The economics of the structure
Open the model. A BlackRock-linked vehicle holds 80% of the roughly 1-gigawatt El Paso project and Meta 20%; the 2048 bonds are secured by 20 years of Meta rent beginning in 2028. The debt does not sit on Meta's balance sheet; the rent obligation is a fixed claim on future cash. The capacity is rented.[1]
The pricing matters: early yields are above 7%, about 0.4 percentage points higher than on Meta's Hyperion project, funded by a bond sale of a record 27 billion dollars in October. On a deal of 12 billion dollars, even a tenth of a point means millions in annual interest. The market is repricing AI infrastructure risk; the cost of the spend is rising even as the structure hides the leverage.[1]
Capital allocation is character
The sensitivity concentrates in a single assumption: if the rented capacity converts into durable incremental cash at Meta, the deal earns its cost of capital; if AI monetization lags, the 20-year rent could be a fixed cost against uncertain returns. The counterargument deserves weight too: real option value and the off-balance-sheet structure can keep Meta's reported leverage low. The story is cheap; the cost of capital is real.[1]
The signal to watch is the coupon the El Paso bonds ultimately clear at, and whether subsequent AI financings widen further against Hyperion. This is an observable test of financing cost; if the spread widens, the bill for the spend gets heavier.[1]