Thirteen buildings across northern Indiana and Mississippi changed hands this week with nothing in them but concrete, copper and switchgear.
Blackstone Digital Infrastructure Trust agreed to pay about $5 billion for a 75% interest in two joint ventures with Amazon Web Services, the cloud unit of Amazon.com Inc. The assets are powered shells inside two campuses AWS is developing: built, energized, and not yet occupied. Amazon keeps the other quarter and will lease back all of the space.
The price reflects a 7.0% initial capitalization rate, the ratio of a property’s first-year net operating income to what was paid for it. Against $5 billion, that implies roughly $350 million of rent a year.
Leases start at twelve years and can extend to 32 years through options, the trust said. The purchase is funded with $1.9 billion raised in its initial public offering and about $2.5 billion of non-recourse mortgage debt secured against the properties. Closing is expected in the fourth quarter.
It is the first acquisition since the vehicle listed, which makes it the reference point for everything it buys next.
Sale-leasebacks have become the standard way to pay for AI construction without carrying the entire bill on a cloud operator’s balance sheet. Amazon gets cash back for land, steel and substations it has already funded, keeps operating control of the buildings, and turns a capital expense into a rent line. The trust gets a tenant with an investment-grade rating and a lease long enough to borrow against.
For Amazon the accounting effect is not incidental. Data centers it owns sit in property and capital spending. Ones it leases sit in operating costs, with the obligation spelled out in the footnotes. Both arrive at similar economics over time; the two lines are read differently by investors who value cloud businesses on free cash flow.
The trust’s arithmetic is plainer. It buys a rent stream at a 7% yield, borrows part of the price, and distributes most of what remains. Return comes from the spread between the cost of that debt and the yield on the lease, plus whatever the buildings are worth when the lease expires.
That last variable is the one nobody can price today. A powered shell designed around one generation of accelerators may be obsolete in a decade. It may also be exactly what a replacement tenant wants, because the hard parts to obtain were never the walls. They were the substation, the water rights and the permit.
The timing is awkward. On the same day, another Blackstone portfolio company, QTS, completed a private placement well below what it had sought: about $300 million against a target of $800 million, on seven-year notes priced at 245 basis points over Treasuries, for an all-in cost above 7%. QTS has issued roughly $11 billion of debt in two years and is building against a pipeline measured in gigawatts.
Read side by side, the two transactions describe a market that has not shut but has repriced. Debt for data centers grew scarce enough this autumn that a serial issuer accepted 60% less than it asked for and paid more for the privilege. Equity for the same asset class, packaged as a listed trust with a named tenant, cleared at a 7% cap rate.
The difference is the tenant. QTS builds capacity for customers it has not signed yet, and that capacity has to be filled after the concrete is poured. The trust is buying buildings Amazon has committed to occupy for at least twelve years. In a market that has lost patience for a story, a lease with a hyperscaler’s name on it trades closer to a bond than to real estate.
Concentration is the obvious cost. One tenant, two campuses, thirteen buildings. If Amazon presses for better terms at renewal, or consolidates workloads elsewhere, the trust owns an empty shell in Indiana with no second taker in line. Amazon’s incentive runs the other way for as long as it needs the capacity, which is the same thing as saying the credit depends on demand for cloud computing holding up.
The structure matters more than the sector label. A long lease to a rated tenant is priced as credit: the questions are the tenant’s rating, the lease’s term, and the recoverability of taxes, insurance and maintenance. Real estate underwriting asks a different set of questions, about comparable rents and exit cap rates, and this deal answers those only at the end of three decades.
Power is the constraint that decides whether the shell was a good purchase. Campuses in Indiana and Mississippi were chosen for available generation and land, and the interconnection queue in most American markets now runs years. A building with an executed power contract is worth more than one without, whatever the square footage says.
For Blackstone the transaction is also a statement about the vehicle. The trust raised $1.9 billion in the offering, which by itself would not cover a $5 billion purchase; the debt and the joint-venture structure do the rest. Its ability to repeat that on the next campus is what turns one deal into a platform.
Watch the next cap rate. If the trust buys again near 7% with a comparable lease, the market is telling you that AI infrastructure has found a clearing price and a buyer base that does not need to believe the demand forecasts. If the next purchase needs a higher yield to clear, the first one was priced on novelty.
The buildings will be full within a couple of years, humming and hot. What they are worth in 2058, when the options run out, is the part of the trade no lease discloses and no filing can settle.


