Big Tech’s $3 Trillion AI Bet Faces A New Skeptic: Credit Markets

On Tuesday, Nikkei revealed that AI-driven off-balance sheet debt by five tech titans—Google’s owner Alphabet, Microsoft, Amazon, Meta and Oracle—amounts to $1.65 trillion, more than the roughly $1.35 trillion in their on-balance sheet debt. On Monday, the cost of protecting Oracle’s debt against default reached a record-breaking high as Goldman Sachs’ trading desk noted “signs of panic” among investors lending to AI companies according to Moomoo.

At the heart of the market’s growing unease surrounding the AI race is a potential mismatch between supply and demand: taking on trillions of dollars in debt to finance the construction of exponentially greater AI-related capacity on the one side to meet limited consumer and business demand on the other. CBS News reported that while generative AI subscriptions are up 155% compared to last year, that increase means that only 2% of U.S. households have one. On the business side, TechCrunch observed that companies are pressuring employees to use fewer tokens in an attempt to reign in explosive AI costs.

The dot-com era in the early 2000s saw a similar dynamic. During the late 1990s, WorldCom’s chief scientist Michael O’Dell famously claimed that internet traffic was “doubling every 100 days” based on a best-case scenario projection developed by WorldCom’s internet service provider a CNBC later investigation found. Unfortunately, that soundbite went viral in the business and investment worlds without the original context or caveats and based on this misunderstanding, WorldCom, Global Crossing, and other telecom companies laid downs tens of millions of miles of fiber-optic cable with the help of investors financed the build out.

To be sure, internet traffic was growing at a rapid rate of approximately 100% per year but the infrastructure build outstripped that rate of growth by several orders of magnitude. When WorldCom collapsed under a massive debt burden due to accounting fraud in 2002, fiber optic utilization rates were still 5% or less and would remain below 50% until into the 2010s with the rise of YouTube and smartphones.

Although the fiber-optic build out ultimately paid off for consumers and businesses decades later, it broke many of the companies that did the work and the investors that financed it lost billions. Besides WorldCom, 22 other communication companies went bankrupt when the dot-com bubble burst and telecom sector bondholders at that time were only able to recover perhaps 20% of what they were owed according to Quartz.

The reason Oracle—survivor of the dot-com era—has become the current focal point of investors’ fears concerning the soundness of AI-related debts is because they in many ways are the weakest player among the tech titans. Based on S&P estimates, Half of Oracle’s future revenue is projected to come from OpenAI, a company that not only has never turned a profit but regularly losses over $10 billion every year reported CNBC. Their cash flows are also much weaker relative to the growth of their AI spending, reaching negative $23.7 billion for 2026 as noted by Yahoo! Finance. As Bloomberg Intelligence tech credit analyst Robert Schiffman put it: “People are starting to ask questions on how much debt capacity do they really have, and the answer starts becoming that they need to show progress in terms of AI monetization.”

The other tech titans will face similar questions as they move towards negative cash flows driven by AI-related spending.