$300 billion in debt and collapsing stock prices: what Oracle's bet on AI is costing it

Oracle, one of the largest U.S. software makers and server hardware vendors, is actively engaged in the artificial intelligence race, and this is already worrying investors. The company has a growing debt load and risks concentrating large customers. Its case mirrors the story of the development of AI itself: investors first skyrocketed the quotations of companies related to it, and then began to wonder whether a giant bubble had burst in this sector.

On Sept. 9, Oracle reported in its fiscal 2026 first-quarter filing that its computing capacity orders - that is, actual contractually confirmed future revenue - increased 359% to $455 billion, driven by contracts with AI developers. Most of that amount - $300 billion - will come from OpenAI, with which Oracle has signed a five-year agreement to supply computing power. And it's the largest cloud contract in history. To fulfill its obligations to OpenAI, Oracle has signed at least five long-term leases for data centers in the U.S., which are at various stages of construction.

On September 10, Oracle shares soared 36%, making its co-founder and chairman Larry Ellison the richest man in the world. He even briefly surpassed Tesla CEO Elon Musk ($386 billion vs. $384 billion, according to the Bloomberg Billionaires Index).

"AI is fundamentally changing Oracle and the entire computer industry, although not everyone fully realizes the scale of the approaching tsunami," Ellison himself told investors(quoted in the Financial Times).

But the excitement didn't catch on with everyone. Brent Till, an analyst at Jefferies, told the newspaper at the time: It remains unclear what Oracle's profit margin on these deals will be and whether the company will be able to fully fulfill the contracts, which require huge investments. Oracle's promise to its AI customers, Till said, still goes something like this: "We'll build it. Trust me."

A tsunami did indeed unfold, but not the kind that Ellison had promised. After the initial euphoria, it was as if the market decided: "Let me not believe you" - and the company's stock has plummeted below the level from which the rally started. This is unprecedented, The Wall Street Journal reports. In the nearly 40 years of Oracle's presence on the stock exchange, its shares have three times risen in price in one day by 30% or more. And until now, they have never returned to the closing level before such jumps, according to Dow Jones Market Data.

For equipment in data centers that will provide computing for AI models, Oracle will have to raise huge funds. The company has already placed bonds for $18 billion, the volume of bonds it has issued has already exceeded $100 billion - the highest level of debt among large technology companies with investment grade credit ratings. In addition, Oracle is now discussing with U.S. banks the possibility of raising another $38 billion, the FT writes.

If you add future data center lease payments to this type of debt, which will continue to grow, Oracle's total debt will approach $300 billion in 2028, more than doubling, according to Morgan Stanley's forecast.

"The market has moved to the position of, 'You have to prove to me'" that it is capable of meeting its obligations, Rishi Jaluria, an analyst at RBC Capital Markets, told WSJ.

Investors and analysts point to Oracle's over-reliance on OpenAI. S&P Global warns that in 2028, a third of Oracle's revenue will be dependent on this customer.

That said, the business models that will make money from Xi are still unclear.

OpenAI itself predicts that because of rising computing costs (it has pledged to invest $1.4 trillion in AI infrastructure over the next eight years), its operating losses will grow to $74 billion in 2028, about three-quarters of revenue, and its free cash flow won' t turn positive until 2030.

"It's a very different business model to what investors value in cloud services," Alex Heisle, an analyst at Rothschild & Co Redburn, told the FT. He said Oracle's deals with AI companies, which also include Ma, xAI and Nvidia, "look fantastic when you look at the revenue numbers, but they are very capital intensive, so they create very little value".

In fact, Oracle has become a bet on AI development, but as a trader specializing in downside plays told the FT, "the market is clearly making it clear that it is no longer interested in companies that are relentlessly spending money on artificial intelligence."

Shares of technology giants, which until recently were soaring mainly due to the euphoria associated with the deployment of AI infrastructure, have fallen heavily in recent weeks. Market participants began to fear that a bubble had inflated in the sector.

The fall of Oracle turned out to be the strongest. By November 19, from their historical peaks, shares of Broadcom, for example, fell by 8.2%, Nvidia - by 9.9%, Microsoft - by 10.1%, Amazon - by 12.3%, AMD - by 15.4%, Meta - by 25.3%, and Oracle - by 31.3%, according to WSJ.

The sell-offs did not stop there: Oracle securities lost 11.9% on Thursday and Friday, November 20 and 21, while the Nasdaq-100 technology index lost 1.63%. On November 19, Nvidia's report on quarterly profit growth improved investor sentiment, but their optimism was not enough for long. Nvidia shares, which jumped by 4.9% at the beginning of trading on November 20, ended trading on November 20 with a decline of 3.15%. On Friday, the sell-off continued, but then the market turned around. The Nasdaq-100 closed Friday with a slight gain of 0.77%.

"It will take time to recover from the kind of decline we're seeing now - and that's after a change in sentiment," the WSJ quoted Ross Maefield, investment strategist at U.S. investment firm Baird, as saying.

About half of institutional investors who participated in Bank of America's monthly survey (November 7-13) said that there is a bubble in the AI segment. The main risk that could lead to extreme changes in quotations was named by 45% of managers as a bubble in the stock market of AI-related companies. At the same time, 53% believe that a bubble in this segment has already formed.

For the first time since 2005, a majority of respondents expressed concern about overinvestment by companies, "driven by concerns about the scale and financing of the AI investment boom," BofA noted.

Morgan Stanley estimates that technology companies may spend $2.9 trillion on data centers in 2025-2028, but their own funds will be enough to finance only half of these investments. A significant part of the remaining funds will have to be borrowed.

And this is in a situation where the real return on these investments has not yet been confirmed, points out market veteran investor Mark Mobius. In his opinion, the AI sector is in the overheated zone, the gap between expectations and profits has become too large, which may lead to a correction in the shares of AI companies by 30-40%. This, however, will open up new opportunities for long-term players, Mobius believes.

Surprisingly, famed short-shorter Carson Block, whose firm Muddy Waters Capital scouts out companies whose stock is down for a play, isn't willing to do that with tech giants.

However, he explained his position by the popularity of these stocks among investors and exchange-traded funds, who simply "buy the index every day at any price if they have an influx of money." Block is looking for targets to open short positions among smaller companies: "There are a lot of AI-related companies, ones that are just pretending to be in the AI business, and those are the ones to look at."

"A bubble usually results from overinvestment in a new sector, where more capacity is created than companies can afford to build. Overvaluations become a source of future problems," argues Jon Treacy, publisher of investment newsletter Fuller Treacy Money.

He believes a conclusion about whether there is a bubble in the AI sector can be drawn by finding answers to two questions: how long an expensive chip will last before it has to be replaced, and whether chip buyers can generate revenue growth to justify the cost of maintaining the service.

"If the chips last long enough for revenues to exceed the cost of running the service, then there is no bubble," writes Treacy. - If, however, the chips need to be replaced regularly and the costs are not covered, then we are dealing with a bubble."

This article was AI-translated and verified by a human editor