For a few brief weeks in September and October, Oracle was on top of the artificial-intelligence revolution.
The storied business-software and database company had a burgeoning $300 billion relationship with OpenAI that would more than double its revenue in just three years, according to management's new projections. A huge stock surge put Oracle on the doorstep of a $1 trillion valuation. Founder Larry Ellison reclaimed the title of world's richest man, and laid out a vision that put Oracle $(ORCL)$ at the forefront of AI innovation - including the key inferencing stages where trained models use new data to make predictions.
"Oracle is aggressively pursuing the inferencing market, as well as the AI-training market," Ellison said on the company's September earnings call with investors. "We think we are in a pretty good position to be a winner in the inferencing market, because Oracle is by far the world's largest custodian of high-value private enterprise data."
Oracle's accomplishments were the fruit of an almost decades-long investment into its cloud-computing business. Then it all came crashing down.
Shares of Oracle have given up their gains and then some, falling 36% from their September peak. Its credit default swaps, or the price of insurance on its bonds, have widened sharply in recent weeks, an indication that the market perceives the company is on increasingly shaky financial footing. And traders have upped their bets against Oracle's shares: While only 1.5% of its total float, short interest on the stock has climbed from 15.7 million shares at the end of May to 24.4 million shares currently, according to FactSet.
The problem is that while the company wants to compete with the lions of cloud computing, getting there will be expensive, and Oracle doesn't have the cash flow to fund its dreams. Instead, it has turned to the debt markets and leveraged up its balance sheet. Wall Street has become increasingly worried about the debt that Oracle is taking on to become a major partner to OpenAI, which itself is burning cash in pursuit of AI profits that have so far been elusive.
The AI arms race will likely cause the tech "hyperscalers" - like Meta Platforms (META), Amazon.com (AMZN), Google parent Alphabet $(GOOGL)$ $(GOOG)$, Microsoft $(MSFT)$ and Oracle - to spend $2.9 trillion in capital expenditures to build out the technology, Morgan Stanley estimated in a July report, meaning that at least $1.5 trillion will have to come from debt and other financing arrangements.
More than any other Big Tech company, Oracle is testing the limits of this debt-fueled spending spree and is serving as a canary in the AI coal mine, financial analysts say. On Wall Street, the company's credit default swaps have become a way to monitor the sustainability of the debt load being taken on by Big Tech to fuel its AI ambitions, and guard against it. Look at Oracle if you want to assess the health of the AI investment boom, they say.
Compared to its peers, Oracle has much weaker cash flows, higher debt levels and lower credit ratings. It also has significantly higher counterparty risk from OpenAI, Nathan Liddle, a fixed-income research analyst at Columbia Threadneedle Investments, told MarketWatch.
"Oracle has far less flexibility than its peers in this investment cycle," Liddle said.
The cost of competing
In September, Oracle issued $18 billion of bonds, one of the biggest-ever corporate bond issuances by a tech company. The maturities on those bonds, or the date at which they need to be fully repaid, extended out 40 years to 2065.
Oracle is far from the only Big Tech company borrowing money to finance ambitious AI spending; Alphabet and Meta both closed massive bond offerings in 2025. But Oracle is in a more precarious financial position than any of the other hyperscalers, Liddle emphasized. Alphabet and Meta both have pristine AA credit ratings from S&P Global Ratings - giving them more borrowing power compared to Oracle, which has a BBB rating hovering at the lower end of investment grade. And Oracle's debt-to-equity ratio exceeds 462% while the other hyperscalers' ratios are under 50%, according to FactSet.
What originally excited Wall Street about Oracle's AI opportunity was the company's September disclosure that its remaining performance obligations, or future contracted revenues not yet recognized, had surged nearly 360% to upwards of $450 billion. By mid-October, that number had exceeded $500 billion. That meant that OpenAI and other companies planned to do more than $500 billion worth of business with Oracle over time, though Oracle would have to invest heavily in AI infrastructure to translate those contracted agreements into actual revenue, D.A. Davidson Gil Luria wrote in a September note following Oracle's initial disclosure.
But the optimism deflated significantly in recent weeks, as questions arose about whether Oracle's customers could renege on these contracts and leave it holding the bag for billions in underutilized infrastructure.
For now, Oracle is busy building data centers in Michigan, Wisconsin and Texas. With nearly $6 billion in negative free cash flow for the 12 months ending Aug. 31, the company is burning far more money than it generates through its business, leading it to take on significant debt to fund its AI buildout.
Oracle declined to comment for this story.
The $18 billion September issuance is likely to be just the beginning for Oracle's borrowing spree. In a November note, Morgan Stanley analyst Lindsay Tyler estimated that the company's combined debt and leases could rise to $290 billion over the next three years, a massive leap from the roughly $100 billion it held earlier this year.
Further debt issuances could bring Oracle to a "tipping point where the [investment-grade bond] market is reluctant to continue to lend," Columbia Threadneedle's Liddle said. That would increase the cost of borrowing for Oracle, creating a vicious spiral accelerating the company's cash burn and further increasing its need to borrow.
Since inking the $300 billion agreement with Oracle, OpenAI has gone on its own $1 trillion dealmaking spree, and investors are now questioning the stability of Oracle's future revenues from the private company behind ChatGPT. In a November note, D.A. Davidson's Luria characterized Oracle as a "pawn" in OpenAI's "grand game of 'fake it 'till you make it,'" borrowing money for a customer that might never fulfill its promise.
