Oracle's aggressive push into artificial intelligence infrastructure has pushed its debt into junk bond territory. The company's latest earnings report showed increased revenue and profits, but it is doubling down on AI spending and borrowing heavily to fund it. Capital expenditures in the most recent quarter jumped 200% year over year and exceeded Wall Street expectations by 50%. Management now projects roughly $50 billion in capital expenditures for fiscal 2026, up from a previous estimate of $35 billion.
To finance this buildout, Oracle raised $18 billion in one of the largest bond sales in tech sector history in September and plans to target even higher amounts in the coming year. Although the company maintains an investment-grade credit rating, yields on its bonds have slipped into junk bond territory. The price of Oracle's five-year credit default swaps, which act as insurance against debt default, has tripled in recent months and is trading at levels not seen since the global financial crisis.
This borrowing is largely driven by Oracle's reliance on one customer: OpenAI. OpenAI has committed to spending $300 billion over the next five years on Oracle's services. OpenAI remains deeply unprofitable, with annualized revenue roughly a fifth of its annual commitment to Oracle. The company will need to continue raising unprecedented capital to meet its obligations.
Meanwhile, AI data center operator CoreWeave has tripled its revenue over the past year, but this growth is financed with substantial expensive debt. Including lease obligations, CoreWeave carries about $15 billion of debt, nearly four times its total revenue over the last 12 months. Last quarter, the company paid $311 million just to cover interest on its debt, an expense that has increased nearly 200% year over year. Interest costs now exceed a fifth of its total revenue and are roughly six times its gross profit.
CoreWeave also faces significant customer concentration, with nearly all revenues coming from a handful of customers, including Microsoft and other hyperscalers. If AI demand cools, these customers, who are also competitors, may bring workloads in-house, eliminating CoreWeave as a middleman. The company has a $6.3 billion backstop agreement with Nvidia, but this may not be sufficient if demand for AI processing power declines meaningfully.
Other companies, such as neocloud provider Nebius, AI hardware providers like Super Micro Computer, and startups like Oklo, Rigetti Computing, and D-Wave Quantum, could also be at risk if an AI bubble bursts. The scale of spending in the AI sector is unprecedented, and the fervor mirrors past bubbles. If this is a bubble, CoreWeave and Oracle are unlikely to survive its bursting, according to the analysis.