Grid access becomes a customer class, not a queue

The Midcontinent Independent System Operator has just given data centers their own load class, setting a 25-megawatt IT threshold that lets MISO aim reliability requirements directly at computational loads—and telling the market that grid access for hyperscale and colocation providers will be managed differently from industrial, commercial, or residential load. That makes computational load a separate category of reliability risk, beyond a filing change. The queue is no longer the only gate; the customer class is now an explicit variable.

DayOne made the consequences visible by signing a non-binding memorandum with Tenaga Nasional to add 1.5 gigawatts of on-site generation and battery storage to a planned Selangor campus before the campus has broken ground or closed its land purchase. A data center developer is booking grid-scale power with the Malaysian utility while land rights are still unsettled, treating electricity as the first asset to secure ahead of the real estate.

California's budget made the same point from the demand side, dropping funding for Demand Side Grid Support, its largest virtual power plant program. The program had aggregated distributed batteries and demand response into grid-scale capacity; without new money, that capacity is a political line item rather than a rate-based asset, and a region that spent years cultivating virtual power plants has just shown the largest one can be defunded by a single budget decision. It is policy rationing by customer type as much as MISO's rule, only from the other direction: the grid operator formalizes large loads while the state de-funds small, aggregated ones.

Fervo and Google priced nearly 400 megawatts of enhanced geothermal as infrastructure, with a hyperscaler offtake that gives the subsurface a revenue floor. That is private demand clearing outside the rulemaking—a bilateral contract between a geothermal developer and a data-center buyer that needs no load-class designation because the offtake itself performs that function. A private contract credible enough makes grid-operator classification less binding; it is also the exception, because enhanced geothermal at that scale is scarce and most power-hungry infrastructure cannot rely on a bespoke offtake with a hyperscaler.

Exascale Labs and EnergyBank are testing the opposite end: an 800-kilowatt modular data center on a Norwegian wind turbine, built on the bet that power delivery is the binding constraint in AI infrastructure, ahead of chip supply. At a fraction of the MISO threshold, it still represents the same reordering of priorities—locate the generation, then attach the compute—and makes the grid optional. The project responds directly to grid constraints, though its scale suggests a hedge against being priced out of grid-connected campuses more than a substitute for them.

Storage shows the same asymmetry: the U.S. added 20.2 gigawatt-hours of grid batteries last quarter—a record, by PWD's tracking—but the project names and offtake terms that would make that deployment an infrastructure asset are absent from the public record. Aligned Climate has targeted a $500 million solar-storage fund with no vehicle, anchor investor, or first-close date. The capital is being announced; the contracts that would give it a price are not. A record quarter of batteries without offtake is less a sign of a mature storage sector than proof that deployment can outrun underwriting.

Policy now sits inside the underwriting stack in a more direct way: BloombergNEF expects the Aug. 26 executive order and Treasury's FEOC guidance to delay or cancel storage projects, adding a new gate on battery and inverter supply chains layered on top of the merchant risk that already exists. A developer that has priced a storage project against a PPA now has to reprice it against a policy memorandum. The line between project finance and public policy was never clean; this week it became a hard boundary.

The same reordering runs through the week's other deals. OnZero's agreement with utility Helen prices data center waste heat as a utility product, a step toward making merchant AI capacity bankable, while Green Mountain signed an unnamed neocloud to 14 megawatts in Romford to test whether East London can attract AI demand without a named anchor. Australia's Waroona hybrid is readying for construction with no offtake in sight; Ørsted's 920-megawatt buildout lands with offtake terms still in shadow; and France's second hydrogen auction offers 250 megawatts of price discovery, small capacity but a real benchmark for a sector that has lived on unpriced announcements. None of these are grid-classification stories, but all of them circle the same scarcity: a customer that can actually pay for power, or a policy that can stand in for one.

Grid access is now being formalized, priced, and rationed by customer type and policy decision rather than by interconnection order—MISO's load class and California's budget line are the same instrument applied at opposite ends of the market. For infrastructure capital, the underwriting variable has shifted from how many megawatts can be connected to which class the regulator assigns, and whether a budget line or an offtake survives the next session. The next data center campus will be priced less by its megawatts than by that classification. Watch whether DayOne closes its land purchase on the strength of a non-binding power MOU; if it does, the classification will have become the collateral.