The Story
Let's imagine you're a landlord in Bengaluru.
A hot young startup walks in and says, "We need a huge office. We'll pay you handsomely for five years." You're thrilled. But you don't own a building that big. So you sign a 15-year lease on a new tower, borrow money to fit it out with furniture and servers, and wait for the rent to roll in.
Now ask yourself one question. What happens in year six?
That, in a nutshell, is the question hanging over Oracle right now.
Last Thursday, Oracle sent a legal letter about its giant New Mexico data center campus, called Project Jupiter. It was a "force majeure" notice, which declares that an uncontrollable event is getting in the way of a contract. In plain words, it's the contract version of "this delay isn't my fault, so don't make me pay yet."
Markets didn't like it. Oracle's stock fell 4.4% on the day, part of a 29% slide this year. And the bond market reacted even more sharply, which is the part we really want to talk about.
But to understand why a single letter caused so much noise, we need a little context first.
#1 What actually happened
A year ago, Oracle was the darling of the AI boom. It signed massive deals to rent out computing power to AI companies, most famously OpenAI. The stock went as high as $345.72 last September. Today it trades around $139.
The first big blow came in July. S&P Global Ratings cut Oracle from BBB to BBB-, one notch above junk.
What's that, you ask? Well, you see, credit ratings are like a report card for borrowers. Anything from BBB- upwards is "investment grade," meaning pension funds and insurers can comfortably lend to you. Anything below is "junk," or high-yield. Oracle is now standing on the very last step before that line.
S&P's reasons were blunt: rising AI spending, a forecast free cash flow deficit of nearly $42 billion for fiscal 2027, leverage in the mid-4x range (above the 4x it's comfortable with for a BBB), and OpenAI accounting for roughly half of Oracle's then $638 billion backlog.
Then came September. Oracle's quarterly results on September 10 looked excellent on the surface. Revenue was a record $19.3 billion, up 30%, and cloud infrastructure revenue jumped 121% to $7.4 billion.
But there's a catch.
Capex, the money a company spends on physical stuff like buildings, chips and power equipment, hit $28.5 billion in a single quarter, against $8.5 billion a year earlier. Free cash flow was negative $5.4 billion.
And then the Jupiter letter landed. Within days, Oracle's credit default swaps hit a record, the yield on its 2056 bonds crossed 8% for the first time, and about $18 billion of loans tied to Project Jupiter traded below face value.
A credit default swap, or CDS, is basically insurance against a borrower not paying you back. When its price jumps, it means more people are worried enough to buy that insurance.
So the question isn't whether Oracle's AI business is growing. It clearly is. The question is whether the money arrives before the bills do.
#2 The four mismatches at the heart of the problem
We read through the rating agency notes, Oracle's own filings, bond market reports and analyst takes. The same handful of worries kept popping up across all of them. We'd put them into four buckets.
The time mismatch
Remember our Bengaluru landlord? Oracle is doing exactly that, at a scale that's hard to picture. Its latest quarterly filing lists $288 billion of additional data center lease commitments, starting between late 2026 and fiscal 2029, for terms of fifteen to nineteen years, that aren't yet on its balance sheet.
For context, the lease liabilities already on its books are about $34.6 billion. So the commitments waiting in the wings are more than 8 times bigger than what you see today.
Meanwhile, the customer side is much shorter. The same filing says Oracle's cloud arrangements generally run one to five years and are renewed at the customer's option. Long promises going out, shorter promises coming in. That's exactly the scenario S&P flagged: if demand or a key customer's ability to pay shifts, Oracle could be left with stranded assets and costly re-leasing.
The customer mismatch
Oracle's backlog (accountants call it RPO, or remaining performance obligations, which basically means signed contracts it hasn't been paid for yet) is a jaw-dropping $664 billion. But roughly half of it traces back to one customer, OpenAI.
And OpenAI is burning cash at a remarkable pace. It reportedly expects to burn around $278 billion between 2026 and 2030, according to a Financial Times report. So Oracle's biggest customer is itself relying on investors to keep funding it. If that tap slows, Oracle feels it.
The cash mismatch
Here's a number that doesn't get enough attention. Oracle's filing says it expects to recognize about 13% of that $664 billion backlog as revenue in the next 12 months, which is roughly $86 billion. Another 37% (about $246 billion) comes in months 13 to 36, and 34% (about $226 billion) in months 37 to 60.
Now compare that with Oracle's spending plan. Full-year capex guidance for fiscal 2027 is $90 to $95 billion. In other words, Oracle is spending roughly an entire year's worth of revenue on construction, while most of the revenue that pays for it arrives in 2028 and beyond.
The physical-world mismatch
This is the one Jupiter exposed. You can sign a contract in an afternoon. You can't build a power supply that fast. The 2.45-gigawatt campus is meant to run on gas-powered fuel cells, but the pipeline serving it has slipped nearly six months to February 2027, after regulators repeatedly denied permits. The site has also faced protests and lawsuits over its environmental impact.
What does the notice actually buy Oracle? If both sides agree a force majeure event has occurred, Oracle could get a three-year delay on rent once payments begin.
To be fair, Oracle insists nothing is wrong. It told CNBC that Project Jupiter remains on schedule, and the developer says such notices are routine for projects this size. That may well be true. But the bond market is paid to imagine the bad version of every story. And right now, it's imagining.
#3 So where does the money come from?
Oracle basically has five taps it can open.
- Its own cash machine. Oracle still runs a very profitable software business, and operating cash flow hit a record $23.1 billion last quarter. Sounds great, right? Not so fast. $11.4 billion of that was customers paying in advance, money Oracle will have to earn later by delivering the service.
