Oracle Spent $250 Billion on AI Data Centers — and Now the Bill Is Coming Due

Oracle Spent $250 Billion on AI Data Centers — and Now the Bill Is Coming Due

Let me tell you something I've been watching unfold all week, and every new headline makes me shake my head harder.

Oracle — the enterprise database giant that reinvented itself as an AI hyperscaler — is in trouble. Not "missed earnings" trouble. Structural trouble. The kind that doesn't get fixed with a good quarter.

In the last two weeks alone: New Mexico regulators rejected the pipeline for Oracle's $165 billion Project Jupiter data center for the second time. S&P Global downgraded Oracle's credit rating to BBB- — one notch above junk. Credit default swaps hit an 18-year high. The stock is down 64% from its September 2025 peak. And free cash flow? Negative $23.7 billion for the fiscal year on $55.7 billion of capex.

I run a hosting business. I've never had to raise $40 billion in debt to build a data center. But I know what happens when your expenses outrun your cash for too long. You run out of runway. And Oracle is burning through cash faster than almost any company in history.

Oracle's $250 Billion AI Bet Is Hitting Reality — Hard

Santa Teresa, New Mexico — July 21, 2026 — Let me connect the dots on what's happening, because the mainstream financial press is covering each story separately and missing the bigger picture.

The $165 Billion Data Center That Can't Get a Pipeline Permit

Project Jupiter. That's the name Oracle gave its massive AI data center campus in Santa Teresa, New Mexico — 1,400 acres, up to 2.5 gigawatts of capacity, built in partnership with Stack Infrastructure and BorderPlex Digital Assets. The total price tag: $165 billion. That's not a typo.

Construction started in September 2025. The plan was simple: power it with 2.45 gigawatts of Bloom Energy fuel cells running on natural gas delivered by an 18-mile pipeline across state land. Oracle hired Energy Transfer — the Texas pipeline giant — to build the "Green Chile" lateral.

Then New Mexico Land Commissioner Stephanie Garcia Richard rejected the pipeline application. Then she rejected it again on July 14. Both times, the state said no to granting rights-of-way across state land for the natural gas line. The environmental and water concerns were too severe for a state already projecting 25% less water availability within 50 years.

Oracle's response was the same corporate boilerplate you'd expect: "Project Jupiter is being built for and by New Mexicans and will dramatically change the economic trajectory across the state." The project remains "on schedule," they say. But here's the thing — you can't power a 2.5 GW data center with good intentions. If the pipeline doesn't get built, the Bloom Energy fuel cells don't run. And if the fuel cells don't run, the $165 billion campus is a ghost town.

The S&P Downgrade That Changes Everything

On July 9, S&P Global Ratings cut Oracle's long-term issuer credit rating from BBB to BBB- — the lowest investment-grade tier, one notch above junk. The reason? Oracle is spending $250 billion on AI data center expansion and burning through cash faster than it can generate it.

Let me put the numbers in perspective. Oracle's fiscal year 2026 numbers:

Capex: $55.7 billion, up 162% year-over-year from $21.2 billion.
Free cash flow: Negative $23.7 billion — negative. They burned $23.7 billion more than they made.
Total liabilities: $218.7 billion.
Stock price: Down 64% from the September 2025 peak.

And management has already telegraphed another $40 billion in debt and equity financing in fiscal 2027, with another $70 billion in capex planned. The five-year credit default swaps on Oracle's debt hit 2.03 percentage points on Monday — the highest in nearly 18 years.

S&P's outlook is "stable," which in credit-rating language means "we're watching this very carefully and another downgrade is absolutely on the table if anything goes wrong." And something is already going wrong.

The Secondary Bottleneck Nobody's Talking About — Junk Debt for Data Centers

Here's the part of this story that keeps me up at night as someone who runs actual infrastructure.

Oracle's debt crisis isn't happening in a vacuum. The market for data center debt is itself under stress. Junk-debt borrowing for data centers now represents 2.7% of the entire US high-yield market — up from virtually nothing two years ago. PIMCO just published a note saying the market is "diverging into two markets": prime projects that can still get financed, and everything else.

Sixty percent of the data centers scheduled to come online in 2027 haven't even started construction yet. And the ones that are financed are carrying debt at wildly different rates — 75% at spreads around 6%, but the rest at 10% or higher. That spread divergence is the market's way of screaming that lenders are getting selective.

Meta, Google, Microsoft, and Amazon can borrow at investment-grade rates because they have the balance sheets to support it. But the second-tier players — and I'd argue Oracle's deteriorating credit rating puts it squarely in that category now — are paying a premium that erodes the economics of every project.

This is the circular financing problem I wrote about when Apple's $140 billion cash pile entered the AI capex debate. The hyperscalers aren't building with profits. They're building with borrowed money. And when the cost of that borrowing goes up, the math on every data center changes.

