Data center capital follows the power calendar, county by county

Ten U.S. counties hold 42 percent of the nation's data centers, and 92 percent of the country's counties hold none of them. That distribution is the fact that has begun to reorganize how private capital underwrites the asset class, because a national demand model no longer tells you where megawatts get built: the county does, and the county's answer arrives on an energization calendar that has displaced acreage and fiber as the variables buyers actually model.

Massachusetts made the shift explicit. Healey's data center order requires developers to clear community benefits agreements before state agencies will act, which turns grid cost and town consent into conditions of entry and adds unquantified duration to every Bay State schedule. Read the order as a subtraction, one jurisdiction removed from the national opportunity set, and you miss the harder point: it reads as a template other states can copy, converting a question that used to be answered by an interconnection queue into one answered by a town meeting.

Concentration at that level is a product of three things that are all local: a site with power capacity to spare, a zoning code that allows a load that size, and a political body willing to say yes. None of the three travels inside a national demand forecast, and the last one is decided by people whose names never appear in a data center pitch deck. That is the gap the Massachusetts order widened, because the consent question now sits on the critical path as a document a developer has to produce before the state will act, and private capital's response, on the evidence of this week, is not to fight the order but to sit on the other side of it.

The two largest data center capital moves of the week point in opposite directions and describe the same insight. I Squared is buying existing megawatts in nine markets that have already decided to allow them, and Blue Owl is buying a public mark on assets the private market has carried at appraisal. Neither is a greenfield gigawatt pledge, and the absence of one from a week this busy is the tell.

I Squared backed Saragon's launch with $1 billion, seeded by a $225 million cash purchase of ten Cogent facilities across nine markets and 53 megawatts, a strategy the firm describes as densifying urban AI inference capacity. Nine markets means nine local approval regimes, which reads as nine separate ways to get stopped until you price the alternative: a single greenfield campus in a single county is one town board away from zero, and the acreage-and-fiber model that served the first data center cycle has nothing to say about which nine markets are the right nine. Ten operating facilities amount to ten already-answered consent questions, and a billion dollars is what that portfolio costs.

Blue Owl's $6.5 billion seed for a data center REIT is the other half of the week's capital, and size is the least interesting thing about it: the vehicle's function is to put a public number on private data center valuations, aimed at the private side of the market as much as the public one, which makes the seed a pricing event before it is a capital raise.

A private data center fund marks its assets through a quarterly appraisal process and a set of comparables most limited partners see in a PDF, whereas a public REIT with $6.5 billion of seed capital would publish an implied valuation every day the market is open. If the public number comes in above the private marks, the private funds have room; if it comes in below, every owner of a private data center position will carry that comparison into their next investment committee. That is why the seed is the week's most consequential data center story even though it involves no construction.

The money is going into assets that already have power and already have permission, and the public market is being invited to price what the private market has been told to believe. Neither move depends on a forecast about national demand; both depend on the local answer to a local question.

The industrial outdoor storage trade belongs to the same story, even if it sounds like a side market. PWD's tracking has institutional capital's share of IOS investment climbing from 30 percent to 45 percent in four years, and construction staging is the demand pulling it in. Staging is a use that ends when the buildout does, which means a buyer paying a permanent cap rate for a tenancy with a maturation date is underwriting a re-tenanting event as much as a rent roll, and two years of IOS bids justified by staging demand alone have been priced off occupancy that will not renew.

The better reading of the same 45 percent is that electrified IOS pairs infill zoning with excess power capacity, which makes the yard next to the substation a different asset from the yard next to the interstate. The first is a data center site in waiting, and the institutions that worked that out are bidding it accordingly. The staging demand ends when the buildout does; the power capacity does not, and that is why the land underneath an electrified IOS parcel outlasts the tenancy currently paying rent on it. The truck yards are the same trade as the data center story, priced earlier.

The 42 percent and the 45 percent describe one rotation: capital is not abandoning real estate for data centers in the aggregate but concentrating in the counties where power, zoning, and a board willing to sign a benefits agreement intersect, and doing so through existing structures because new ones are gated by a consent process Massachusetts just formalized. A manager running a national demand model and a county power map is running two businesses and billing for one.

The I Squared structure — ten operating facilities across nine markets for $1 billion committed — is the shape of the business that gets paid, and the funds still underwriting greenfield campuses against a gigawatt curve are solving a problem the counties have already finished pricing. The order in Massachusetts did not remove megawatts from the country; it removed a jurisdiction from the set a national model would score as available and signaled that the scoring function itself has changed.

The number to watch is how many of the 92 percent of counties without a data center can produce a benefits agreement a developer will sign, because that is the count that decides whether the next generation of megawatts lands in ten counties or a hundred. It will not appear in any model that begins with the country instead of the county.