Meta's fourth Utah build buys what capital can't
The fourth Eagle Mountain expansion pushes Meta's Utah spend past $3 billion, and the missing square footage says as much as the liquid-cooling spec about how hyperscaler capex gets allocated.
Meta's fourth expansion at Eagle Mountain, outside Salt Lake City, is the most useful item in this quarter's data center pipeline for anyone pricing digital infrastructure, and it arrived without a square footage figure or much else. The company said this week that it will add buildings to its Utah campus, taking its total investment in the state past $3 billion, and that the new halls will use closed-loop liquid cooling with dry cooling. The omission fits a site Meta has been expanding since 2018, where no tenant, no lease and no counterparty sets the terms.
The campus reached its current shape through a $750 million first phase, a 500,000 sq ft addition planned in 2019, 900,000 sq ft in 2021, and two buildings totaling 2,000,000 sq ft in 2022 — roughly 4.4 million sq ft, all inside the same $3 billion Meta now cites for the state and all standing before the newly announced round. That history is why the fourth round is the cheap round: a greenfield data center carries every cost not yet paid, from land assembly and an interconnection agreement to water, local consent and queue position, and at Eagle Mountain those costs are sunk.
The new buildings buy steel, cooling loops and servers against a grid connection and a permitting record that a competitor cannot assemble in the same quarter at any price.
What the first three rounds bought
Consent is now a priced good, as this publication has argued. The $18 million in Ohio grants that bought community patience were a rounding error next to grid cost, but they priced the one input a take-or-pay tariff cannot reach. Eagle Mountain shows what a consented site looks like once that price is settled: a jurisdiction that has approved four expansions is not a permitting risk, it is an annuity.
The jobs numbers put a price on that consent: more than 1,200 construction workers at peak and 300 permanent roles is what a county gets for its land, its water and its patience, and Meta has now paid it four times. Set that against the cost of a first approval elsewhere — the Ohio grants, the tariff structures that price community toleration — and a fourth expansion in a jurisdiction that keeps saying yes is the cheapest capacity Meta can add in any given quarter.
The cooling choice is the more interesting disclosure. Closed-loop liquid cooling serves higher rack densities than the air-cooled halls Meta has historically run, and pairing it with dry cooling limits on-site water use — a design that points to water as a constraint, though that reading is inference from the specification rather than a stated reason, since the company did not explain the design. Whether the next buildings are social-feed capacity or model-training capacity is a question the cooling spec answers faster than any press release will.
The $3 billion figure is cumulative, which makes it a stronger jobs headline than an underwriting datum; the incremental cost of this round is not public. With the square footage missing too, Eagle Mountain sits in the same half-disclosed category as most of the sector's announcements, enough to establish commitment and not enough to model returns. The difference here is that Meta is spending its own balance sheet, so the only party that needs the increment is the market.
The watt it already holds
Meta counts more than 30 self-built campuses in operation or development, has said it is targeting tens of gigawatts of compute within the decade, and leases third-party data center and cloud capacity at volume. Build where the site is settled, lease where it is not: that stance says more than the wattage on any single press release.
Self-building at this scale is a statement about the lease market too. Meta rents third-party capacity in volume, but ownership converts recurring rent into depreciating assets and, at a site like this one, into repeat expansions with no landlord standing in the middle. That trade only works where the site is settled; where it is not, the lease is cheaper than the delay, and much of the current buildout sits on the wrong side of that test.
Power is the input this expansion does not advertise. Meta's 125 MW Texas solar agreement did more underwriting work through the offtaker's name than through the megawatts, and Eagle Mountain runs the same logic in reverse: the company is converting an interconnection it already holds into more compute rather than buying power in a market. The utility relationship, like the permit, behaves as a fixed asset by now, which is why the marginal building is priced against Meta's own cost of capital and not against a merchant power curve.
The same hierarchy shows up in miniature: hyperscaler-anchored capacity clears at infrastructure pricing because the anchor's balance sheet absorbs development risk, while unanchored AI capacity has to fund grid services and governance costs that no lease covers. Growing a campus that already holds its permit, its interconnection and its cooling design is a better trade than permitting a new one, and the market keeps pricing that trade backwards. Eight years took Eagle Mountain from a single $750 million phase past $3 billion. The square footage on this round, when it comes, will say whether Meta is still scaling the site or just topping it up.
a jurisdiction that has approved four expansions is not a permitting risk, it is an annuity