Who outbid the homebuilders?

In November 2025 the largest land purchase in the history of American data centres was recorded in Prince William County, Virginia: Amazon paid $700 million for 189 acres, roughly $3.7 million an acre, in a county where the National Association of Home Builders puts the median price of land at about $94,000 an acre. The detail that matters is the seller. This was not a speculator cashing out. It was Stanley Martin Homes, a residential builder that had reportedly assembled the site for less than $60 million. Land bought to become houses became, in one transaction, land for machines.

It is not an isolated deal. The NAHB, the homebuilders' own lobby, has assembled the tally: $615 million for 97 acres in neighbouring Loudoun County, over $6 million an acre against a $125,000 median; a Texas corridor where ranch land that traded at $20,000 to $40,000 an acre now asks $350,000; an Illinois subdivision of 55 existing homes bought and razed for a server campus. Behind the bids sits the spending programme funding them: the four technology megacaps were expected to put more than $300 billion into capital expenditure in 2025 alone, most of it AI infrastructure.

The claim on the table

Chapters 2 and 6 of the book make a specific argument: the AI build-out is the largest private capital-formation event in history (Alphabet alone went into 2026 guiding to roughly $185 billion of AI-driven capital spending for the year, the figure the book went to press with; it has since raised that to as much as $205 billion, two-thirds of what all four megacaps together spent in the whole of 2025) and the gains are being banked as assets owned by the frontier, with not one line reserving a claim for anyone else. If that is right, the spending should show up as a bid: frontier balance sheets competing for real inputs such as land, power and water against everyone who needs the same inputs to live. The prediction is not that machines take jobs. It is that machine capital and human need end up on opposite sides of the same auction, and the auction is not close. A data-centre developer bids with retained earnings; a homebuilder bids with project finance priced against what families can afford to pay for the finished houses. The family is in the auction whether it knows it or not, one remove down the chain. Nearly forty times the median is not an anomaly in that contest. It is the clearing price.

The honest complication

The NAHB is a lobby, and its numbers serve an argument; so do mine, which is why both interests belong on the page. America was short roughly 1.5 million homes before the first hyperscale campus broke ground, and Virginia's shortfall of about 188,000 predates this boom too: data centres did not cause the housing crisis. The headline parcels are also idiosyncratic (flat, fibre-adjacent, near substations) and a median-versus-deal comparison flatters the contrast. If the claim were "data centres made housing unaffordable", this evidence would not carry it.

But that is not the claim being scored. The ledger claim says the gains accrue to capital and the costs land wherever the frontier's bid lands, and here the mechanism is visible in a single county's property records. Virginia's own audit commission adds the sharpest fact: the state granted data-centre operators $2.7 billion in tax exemptions over the decade to 2024, a billion of it in the final year. The exemptions attach to the equipment inside the buildings, not to the land bid itself, but the effect runs uphill: every dollar the state forgives on the build-out raises the value of winning the acre it stands on. This is the book's refrain in municipal form: not a law of physics but a design choice, renewed each budget cycle by institutions that have decided which side of the auction to stand on.

Verdict

Confirms. The frontier's capital formation is now repricing the ground itself, and no one outbid for an acre holds any claim on what that acre will earn. The homebuilder, to be fair, did fine: a better-than-tenfold return for selling its land up the stack. The people who would have bought the unbuilt houses got nothing, and will get nothing, because nothing in the structure entitles them to anything. That is the predistribution argument in one transaction. The servers will hum on the old Devlin site for decades, throwing off returns to whoever holds the asset.

Who owns the machine?