unrealistic for our industry to ask the public to shoulder added electricity costs for AI."
What August added is a test of instruments. The stated policy objective across the proceedings is the same: prevent data-center costs from shifting to other customers. Ohio's answer is actual-cost recovery with but-for supplemental projects included and revenues credited back. ATC's answer is a minimum charge at the network rate under a liability cap; Microsoft and CUB want incremental-cost pricing with true-ups. TVA's answer, as reported, is demand-based rate classes with a capacity provision above 5 MW. Texas operates a layer earlier, testing whether the projected loads and their impacts are credible. Each confronts the tension FERC flagged for the whole MISO footprint in its June show-cause order on large loads: how to price infrastructure whose cost is certain against a load that may not fully materialize, without locking the error in for decades either way.
The milestones are identifiable, if uneven. ERCOT's report is due December 10, and Ohio's first-year rates would take effect January 1 if PUCO approves the stipulation; FERC has set no date for acting on the Wisconsin protests. The generation bottleneck told the industry when the power could arrive. These proceedings show the policy debate moving from whether power and infrastructure can be built to how utilities allocate the cost and nonperformance risk of assets expected to remain in service for decades.
The SCN Staff is a small AI editorial squad working under human direction. Each agent owns one job.
Scout does the research. It runs down primary sources and checks what's already been published, on SCN and everywhere else, before a story gets written. If a claim can't be traced back to a real document, Scout flags it.
Forge writes. It takes what Scout found and turns it into a draft, argument and sentences and all. Every SCN piece starts here, then gets sharpened.
Cipher handles search: the titles, descriptions, and keyphrase work that decides whether a good article ever gets found. Least glamorous job on the squad. Also one that matters more than it looks.
Pixel makes the visuals. Images, charts, the occasional diagram, all built to SCN's brand instead of pulled from a stock library. When something's easier to see than to read, it goes to Pixel.
Editorial judgment and the final call stay with the humans. So does the fact-checking.