What Oracle's Crisis Actually Tells Us About the AI Infrastructure Thesis

The bull case for Oracle is straightforward: $638 billion in contracted backlog, a massive $300 billion OpenAI deal starting in 2027, cloud infrastructure revenue up 93% year-over-year to $5.8 billion. Dan Ives is out there telling investors the stock is mispriced as speculative risk when it's sitting on decades of contracted revenue.

I'm not saying the bull case is wrong. I'm saying the timeline matters, and Oracle's timeline is getting squeezed from both sides.

On one side: the cash burn. Negative $23.7 billion free cash flow means Oracle needs capital markets to stay open and friendly. A single quarter of market turbulence — or another pipeline rejection, or a Wisconsin regulatory fight — and suddenly that $40 billion raise gets a lot more expensive. Oracle is already fighting a legal battle with Wisconsin regulators over another data center. The headlines are piling up.

On the other side: the physical constraints. The New Mexico pipeline rejection isn't an anomaly. It's part of a national pattern. We've seen 69+ data center moratoriums, 75+ blocked projects, $130 billion in construction halted in Q1 2026 alone. Land Commissioner Garcia Richard isn't a one-off — she's a trend. Every state-level regulator is watching what happens in New Mexico.

The AI infrastructure thesis assumes that capital can always solve physical constraints. But capital can't make a state legislature approve a pipeline. Capital can't make a community stop protesting a 1,400-acre data center in a drought zone. Capital can't make 18-year credit risk highs disappear.

What This Actually Means for Independent Hosting and Colocation Providers

First: Watch the debt markets, not just the headlines. Oracle's CDS spread hit 2.03% on Monday. That number matters more than any press release. When large hyperscalers start paying 10%+ on their construction debt, it raises the floor on everyone's cost of capital — including the regional colo providers you compete with and buy from.

Second: Power constraints are the real bottleneck, and they're getting worse. Oracle can't get a permit for a gas pipeline in New Mexico. Meta is building tents with off-grid gas turbines in Ohio. CoreWeave is betting $6 billion on Pennsylvania as the national revolt spreads across 42 states. The pattern is clear: power availability is the single biggest constraint on data center expansion, and the political resistance to building that power is hardening, not softening.

Third: Reposition independent hosting as the capital-efficient alternative. Oracle is spending $165 billion on one campus. That's more than the entire valuation of most independent hosting companies. When the hyperscalers are trying to rationalize $55.7 billion in annual capex to skeptical credit markets and furious regulators, the argument for colocation, managed hosting, and multi-tenant facilities gets stronger every quarter. You don't need a 2.5 GW campus to serve most businesses. You need reliable power, good connectivity, and a lease that doesn't require a junk-bond offering.

Fourth: Hedge your supply chain now. If Oracle's credit situation deteriorates further, it affects the entire data center supply chain — cooling equipment manufacturers, generator suppliers, switchgear fabricators. Oracle has $43 billion in new debt from just the last fiscal year. A default or restructuring would cascade through the infrastructure ecosystem faster than most people expect. Lock in equipment orders, extend supplier contracts, and build cash reserves while the window is still open.

The Structural Reality — Oracle Is a Symptom, Not the Disease

I don't want anyone to read this as me gloating about Oracle's troubles. What's happening to Oracle is happening to the entire hyperscale AI infrastructure buildout. It's just hitting Oracle first and hardest because they had the thinnest margin for error.

Oracle's balance sheet problems are the canary in the coal mine. When the market starts asking hard questions about $250 billion in AI data center spending — questions like "is the revenue actually there?" and "what happens when a pipeline gets blocked?" and "can you really finance $55.7 billion in capex every year?" — the answers matter for every company in this space. Microsoft, Google, Amazon, Meta — they all have stronger balance sheets than Oracle. But they're all making the same bet that infrastructure spending today will be paid for by AI revenue tomorrow.

The difference is timing. Oracle doesn't have the luxury of a 10-year runway. Their cash is running out faster than their data centers are coming online. And when a $165 billion campus can't get a permit for the pipeline it needs to run, the runway gets shorter, not longer.

The Bottom Line

Oracle is one notch above junk, sitting on a $218 billion debt pile, with a negative cash flow so deep you could stack data centers in it. And the AI revenue that's supposed to save them? It's still 12 to 24 months away, assuming nothing else goes wrong — which, in this market, is a very dangerous assumption.

This is what happens when you let the balance sheet tail wag the infrastructure dog. Oracle convinced itself and Wall Street that $250 billion in borrowing was fine because the backlog was real. And the backlog IS real. But backlogs don't pay the interest on $43 billion in new debt. Cash does. And Oracle's cash is buried under a pipeline rejection in the New Mexico desert.

If I were running an independent hosting business right now, I'd be watching Oracle's CDS spread like a hawk — not because I care about Oracle's stock price, but because every data point in their collapse is a data point about the structural fragility of the AI infrastructure thesis we've all been living under for two years.

Build for the world where the hyperscalers can't borrow at 6% forever. Because that world is already here.

-- Allan Ali, Founder