"Earlier in the year, when someone would announce a deal, that money was going to come through no matter the time horizon," said Michael Treacy, head of market risk at Apex Fintech Solutions. "Only recently have investors started to actually ask the question: Will these payments come to fruition? Through the life of these contracts, will the vendor actually get paid?"
The concern is apparent in Oracle's five-year credit default swap (CDS) spreads, which have decoupled from the broader investment-grade Credit Default Swap Index - widening from 60 basis points at the end of October to 125 basis points as of last week, while the underlying index remains consistent. That means it costs significantly more to buy protection against the company defaulting on its debt. In fact, Oracle's five-year CDS spreads closed at 128 basis points on Dec. 5, the highest level since the height of the financial crisis in March 2009, according to Bloomberg.
Oracle's five-year CDS spread reached a recent high of 128 basis points last week.
The spread could reach 150 basis points in the near term, Morgan Stanley's Tyler predicted, as uncertainty remains about Oracle's future funding plans and whether it can count on OpenAI to deliver the promised revenue.
While both Liddle and Tyler believe the CDS pricing is primarily due to Oracle-specific risk, others believe it's a signal of systemic concerns over Big Tech's appetite for AI leverage. The company's CDS movement is emerging as a way to "hedge the entire debt cycle within AI," Apex's Treacy said. He added that the "presumption is that if there are tremors or fears or issues in the debt market, Oracle's CDS and debt will be the most correlated with that broader narrative" compared to the "Magnificent Seven" names - companies like Google or Microsoft that have the cash flows to withstand a downturn.
Matthew Unterman, managing director at S3 Partners, wrote in a note earlier last week that Oracle's increasing short interest and widening CDS spreads indicate the company is "trading as a proxy for investor anxiety around AI infrastructure leverage, making it a barometer for broader credit risk in the megacap tech complex."
Even the most well-capitalized companies are receiving scrutiny for their massive AI spending, as traders begin to bet on Nvidia (NVDA) and Meta defaulting on their debts. A liquid market for Meta's CDS sprang up after the company's debt issuance in late October. Nvidia's CDS market emerged after key partner CoreWeave (CRWV) announced delays in data-center construction in November.
The poster child of AI bubble fears
To avoid taking on additional debt, some companies have turned to alternative financing strategies such as private credit, the lightly regulated Wall Street industry that lends money it raises from investors as opposed to banking deposits.
Most notably, in the largest-ever private-credit transaction on Wall Street, Meta entered into a $27 billion joint venture with Blue Owl Capital $(OWL)$ in October to finance the construction of its Hyperion data center in Louisiana. The deal created a special-purpose vehicle called Beignet Investor LLC, which holds the debt and does not appear on Meta's balance sheet. Upon completion of the data center, Meta will enter into short-term leases with the Beignet entity.
There's been much speculation over whether Oracle will follow Meta's footsteps to derisk its financial profile. However, D.A. Davidson's Luria believes private-credit financing might have the opposite effect - creating more opacity that would set off red flags for investors.
"Oracle could pursue off-balance-sheet arrangements with special-purpose vehicles going forward, but that would be even more concerning, since in many ways they are now acting as OpenAI's off-balance-sheet arrangement," Luria told MarketWatch over emailed comments.
Instead of taking on debt to build its own data centers, OpenAI has opted to rent the infrastructure and outsource the financing responsibility to Oracle.
Oracle will have an opportunity to address funding concerns on Wednesday, when it reports earnings for the second quarter of its 2026 fiscal year.
"The main thing that Oracle could do to quell fears would be to start demonstrating positive free cash flow and return on the investments that they're making," Apex's Treacy said. Although that outcome seems highly improbable in the near term, announcements of new contracts and updates on customer diversification and revenue timing could help improve the current narrative. More clarity on the company's future fundraising plans would also ease investor concerns, Morgan Stanley's Tyler added.
Right now, it's unclear how binding OpenAI's $300 billion contract is with Oracle, or the extent to which Oracle can pull back AI spending if demand doesn't show up. "To the extent there's flexibility on either side, [Oracle] can manage a changing outlook for OpenAI," Columbia Threadneedle's Liddle said.
Beyond contract details, the ultimate question for credit markets is whether Oracle can maintain its investment-grade credit rating. If the company "were to give even minimal commitments to the balance sheet, that could have an outsized impact on the sentiment, and therefore likely the price of the bonds," Liddle added.
As Oracle emerges as the poster child for the AI bubble, the financial landscape beneath it is bending to accommodate a future of perpetually higher leverage. Credit-rating agencies and other financial institutions are ushering in new, more lax borrowing frameworks as AI becomes too expensive to play by the old rules.
Meta's Beignet offering doesn't appear as debt on its balance sheet, but the terms of the agreement are structured incredibly similar to a debt instrument. While Meta's leases only span four years, it's effectively locked in for 16 years via a "residual value guarantee" clause that forces the company to make a cash payment to investors if it chooses not to renew the lease early.
Meta did not respond to a MarketWatch request for comment.
"There's reason to think that that should be fully accounted for on Meta's balance sheet, but the agencies are not making that adjustment today," Liddle said.
Rating agencies are also giving Oracle more leeway to maintain its investment-grade credit rating. In September, S&P Global loosened the downgrade trigger for Oracle's credit rating to a ratio of 4 times debt-to-Ebitda, from a ratio of 3.5 times debt-to-Ebitda previously, citing how the company's AI business transition would require significant upfront investment.
The financial goalposts are shifting as companies prepare to take on increasing amounts of debt. "We're in the early stages of the AI debt cycle," Treacy said. "What we're seeing today in terms of leverage coming into AI is only going to grow."
Comments