Earlier this month, Meta announced that it would finance the construction of a new data center through debt issued via an SPV, while Oracle is setting up two debt facilities for two data centers. Investors do not see this move toward more “exotic financing” as a sign of weakness. “In a sense, we’re now seeing big tech applying the same financial discipline that has traditionally been associated with infrastructure and utility companies.”
According to a Goldman Sachs report, the five largest US cloud companies invested 342 billion dollar in AI infrastructure this year—a figure the bank expects to rise to 394 billion dollar by 2026. Rising costs are forcing technology companies to rethink their financing strategies. While expansions were long funded directly from cash flow, they are now increasingly being financed off balance sheet.
Earlier this month, Meta announced a joint venture with private credit giant Blue Owl Capital. Together, they will finance Meta’s Hyperion project, a 30 billion dollar data center in Louisiana. Most of the loan—27 billion dollar—will be financed through debt issuance via a vehicle called Beignet Investor LLC.
Meanwhile, according to Bloomberg, Oracle is arranging two separate debt facilities totaling 38 billion dollar for the construction of two data centers in the United States. The financing is led by JPMorgan Chase and Mitsubishi UFJ Financial Group, Bloomberg reported last week. The loan consists of two secured tranches: one worth just over 23 billion dollar for a Texas data center, and another of nearly 15 billion dollar for a site in Wisconsin. Bloomberg noted that the loans resemble project or real estate financing: during construction, only interest is paid, and repayments begin once the data centers are operational.
A turning point
Lisa Shalett, CIO at Morgan Stanley, described the Meta deal as a turning point in an interview with Fortune. The deal, structured as a Rule 144A offering, means the loans are not publicly traded but can be sold among large institutional investors. The bonds are backed by Meta’s future lease income and are included in iShares ETFs managed by Blackrock.
The bonds carry an A+ rating and a 6.6 percent coupon. According to bond specialists, this may seem high but is justified by the long maturity and the specific structure of the issuance. “Meta’s project is financed through a special purpose vehicle (SPV) that holds only the data centers,” said Richard Abma, CIO at OHV Vermogensbeheer. “The bonds are asset-backed and legally ring-fenced from the rest of Meta, which limits risk to the project itself. Using an SPV in this way is quite common. Incidentally, the bonds have already risen by nearly 10 percent since issuance.”
Abma points to the long maturity (until 2049), for which investors demand a premium, as well as a structural premium for the novel, complex setup and the higher risk-free rate on long-term US Treasuries, which now exceed 4.5 percent.
Erik Schmahl, bond strategist at Rabobank, says he is not familiar with the exact transaction but calls it a “special bond”. “The cash flows from the transaction come in as repayments starting in 2029. This type of bond is called a sinker. We rarely see them, though they appear in, for example, mortgage-backed securities. When assessing the price, you must account for the fact that there are no repayments in the early years, followed by accelerated ones later. That can actually make it more attractive than a typical bullet bond, though it’s not suitable for everyone because you have to value all those cash flows.”
The fact that the coupon is about one percentage point higher than on a regular Meta bond with similar maturity reflects, according to Schmahl, the lower rating, different cash flows, and uncertainty around the risk profile. “What happens if, in twenty years, there’s no longer demand for such a data center? Still, the bond clearly offers sufficient value for those who understand its structure.”
“An interesting case”
Schmahl calls it “a very interesting case” and the next step in off-balance-sheet financing. “What we’re seeing is Meta’s willingness to pay extra interest to truly distance the financing from its core operations. The bond is backed by the data center’s cash flows, not by a Meta guarantee. This gives the company the option to part ways with the asset in the future and keeps the large investments off Meta’s balance sheet—allowing it to maintain a comfortable AA rating.”
Asked whether there’s a risk in this type of financing, Abma says he doesn’t view Meta’s or Oracle’s approach as a sign of weakness. “Quite the opposite. It shows that they’re using capital structure management strategically: long-term funding, strong credit quality, and preservation of financial flexibility. In a sense, big tech is now applying the same discipline that has traditionally characterized infrastructure and utility companies. This financing model could easily be adopted by other big tech firms.”
A logical consequence
Garreth Melson, portfolio strategist at Natixis Investment Managers Solutions, sees the shift toward debt financing as a logical outcome of the scale of the AI boom. “New data centers reach full capacity almost immediately, and demand for additional computing power remains high,” he said. “The move toward debt capital is only just beginning, but for now there’s little sign of overcapacity or weakening demand.”
Shalett (Morgan Stanley) told Fortune that the transition from self-funded growth to financing via the shadow banking system and private credit players makes the “simple growth story” of big tech companies “a lot more complex”. According to her, it will become harder to track—and the pressure to actually deliver returns will increase.
High expectations
Robert Almeida, investment strategist at MFS, views the trend with mixed feelings. He notes that this new form of financing leaves “little margin for error,” while much about the ultimate impact of AI technology remains uncertain. “The market is pricing AI as if it will be as transformative as electricity or the internet. If that proves wrong, the risk of write-downs increases.”
Wim Zwanenburg, investment strategist at Stroeve Lemberger, also raises questions about AI companies’ business models. “The question is whether consumers are willing to pay extra for AI functionality in their subscriptions,” he said. “If not, revenues could come under pressure over time. In addition, data centers are heavily dependent on the power grid—and if that lags behind, it could slow growth.”
When asked whether financing structures like Meta’s are possible for smaller players, Zwanenburg says the setup underscores the massive cash flows and credit strength of big tech. “Companies like Meta can afford such long-term, complex financings without straining their balance sheets,” he said. “For smaller players, that would be far riskier.”