On October 30, 2026, Michigan Attorney General Dana Nessel filed a petition for rehearing before the Michigan Public Service Commission, asking it to revisit its conditional October 1 approval of DTE Electric’s special 20-year contracts to serve a 1-gigawatt Google data center in Van Buren Township. The case, U-22058, is Michigan’s first hyperscale data-center contract decided through a contested case rather than an expedited, uncontested process — and the rehearing petition is now the clearest test of whether the protections the commission approved are strong enough to hold for two decades.
The dispute is not about whether the campus gets built. The Van Buren Township Planning Commission approved the final site plan for the 1-gigawatt facility and its accompanying substation on October 14, on 282 acres north of Interstate 94 between Haggerty Road and Interstate 275. Service is expected to begin in December 2027, with the campus reaching maximum load by December 2028. The dispute is about the bill — specifically, about who pays when the arithmetic of a 20-year deal turns against the utility’s other customers for a stretch of years in the middle of the contract.
The aggregation fight
The petition’s central target is a methodological choice that sounds technical and turns out to be everything. The commission adopted DTE’s contract-long “aggregation” approach to weighing projected ratepayer benefits against the actual costs of serving the Google campus. Under that standard, the question is whether other customers come out ahead across the entire 20-year term, not whether they come out ahead in any given year.
Nessel’s argument is that this standard blesses years in which other customers plainly subsidize the data center, so long as the long-run ledger still looks favorable. The commission itself concedes that will happen for at least four years, from 2035 to 2038, according to the Attorney General’s office. DTE’s own modeling, as described in the petition, shows other customers facing net cost increases in that window — a window that opens roughly seven years after the campus reaches full load and closes a decade before the contracts expire.
That is the petition’s most damaging sentence, and it is worth pausing on. The commission did not deny the four-year gap. It accepted it, on the reasoning that the benefits projected over the full term outweigh it. Nessel’s counter is that a benefit projected over twenty years is not a benefit realized in 2036, and that a household paying higher rates in 2036 cannot spend a benefit projected for 2043. Her petition asks the commission to require costs and benefits to be accounted for and reconciled annually, rather than netted across the life of the deal.
The distinction matters beyond Michigan because aggregation is becoming the quiet default in large-load cases. Every utility witness in these fights argues that a big new customer spreads fixed costs across a larger base, lowering the per-customer cost of the system. That argument is true in the aggregate and can still be false in a given year, when the generation and transmission built for the campus are in rate base but the campus’s payments have not yet caught up, or when the customer’s load profile shifts. An annual reconciliation requirement would force the utility to show its work year by year. A 20-year aggregation standard lets the utility show it once, in advance, on the basis of a model.
What the contracts actually require
To be fair to the October 1 order, it is not an empty document. The commission voted unanimously to approve the contracts conditionally, and the conditions are the strongest Michigan has yet attached to a data-center deal. The primary supply agreement runs 20 years, sets a minimum monthly billing demand of 80 percent of contract capacity — the take-or-pay floor — and includes a termination payment designed to ensure Google pays for at least 15 years of minimum monthly charges if it walks away early. The contracts are structured to cover generation, transmission, distribution, and other costs of serving the campus, and the commission required DTE to track project-related costs.
Google also funds, through a separate agreement, DTE’s development of up to 1,600 megawatts of renewable generation and 480 megawatts of battery energy storage, which DTE will own and operate. DTE has said Google will cover the full cost of powering its data center, and the utility’s 20-year model projects roughly $1.7 billion in system-wide benefits to other customers over the term — the figure that underwrites the aggregation argument.
Nessel’s petition does not dispute that these terms exist. It disputes that they are enough. In an August 3 brief in the case, her office pressed for a 90 percent minimum billing demand rather than 80 percent, and for an early-termination payment calculated at 100 percent of minimum billing demand multiplied by the remaining months of the 20-year term, rather than the 15-year floor the commission approved. The difference between 80 and 90 percent is the difference between a campus that can idle a fifth of its contracted capacity at no penalty and one that can idle a tenth. Over a 1-gigawatt contract, that ten-point gap is 100 megawatts of capacity whose costs somebody else carries if Google scales back.
The petition also attacks a collateral cap the commission adopted, which limits the liability of Google and its parent company, Alphabet, for costs of service. Nessel urges the commission to reject the cap outright on rehearing. The logic is straightforward: a take-or-pay promise is only as strong as the balance sheet behind it, and a cap on the parent’s liability weakens the promise precisely in the scenario — a scaled-back or abandoned campus — where other customers would need it most.
The transmission costs nobody accounted for
A second thread of the petition will sound familiar to anyone who followed Michigan’s other data-center fight. Nessel argues that the commission, as it did when it approved DTE’s contracts for the Oracle and OpenAI campus in Saline Township in December 2025, ignored significant transmission costs her office identified that DTE has refused to account for in its contracts and assessments.
This is the same fault line this desk has documented in Indiana, where the contested remainder in every large-load case turned out to be network upgrades: the headline generation deal protected other customers, and the transmission upgrades became the argument. The reason is structural. A gigawatt of new load changes flows across the shared network in ways that cannot be neatly fenced off to one customer, and every utility witness argues that shared network investment benefits shared network users. Consumer advocates counter that upgrades that would not exist without the data center should be paid for by the data center, whatever their incidental benefits.