- Borrowing. This route is getting narrower. In fiscal 2026 Oracle raised $43 billion in debt. Interest expense last quarter was $1.43 billion, up from $923 million a year earlier, a jump of over 50%. And every extra dollar of debt pushes up the leverage ratio S&P is watching. One more notch down and Oracle becomes junk.
- Selling new shares. Since borrowing is harder, Oracle has leaned on shareholders. It sold 141 million new shares for $19.9 billion last quarter, taking its share count from about 2,880 million to 3,024 million. That's roughly 5% more slices of the same pizza. It has also paused buybacks: zero repurchases in the quarter.
- Making the customer pay upfront. This is Oracle's cleverest move. Some customers now have to buy or bring their own AI chips into Oracle's data centers. Think of it as a landlord saying, "Sure, I'll give you the office. But you bring your own furniture." Most of the more than $30 billion in new contracts signed last quarter came through prepayment or this bring-your-own-hardware model.
- Letting partners carry the building. Oracle doesn't own Jupiter. It's the tenant. The campus is being built by Stack and BorderPlex, with plans to invest up to $165 billion. Developers, private credit firms and banks put up the money, and Oracle signs the long lease. That's why the loans financing Jupiter matter so much, even though they aren't technically Oracle's debt.
#4 How Oracle plans to manage it, and the catch
Oracle's argument goes something like this: yes, we're spending a lot now, but each data center becomes a cash cow once it's up and running. CFO Hilary Maxson said each project converts roughly 100% of its post-tax EBITDA into free cash flow shortly after ramp-up, making the business "somewhat self-funding."
Co-CEO Clay Magouyrk went further, arguing that the capital doesn't all have to be Oracle's own capex. To make that visible, Oracle now reports a new number called "net cash capex." Full-year capex is $90 to $95 billion, but after customer prepayments and financing arrangements, net cash capex is guided to stay under $70 billion.
It's also cutting costs. Its 2026 restructuring plan will cost up to $2.1 billion, and management has since added about $700 million more for further actions. Much of that is severance.
And that's a genuinely reasonable plan. If the data centers fill up, get paid for, and run for years, Oracle comes out the other side with a huge new business.
But here's the thing.
When we lined up all these sources side by side, a pattern jumped out. Almost every tool Oracle is using doesn't make the risk vanish. It just moves it somewhere else.
Selling shares moves the risk to shareholders. Prepayments bring cash forward but create an obligation to deliver later. Letting developers build the campus moves the construction debt off Oracle's books, but Oracle is still the one promising 15 years of rent. And when a site like Jupiter hits trouble, the risk comes right back to Oracle's doorstep. That's exactly what the force majeure letter shows.
Even the "net capex" number tells you something. The gap between $28.5 billion and $18 billion last quarter didn't disappear. Someone else paid it, on the condition that Oracle delivers.
#5 The two markets that can't agree
This might be the most interesting pattern of all.
On the day S&P cut Oracle's rating, the stock actually rose about 2.7%, while bond investors demanded higher interest to hold Oracle's debt. Even now, Wall Street's average price target is around $245, well above where it trades.
Why the split? Well, stock investors get the upside. If Oracle's AI bet pays off, the share price could multiply. Bond investors get no upside at all. The best they can hope for is getting their money back with interest. So they obsess over the downside.
And the bond market is quietly telling you what it thinks. Oracle's own filing shows its $125 billion of long-term borrowings had an estimated market value of $105.7 billion at the end of August. That's roughly 85 cents on the dollar, and that was before the Jupiter news pushed yields higher.
The next domino to watch is Moody's, which rates Oracle Baa2 with a negative outlook, citing the risk tied to its largest AI customer. A cut from Moody's wouldn't push Oracle into junk right away, but it would add to the pressure.
So what does this mean for you?
If you don't own Oracle shares, you might wonder why you should care.
For one, Oracle has deep roots in India. It counts Oracle Financial Services Software, listed on Indian exchanges, among its majority-owned subsidiaries, and it employs a large workforce here. When a parent company is watching every rupee, cost-cutting tends to travel. Oracle cut 21,000 jobs in fiscal 2026 to fund its AI buildout, and its filings point to more restructuring ahead.
But the bigger lesson is about the AI boom itself. Oracle has become the credit market's canary in the coal mine for AI spending. If demand stays strong, its bet looks brilliant in hindsight. If demand wobbles, Oracle is the company most exposed, because it borrowed and leased the most relative to its size.
So is Oracle in trouble? Not today. It has $36.4 billion in cash, a profitable software business, and customers genuinely lining up. What it doesn't have is much room for error. Everything has to go right: OpenAI has to keep raising money, power has to reach the sites on time, and the rating agencies have to stay patient.
Whether all three happen together is something only time will tell.
Until then…
If this story helped you understand why a data center in New Mexico is rattling the global bond market, share it with your friends on WhatsApp, LinkedIn and X.
Disclaimer: This story is not investment advice. The examples and data shared here should not be taken as recommendations to buy or sell any asset. Investments are subject to market risk, and past performance does not guarantee future returns. Always do your own research or consult a qualified financial advisor before making investment decisions. Do not YOLO.
Sources
- Oracle Form 10-Q, fiscal Q1 FY27 (Jun–Aug 2026): leases, backlog timing, debt fair value, share issuance, restructuring
- Oracle Q4 and FY2026 results release
- Coverage of S&P's July 2026 downgrade
- TechCrunch on the Project Jupiter notice
- TipRanks on record CDS and bond yields
- Q1 FY2027 earnings call highlights
- Oracle's 2026 financing plan