Digital Realty to buy Blackstone’s stake in three Virginia data centres

Northern Virginia is where the internet physically lives, and a large piece of it just changed hands.

Digital Realty has agreed to buy Blackstone’s majority stake in three fully leased data centres in the region, in a transaction that values the assets at $7.8 billion.

It is the kind of number that has become routine in a sector where AI demand has turned warehouses full of servers into some of the most contested real estate in the country.

The mechanics are specific. Digital Realty is paying $3.5 billion for Blackstone-managed funds’ blended 64% equity interest in the portfolio, split between $1.2 billion of cash and $2.3 billion in Digital Realty shares.

Paying for two-thirds of the equity in stock rather than cash tells you something about both the price of capital and Blackstone’s willingness to stay invested in the upside, this time through the buyer rather than the buildings.

The portfolio itself is a tidy thing to own. It comprises two data centres in Manassas and one on the Digital Dulles campus in Sterling, each with 96 megawatts of IT capacity, all of it fully leased to three separate investment-grade hyperscale customers.

In a market where the scarce input is not floor space but power, 288 megawatts of contracted capacity with creditworthy tenants is close to the platonic ideal of a data-centre asset.

The deal carries an expected initial stabilised capitalisation rate of over 6.5%, the yield figure that tells institutional buyers what they are getting for their money.

The purchase was expected to complete on 30 June, subject to customary closing conditions, which makes this less a proposed transaction than a near-finished one.

For Digital Realty, a listed real estate investment trust and one of the largest data-centre operators in the world, the logic is straightforward: acquire stabilised, leased capacity in the single most important data-centre market in the United States, and do it without the years of permitting, construction, and power-procurement risk that a ground-up build would carry.

Buying finished, occupied megawatts is the fastest route to scale at a moment when power is the bottleneck for everyone.

Northern Virginia’s Loudoun and Prince William counties handle a disproportionate share of global internet traffic, and grid constraints there have already pushed some operators to queue for years before they can energise new capacity, which makes a fully powered, fully leased portfolio rarer and more valuable than the headline price alone suggests.

For Blackstone, the sale is a partial exit from a position it has ridden well. The firm has been one of the most aggressive private-capital players in digital infrastructure, and crystallising a gain on prime Virginia assets while rolling part of the proceeds into Digital Realty stock is a characteristic move, banking the win without fully leaving the table.

Retaining a minority interest alongside an experienced operator also lets it stay exposed to the leasing economics it knows well, without carrying the operational burden of running the sites itself.

The transaction sits inside the broader capital surge reshaping AI’s physical layer, where the constraint has shifted decisively from chips to the buildings and electricity that house them.

The investment thesis driving these deals is the same one underpinning the chip side, where firms like SiFive frame data-centre silicon as a $100 billion-plus market created by agentic AI. Concrete, power, and cooling are simply the same bet expressed in real estate.

That bet keeps drawing capital. Goldman Sachs and Nomura have attributed a large share of China’s recent export growth to AI-related goods, the semiconductors and data-centre components feeding infrastructure build-outs worldwide, a demand signal that runs straight through deals like this one.

With the purchase due to close, attention turns to where the next tranche of contracted megawatts comes from, and how much the buyers are willing to pay to lock it down before someone else does.