Bond prices tied to data centres that underpin artificial intelligence infrastructure have dropped sharply in recent weeks as investors question whether AI-driven revenues will be sufficient to service large accumulated debts, an analysis in the Financial Times reports.
Which bonds have fallen?
Several debt issuances backing major U.S. data-centre projects have traded below par. Bonds for the Hyperion project, backed by Blue Owl and Meta, fell from above par in July to 92 cents on the dollar. Debt for Project Excalibur—the Blackstone-owned QTS campus leased to Microsoft in Georgia—has dipped below 88 cents. Oracle-backed projects in New Mexico and Michigan have also moved into distress territory.
New cloud providers hit hardest
So-called neoclo ud providers have felt the heaviest pressure. CoreWeave’s bonds slid below 80 cents, a level generally considered distressed, with yields approaching 15 percent on Thursday. CoreWeave’s stock fell 17 percent in a single week, while competitor Nebius lost 18 percent of its value. Nvidia lost roughly $270 billion of market value over the same week, although Nvidia CEO Jensen Huang said he remained optimistic, asserting that CoreWeave’s cost of capital is falling and its position is strengthening.
Analysts warn of consolidation and large write-down risks
Gartner forecasts that 80 percent of neocloud providers could cease to operate as independent entities by 2030. Gartner estimates that data-centre owners would need about $8,500 billion in revenues through the end of 2030 to cover investment and financing costs and to deliver acceptable returns. By contrast, consumer and enterprise AI spending may reach at most $3,600 billion, implying more than $4,000 billion of assets at risk of impairment.
Recent corporate and market developments add pressure
The Financial Times previously reported that OpenAI pushed its stock-market listing into next year, and that OpenAI’s annualized revenue fell about $20 billion short of levels the company had signalled—moves that triggered broad selling in equity markets. Anthropic’s quarterly revenue growth has also slowed. Together, these developments have exacerbated risk sentiment across related equities and debt.
Political and environmental obstacles in the U.S.
Political resistance to new data-centre construction has emerged in several U.S. states over concerns about water use, noise and air quality. States including New York and Texas have imposed tighter permitting restrictions, slowing some projects. For example, permitting for power supply at Oracle’s New Mexico facility has stalled amid local opposition.
Circular financing raises systemic concerns
The Bank for International Settlements (BIS) has warned that financing and commercial links in the AI sector are often circular—clients and suppliers effectively finance one another. That structure increases systemic risk because the difficulties of one participant can quickly propagate through the ecosystem.
Why it matters
Stress on bond markets and on project financing could delay or raise the cost of expanding AI infrastructure, potentially slowing deployment of certain AI applications and reducing the expected economic benefits. Analysts expect significant consolidation and restructuring across the sector in the years ahead.
This article was prepared with assistance from an AI tool; the final content was edited and verified by our journalist.