In Michigan, the transmission question is entangled with DTE’s broader generation plan. The contracts DTE filed in March modeled a proxy 727-megawatt gas plant for 2033, similar in size to a roughly 700-megawatt plant the utility is now proposing in Milan Township, alongside a 1,400-megawatt plant proposed in Handy Township. The two plants together carry an estimated capital cost of about $6.4 billion, and both are described as carbon-capture-ready — designed so capture equipment can be added later, not equipped with it now. Michigan’s 2023 clean energy law classifies gas with carbon capture trapping at least 90 percent of emissions as clean energy, which is part of why the “ready” designation matters to DTE’s compliance story. Environmental advocates note, correctly, that carbon-capture-ready plants still emit carbon and other pollutants until capture is actually installed.
DTE’s position is that the gas plants are system resources, studied against the needs of its entire customer base, and that no single project determines the need for any particular generating resource. That is a defensible planning argument, and it is also exactly the argument that makes cost allocation hard: if the plants serve the whole system, the whole system pays for them, and the data center’s contribution becomes a modeling question rather than a billing fact. Nessel’s petition is, at bottom, an insistence that modeling questions get answered with annual accounting rather than 20-year projections.
Why the burden of proof matters
Beyond the contract terms, the petition makes a procedural argument that could matter more than any single number. Nessel asserts that DTE repeatedly failed to meet its burden of proof under Michigan law, that the commission applied an incorrect burden of proof and an incorrect standard of review, and that it failed to meaningfully engage with — or outright ignored — important arguments from her briefing, defaulting instead to acceptance of DTE’s proposed terms despite DTE’s failure to support their reasonableness and prudence.
That is strong language for a rehearing petition, and it is deliberate. Rehearing is not an appeal to a higher court; it is a request that the same commission reconsider its own order, and commissions grant rehearing far less often than intervenors hope. A petition that merely reargues the merits usually loses. A petition that identifies legal error — wrong standard, ignored evidence, unsupported findings — gives the commission a reason to act that does not require it to admit it got the economics wrong. Nessel’s office has framed the case that way on purpose, and the petition asks the commission either to adopt her recommendations on rehearing or to reject the contracts outright.
The petition also lands amid DTE’s pending rate case — Nessel earlier called the utility’s offer to pause future rate-hike requests if its data-center contracts were approved a “ransom note” — and while DTE’s broader integrated resource plan, including the two gas plants, is still under review.
What to watch
The first thing to watch is procedural and soon: whether the commission grants rehearing at all, and if it does, whether it reopens the record or decides on the existing one. A rehearing grant with a reopened record would let Nessel test DTE’s benefit projections under cross-examination in a way the original contested case, by her account, did not. A denial would leave the October 1 order intact and push the fight to the courts, where the standard of review is deferential and the four-year gap becomes a footnote rather than a holding.
The second is the annual-reconciliation question, which is the petition’s most portable idea. If the commission adopts yearly cost-benefit accounting for the Google contracts, that standard will be cited in every large-load case that follows in Michigan — including whatever DTE files next from a pipeline it says includes 2 gigawatts in advanced discussions, with a goal of an additional agreement by the end of 2026. If it does not, the aggregation standard becomes the precedent, and the expectation Nessel warns about takes hold: that data-center costs may be passed to other ratepayers in any number of years, under the presumption of a longer-term benefit that has not been adequately supported.
The third is the collateral cap and the 80 percent floor, which are the contract’s real backstops. A rehearing order that raised the minimum billing demand, removed the cap, or extended the termination payment would change the economics of the deal in a way that matters to the next developer’s filing.
None of this should be read as a prediction. The commission approved these contracts unanimously, and unanimous commissions rarely reverse themselves wholesale. But unanimous commissions do sometimes narrow their own orders on rehearing, and the petition gives them specific, severable things to narrow: the aggregation standard, the collateral cap, the transmission accounting. The most likely outcome, if the commission moves at all, is not rejection but revision — a tighter contract, with annual accounting, that still lets the campus proceed.
That would be a meaningful result, and it would also be an incomplete one. Annual reconciliation tells you when other customers are subsidizing a data center; it does not by itself stop the subsidy. The stop comes from the backstops — the minimum bill, the termination payment, the collateral — and from a commission willing to enforce them when a campus underperforms. The rehearing petition asks for both the accounting and the backstops. Whether Michigan gets them will be decided in case U-22058, and the answer will travel: every state weighing a gigawatt-scale contract is watching how the first contested hyperscale deal in the Midwest holds up when the attorney general asks, a month later, who actually pays.
The pattern worth naming
Step back and the Michigan fight fits the pattern this desk has been tracking all year. In Indiana, the first filing under a new large-load law produced a regulated build with a rate-based share of network upgrades and a confidential contract. In Oregon, the binding answer was a tariff with a 90 percent minimum demand charge and 30-year terms for the largest loads. In Michigan, the answer so far is a 20-year contract with an 80 percent floor, a 15-year payment guarantee, and a 20-year aggregation standard that concedes a four-year window in which other customers pay more.
Each jurisdiction is converging on the same toolkit — minimum bills, long terms, exit payments, customer-funded supply — and each is calibrating it differently. The calibration is the whole ballgame. A 90 percent floor is not an 80 percent floor. A 15-year guarantee is not a 20-year guarantee. Annual reconciliation is not 20-year aggregation. And a projection of $1.7 billion in benefits is not $1.7 billion in benefits, until the years arrive and the accounting shows it.
Nessel’s petition is best read as an argument about that last sentence. The commission’s order treats the projection as the protection. The petition treats the projection as the thing that needs protecting against. Between those two readings sits every household and business customer on DTE’s system, waiting to find out whether the four years from 2035 to 2038 are a modeling artifact or a preview. The rehearing will not answer that question definitively — only 2035 will. But it will decide what rules are in place when the answer arrives.