Meta Owns 20% of Its Own Gigawatt Data Center

Meta and BlackRock announced a strategic venture on July 28 to develop and own Meta's data center campus in northeast El Paso, Texas — roughly 1,000 acres, 1 gigawatt of capacity, about $14 billion in total development cost.

Funds managed by BlackRock will own 80%. Meta will own 20% of the campus it has been building for more than six months, will continue to manage construction and operations, and will be the site's initial sole tenant.

Meta is not selling a data center it no longer wants. It is selling most of one it fully intends to use.

The mechanics

At closing, Meta contributes the land and construction-in-progress assets, valued at approximately $2.3 billion. BlackRock contributes roughly $4.9 billion in cash. To square the resulting ownership split, Meta receives about $1 billion in distributions back out of the venture. Reporting on the deal puts the remainder of BlackRock's side funded through roughly $12.5 billion in debt financing.

Net it out from Meta's seat: it hands over an asset carried at $2.3 billion, takes $1 billion back in cash, retains a fifth of the equity, keeps operational control, and gets a gigawatt of compute delivered — with first capacity targeted for 2028 — without the full $14 billion crossing its own capital expenditure line.

At peak construction the project is expected to support more than 4,000 construction jobs, with roughly 300 permanent operational roles once the campus is running. Those are the numbers El Paso will be quoted. They are not the numbers that make this deal interesting.

This is the second time

In October 2025, Meta did structurally the same thing with Blue Owl Capital for Hyperion, its Louisiana campus. Blue Owl took 80% through a special-purpose vehicle, Meta kept 20%, and the project was financed with about $27 billion in project bonds. Meta leases the finished campus back under a long-term operating lease — converting what would have been capex into operating expense, with a multi-year exit ramp built into the lease terms. Hyperion is targeted at roughly 5 GW.

Two flagship sites, nine months apart, same 80/20 split, same private-capital counterparty class, same leaseback shape. The first one could be read as opportunistic financing. The second one is a template.

And the counterparties are not incidental. Blue Owl is one of the largest private-credit managers in the world. BlackRock, through its acquisitions of Global Infrastructure Partners and HPS, has assembled exactly the infrastructure-equity-plus-private-credit combination this structure requires. The buyers of AI compute infrastructure are increasingly the same institutions that buy toll roads and pipelines, and they are underwriting it the same way: long-lived physical asset, single investment-grade tenant, contracted revenue, twenty-year horizon.

That is a coherent trade for BlackRock. Meta is close to the ideal tenant — enormous balance sheet, structural need for the asset, and operational control of the thing it is leasing, which removes most of the execution risk a normal landlord would carry.

What it does to the capex number

Meta raised 2026 capital expenditure guidance on April 29 to $125–145 billion, up from $115–135 billion, citing higher memory-chip pricing and additional data center costs.

Hold that against what just happened. A $14 billion campus and a Hyperion project financed with $27 billion in bonds are both being built for Meta, operated by Meta, and occupied by Meta — while sitting substantially outside the capex figure investors track. The reported number is real. It is also, increasingly, an incomplete measure of how much infrastructure a hyperscaler is actually commissioning.

Compare the alternative approach. Alphabet guided to $195–205 billion of 2026 capex last week, on balance sheet, and the stock sold off on the spending line despite a quarter that beat on every operating metric. The market is currently punishing visible capital intensity. Meta has found a way to acquire the same physical capacity while showing investors a smaller number and a lease obligation instead.

Whether that is financial engineering or prudent capital allocation depends entirely on a question nobody can answer yet: does a 2028-vintage gigawatt campus still earn its rent in 2035?

If it does, Meta will have secured a decade of compute while conserving balance-sheet capacity, and the 20% equity stub will look like a cheap option on the upside. If AI demand disappoints or the hardware inside these buildings depreciates faster than the leases amortize, Meta is contractually committed to paying rent on a gigawatt regardless — and the obligation does not evaporate just because it was never labeled capex.

The broader signal

The most durable read here is not about Meta. It is about who is financing the buildout.

Hyperscaler cash flows, enormous as they are, are no longer sufficient to fund AI infrastructure at the pace the demand curve is asking for. The marginal dollar is now coming from private credit, infrastructure funds, and project bonds — capital that requires contracted, long-dated, asset-backed structures to move at all.

That has two consequences. It makes the buildout more resilient in the short run, because it taps a far deeper pool than corporate cash flow. And it makes it more rigid, because leases and project bonds do not flex when demand does. Capex can be cut in a quarter. A twenty-year lease cannot.

Meta has now run this play twice, at scale, with two of the largest alternative-asset managers alive. Expect the rest of the field to run it too — and expect the headline capex numbers to keep getting less informative as they do.