In late September, news circulated that the hyperscaler Oracle had sent a force majeure notice to the developer of its New Mexico data center campus, Project Jupiter, seeking to delay payments on the project should it fail to come on line in 2028 as planned. In response, Oracle took to X, writing that “force-majeure notices are commonplace in developments of this scale. They do not, by themselves, establish a project delay or change delivery expectations.”
The hiccup was one in a series. According to Financial Times reporting from September, loans tied to the project had been “privately quoted at 89 to 91 cents on the dollar by syndicated banks like Santander and Jefferies,” reflecting, more broadly, Oracle’s barely-above-junk credit rating and its place as the company to watch for those who are concerned about an AI bubble bursting soon. That fact is made crystal clear by both press surrounding the subject and the strength of the market for Oracle credit-default swaps (essentially insurance against Oracle defaulting on its debt).
Things have since gotten worse for Oracle. Last week, overlooked by many amid the midterms and unfolding drama at Nvidia, the Financial Times reported that Oracle canceled an estimated $7 billion chip deal with the Chinese tech giant Tencent under pressure from the Trump administration. Two days later, Bloomberg reported that Microsoft had been winding down a 2023 deal to rent AI computing from Oracle that had originally been arranged to meet the computing demands of its Copilot AI assistant.
Microsoft reportedly walked away because Oracle’s Nvidia chips were already outdated—a growing concern for investors financing expensive AI infrastructure that risks becoming obsolete before it can pay for itself. As one fund manager put it colorfully to the Financial Times:
“It’s like financing a fax machine and then someone invented email.”
If Microsoft was in fact dissatisfied with the chips, this marks one of the more significant manifestations of the chip-lifecycle concern, which so far has seemingly lived in analysts’ forecasts. It also bears out the problem of mismatched timelines. Oracle is operating on decade-plus horizons, in one sense, with its data center development. But the pressure to meet short-term obligations with big swings and the need for labs to develop their frontier models mean that the build-out goings-on are happening in large part on very different timelines—year to year or even month to month.
Oracle predictably pushed back on X, saying demand for its AI infrastructure remains strong, even for older Nvidia chips. The company said that in its first quarter (which runs from June through August), it renewed or resold capacity using GPUs that were mostly at least four years old, securing long-term contracts at a 20% premium over previous pricing.
While it might seem that Oracle is continuously and unnecessarily sticking its neck out on social media to combatively address every single concern about its financing practices, the company’s posting habits make sense in an environment characterized by rising interest rates and reemergent concern about the AI bubble. Oracle has financed its build-out with more leverage than its peers and similarly negative cash flow. And it’s failed to quell concerns about its borrowing. Earlier this month, it burned through the entire $20 billion fundraising cap meant to last through fiscal 2030, just five days after announcing it. And it’s not only the leverage itself that has watchers concerned—or, doubly the economic consequences of the war in Iran—but also electoral politics.
Compounding Oracle’s troubles are two broader threats to the AI build-out: a growing political backlash against data centers and mounting investor anxiety over infrastructure spending.
In the midterms, voters approved 40 of 41 ballot measures aimed at restricting data center construction. In New Mexico, Governor-elect Deb Haaland pledged to impose a moratorium on new large-scale data centers, declaring during her campaign that Oracle’s Project Jupiter would “not exist” if she were governor.
Meanwhile, the tech-heavy Nasdaq has fallen into a correction, interrupted only occasionally by rallies fueled by Anthropic’s forthcoming IPO and positive earnings from CoreWeave. On Tuesday, SoftBank reported a $7 billion quarterly loss, attributed in part to U.S. data center spending, sending shares of infrastructure companies like the turbine maker GE Vernova lower. On Wednesday, shares rallied again, with the press once more crediting CoreWeave and excitement around Anthropic’s IPO, which has been described as “too big to fail.”
While Oracle has done little to generate any good press, Larry Ellison has not helped. In September, an Oracle proxy filing revealed that he had pledged billions in shares as collateral for personal loans, which reporting would later reveal was to assist with the equity financing in his son’s acquisition of Warner Bros. Discovery. While Oracle and its board seem not too concerned—Ellison is the only executive permitted to pledge stock as collateral—the move’s specifics received some warranted attention.
By pledging stock as collateral rather than selling it, as he had reportedly first intended, Ellison appeared to be attempting to keep his full stake intact while avoiding the headlines a sale would inevitably generate. This in itself isn’t bad news for Oracle, but such a high level of borrowing nonetheless raises more conventional concerns about corporate governance. By tying his personal borrowing to Oracle’s fluctuating share price, Ellison has created a scenario in which a falling stock price could prompt lenders to demand more collateral or force a sale that would push the stock even lower. Unlikely, true—Oracle has written in SEC filings that Ellison has “the financial capacity to repay his personal term loans without resorting to the pledged shares”—but it’s hardly what Oracle needs right now.