Meta Borrowed $27bn Without Borrowing It. This Week, the Bond Hit a Record Low

A campus like this is two assets stacked on top of each other. The top layer is the chips and servers, which are replaced every few years. The bottom layer is land, buildings, power and cooling, which last decades. That bottom layer is what the bond finances ^[2][5].

Why lenders want data centers

Once a data center is built and leased, it behaves less like a technology company than like a toll road: its cash flow is rent, paid by a tenant that is usually one of the largest companies in the world, for as long as that tenant stays ^[2][7]. That is the kind of long, predictable cash flow a lender can size debt against.

And there is a great deal of it to finance. Morgan Stanley puts global data center spending at $2.9tn to 2028, of which $1.5tn cannot be paid from the hyperscalers' own cash flow, and it expects private credit to supply about $800bn of the gap ^[4].

Hyperion is the template for how that money is being lent. Understand it, and you understand where a large part of the next decade's private credit book will come from.

How Meta financed it without borrowing

In October 2025, Meta and funds managed by Blue Owl Capital formed a joint venture to own Hyperion ^[2]. Blue Owl's funds took 80% and put in about $7bn of cash; Meta kept 20% and took a one-time distribution of about $3bn ^[2]. The joint venture then borrowed through a vehicle called Beignet Investor LLC.

On 16 October, Beignet priced $27.3bn of senior secured bonds at 6.581%, maturing in May 2049 and amortising over their life ^[5]. Morgan Stanley ran the books alone ^[5]. PIMCO anchored the order with about $18bn and BlackRock bought more than $3bn ^[1][6]. It was the largest private debt offering ever sold ^[6].

The structure turns Meta from borrower into tenant. Meta leases the campus under operating leases with a 4-year initial term and renewal options, and it guarantees the residual value of the campus for the first 16 years, with capped payments if it walks away ^[2]. The debt sits in a vehicle Meta owns 20% of, not on Meta's balance sheet ^[2][5].

Seen from PIMCO's desk, the trade is the mirror image of Meta's. It is an A+ credit, backed by rent from one of the largest companies in the world, paying about one percentage point more than Meta's own long bonds ^[5][7]. It comes in a size almost no borrower can offer, an $18bn ticket in a single bond ^[1]. And it amortises, so lenders are repaid year by year rather than in one payment in 2049 ^[5].

Why pay more than Meta's own bonds

Two weeks later, Meta borrowed $30bn in its own name. Its bonds due 2045 pay 5.500% and those due 2055 pay 5.625%, rated Aa3 and AA- ^[7]. Hyperion pays 6.581% ^[5].

The difference, close to one percentage point, buys two things: debt that does not appear in Meta's leverage ratios, and a lease Meta can leave after four years. However, the rating agencies did not treat the credit as separate. S&P rated the bonds A+, one notch below Meta, because the contract keeps the risk with Meta ^[5].

Moody's took the argument further in February 2026. It counted $662bn of data center leases signed by five hyperscalers but not yet started, equal to 113% of their adjusted debt ^[8]. Meta itself disclosed residual value guarantees of up to $28bn on leases starting in 2029, none of it recorded as a liability ^[8].

One year later: 94 cents and a second deal

The bond first traded up to 110 cents ^[1]. By late July 2026 it was near 96, and on 23 September it hit 94.4, a paper loss of about $1bn on PIMCO's position ^[1].

The market had also seen a second deal. On 28 July 2026, Meta financed a 1GW campus in El Paso, Texas, through the same 80/20 structure, with BlackRock's infrastructure and private credit arms in Blue Owl's seat ^[10]. The $12.5bn of bonds priced at higher yields than Hyperion ^[11], about 40bp wide of Hyperion ^[12]. Repeating a template, it turns out, costs spread. For a candidate, those 40bp are the most useful number in the story: the market's first price for the structure itself, separate from Meta's name.

Blue Owl itself had a hard winter. In February 2026 it restricted redemptions from a retail private credit fund and sold $1.4bn of loans to pay investors ^[9]. That did not touch Hyperion's bondholders, whose money comes from Meta's rent. It did show who owns the equity in these vehicles.

What could go wrong

The lender's risk sits in the gaps between four clocks. The lease runs four years before Meta must renew ^[2]. The guarantee runs 16 ^[2]. The bonds run to 2049 ^[5]. The servers inside have a useful life that Moody's puts at four to six years ^[8].

In the base case, Meta renews, the campus stays full and the bonds amortise on schedule. In the bear case, a cheaper chip or a smaller model makes part of the campus surplus in the 2030s, Meta pays out under a capped guarantee, and the bondholders own a building whose tenant has left. Critics quoted by Fortune compare that outcome with the dark-fibre overbuild of the 1990s: capacity sitting idle while the debt remains outstanding ^[6].

The price already says lenders want more for the second deal than they accepted for the first ^[11][12]. Our read, which could prove wrong: by the end of 2027, at least one more hyperscaler will finance a campus through the same 80/20 structure, and it will pay a wider spread than Hyperion did.