Investors are beginning to confront potential risks linked to the unprecedented surge in borrowing driving the expansion of artificial intelligence infrastructure in the United States. At the center of this scrutiny is Meta Platforms’ massive AI data center project named Hyperion, currently under construction in Richland Parish, Louisiana. Hyperion represents Meta’s largest data center endeavor, featuring 5 gigawatts of compute capacity and involving more than US$50 billion in investment.
To finance the project, bonds totaling US$27 billion were issued last year through a joint venture vehicle known as “Beignet Investor,” a nod to the iconic New Orleans pastry. Meta holds a 20 percent minority stake in the venture, meaning the project’s liabilities are off Meta’s balance sheet. This approach mirrors financing structures employed by other major technology companies including Nvidia, Broadcom, Amazon, and Alphabet, which collectively utilize off-balance-sheet financing estimated to be worth approximately US$3 trillion, according to Morgan Stanley.
The Beignet bonds, which carry a 6.581 percent interest rate and mature in 2049, were initially rated A+ by S&P Global. However, S&P warned of “substantial credit risk” during both the construction and operational periods of the facility. Recent market movements suggest these risks may be materializing, as the bond prices have declined, trading at roughly 91 US cents on the dollar, pushing yields to 7.55 percent on Monday. This yield premium is about 230 basis points over U.S. Treasuries, wider than the 185 basis points at issuance but narrower than the peak spread of 255 basis points seen in July. While not triggering immediate alarm, these shifts signal emerging stress in certain segments of AI-related debt.
Other indicators point to growing investor concern as well. Credit default swap (CDS) spreads for companies in the AI ecosystem, including Meta, have hit new highs, with Meta’s CDS nearing 100 basis points, exceeding previous records from July. Additionally, Oracle recently declared a “force majeure” for potential delays in expanding an AI data center in New Mexico, citing challenges in securing power, potentially postponing the project by up to a year. Notably, Blue Owl, a co-investor in Oracle’s Project Jupiter, also features in Meta’s financing structures.
Despite these warning signs, major tech companies like Meta remain unlikely to default given their fundamental business strength. The current selloff in their debt raises questions about whether market fears are justified or if some of these securities represent undervalued opportunities. Much depends on the optimistic but uncertain assumptions that AI-related revenues will ultimately outweigh the enormous capital outlays.
This rapid increase in AI investment is striking. Industry estimates suggest AI capital expenditure will reach about US$1 trillion in 2026 and expand to approximately US$1.2 trillion in 2027. Oxford Economics projects cumulative AI investment totaling around US$3.8 trillion from 2024 through 2028, much of which is now debt-financed after cash reserves have been depleted.
Goldman Sachs analysts have calculated that hyperscale cloud providers will need to generate approximately US$300 billion annually in AI revenues over the next few years just to reach breakeven. To achieve desired returns of 15 to 20 percent, this translates to an additional profit between US$570 billion and US$800 billion. However, Oxford Economics highlights a potential disconnect, estimating that AI will increase U.S. GDP by roughly US$850 billion by 2032. Capturing two-thirds or more of this projected GDP growth by a small group of hyperscalers to meet profit targets appears unlikely.
Despite these concerns, borrowing continues to accelerate. Hyperscalers have issued roughly US$250 billion in debt so far this year, with higher interest rates adding to the cost burden. The Federal Reserve’s ongoing rate hikes and rising Treasury yields—the highest in over a decade—are intensifying the financial pressures on these off-balance-sheet and traditional borrowings.
The fate of Meta’s Hyperion financing and the broader “deep-fried” debt market may provide early indications of how much strain the AI investment boom can tolerate as the industry seeks to balance lofty growth expectations with mounting credit risks.