A $12 billion raise, and a rising price for it
Meta has raised roughly $12 billion in debt, led by BlackRock, to fund an estimated 1GW AI data center campus in El Paso, Texas, structured around a 20-year lease. On its own, another multibillion-dollar financing for another gigawatt campus barely registers in a year full of them. The detail that earns attention is the price. The bonds priced at yields above 7%, and that number is the reason this deal is worth a technology leader's time rather than just a treasurer's.
We should flag the sourcing plainly. The terms here are attributed to Bloomberg reporting relayed through a Data Center Frontier roundup, so treat the specifics as reported and worth checking against Bloomberg directly. No direct quotes accompanied the coverage. Even with that caveat, the yield figure is concrete enough to reason from, and it points at a shift in how the market is pricing the risk of financing AI at this scale.
Forty basis points that carry a message
The comparison inside the reporting is the useful part. Meta's El Paso bonds priced roughly 40 basis points higher than the debt it placed for its Hyperion campus in Louisiana. Same borrower, same investment-grade name, similar structure, and yet lenders demanded more compensation for the newer deal. When the spread widens on a repeat issuer whose credit has not deteriorated, the market is telling you something about the category rather than the company.
The message, as the coverage frames it, is that investors want more compensation as AI-infrastructure borrowing accelerates. Supply of these deals is climbing fast, and buyers of the debt are responding the way any market does to a surge in issuance: they raise the price. For anyone tracking the cost of the AI buildout, 40 basis points on a $12 billion raise is a small percentage that compounds into real money over a 20-year lease, and it is a leading indicator worth watching quarter over quarter.
Off the balance sheet, into private credit
The El Paso raise is part of a broader wave of off-balance-sheet and private-credit structures that hyperscalers now use to fund gigawatt-scale buildouts without absorbing all of the capital expenditure directly. Rather than parking hundreds of billions of dollars of data center assets on the corporate balance sheet, the largest technology companies are routing the financing through structures that keep much of the debt at arm's length. BlackRock leading this particular deal fits that pattern, with a large asset manager assembling the capital that funds the campus.
This is now the default playbook for the sector, and understanding it changes how you read a hyperscaler's reported capex. The headline capital-expenditure figure a company discloses may understate the true scale of commitment once leased and privately financed capacity is included. For infrastructure leaders, that gap matters because it shapes how much dedicated AI capacity is actually being built, and by whom, beyond what shows up cleanly in public filings.
Why El Paso, and why 1GW
A roughly 1GW campus is a serious footprint, and the choice of El Paso reflects the same constraints steering every buildout of this size. Site selection at the gigawatt scale is driven by access to power, land and a workable path to bring both online within a lease horizon that runs two decades. The 20-year lease structure is the financial expression of that reality: the debt is being matched to an asset expected to run for a very long time.
For the reader, the takeaway is less about Texas specifically and more about the pattern it confirms. Gigawatt campuses financed on 20-year leases are becoming the standard unit of AI infrastructure. That unit is large, long-dated and increasingly funded through private credit, which means the capacity coming online over the next few years is being locked into fixed commitments today. The flexibility a buyer might hope for later is being priced and committed now.
The signal for the wider market
Meta is among the strongest credits that will ever issue this kind of paper. If its spreads are widening as issuance accelerates, it is reasonable to expect weaker borrowers to face steeper premiums, and to expect the neocloud operators and specialist developers who rent GPUs to feel it first. The cost of money for AI infrastructure is rising at the margin, and the margin is where most of the market that is not named Meta actually lives.
We would not overread a single deal, especially one whose terms still merit a check against the primary Bloomberg reporting. But the direction is consistent with everything else visible in AI finance this summer: enormous sums, long leases, and lenders who are starting to ask for more. A widening spread on a blue-chip name is the kind of quiet data point that tends to matter more than the headline dollar figure it sits beside.
What it means for your compute budget
Financing cost is not a back-office abstraction for the people who rent compute. The yield on the debt that builds a campus feeds directly into the price of the GPU capacity that campus sells. When Meta pays above 7% and 40 basis points more than it did on its last deal, that higher cost of capital eventually surfaces in the rates charged to tenants, and in the rates your own providers charge you. Rising infrastructure financing cost is a slow-moving input into tomorrow's compute pricing.
So the practical move is to fold a rising cost of capital into your sourcing assumptions rather than budgeting as though GPU rental prices only fall. Watch spreads on hyperscaler infrastructure debt as a forward indicator, favor sourcing terms that give you room if capacity pricing firms up, and treat any provider promising cheap long-term rates against this backdrop with appropriate skepticism. The El Paso deal is a small, clear reading on where the cost of the AI buildout is heading, and it is heading up.
