AI Boom to the Test Wall Street Wants Proof: AI Infrastructure Hits a Financing Wall

By AI World Journal

Soaring debt costs and record default-insurance prices show investors are no longer willing to fund the AI buildout without proof it will pay off.

Investors spent the past few years rewarding almost any company tied to artificial intelligence. That patience is wearing thin. The market now wants to know what the AI buildout really costs, who is paying for it, and how long it will take to earn the money back. Fears of higher interest rates have made those questions harder to ignore, especially for companies that borrow heavily to expand. Upcoming elections and policy uncertainty are adding to the strain.

A bruising week for AI infrastructure

The pressure showed up clearly this week, though it was concentrated in one corner of the market. CoreWeave was hit hardest, falling about 15% on Wednesday and another 12% on Friday, on track for a weekly drop of roughly 25%. Peers such as Nebius and IREN also slid sharply. Elon Musk added to the attention with a one-word “Yikes” in response to a post about CoreWeave’s bond yields.

Big tech held up far better. Cloud computing stocks fell about 4% on Friday, while the broader Nasdaq-100 dipped only modestly. Investors weren’t abandoning technology; they were pulling back from the companies most dependent on borrowed money.

The trigger was inflation. A hotter-than-expected inflation report revived bets that the Federal Reserve could raise interest rates, which would make it more expensive for debt-heavy data center and AI firms to finance their growth.

Inside CoreWeave’s “box”

CoreWeave has become a test case for the debt-funded side of the AI boom. The company rents computing power to customers such as Microsoft and has borrowed heavily to build it.

Earlier this year, CEO Michael Intrator explained the company’s model on a widely shared podcast appearance. He described each customer deal as a self-contained unit, which he called a “box.” The process works roughly like this:

• CoreWeave signs a long-term contract with a customer.
• It places that contract in a dedicated financing structure, alongside Nvidia chips and data center leases.
• The customer pays into the structure, not directly to CoreWeave.
• That cash first covers data center costs, power and debt payments. CoreWeave keeps what remains.

According to Intrator, a typical five-year deal pays off its costs in about two and a half years. Each box stands on its own, so trouble in one shouldn’t spread to the others. As lenders gain confidence in the model, he said, they have offered money at steadily lower rates.

On paper, the design is sensible. It ties each loan to a specific revenue stream that a lender can evaluate. But it doesn’t remove risk. Construction delays, rising borrowing costs, shifts in customer demand or trouble refinancing can all squeeze the math.

What the credit market is signaling

Many analysts argue that the most telling signal comes from the credit market, not the stock market. Credit default swaps work like insurance for bondholders, and their price shows how much risk lenders see.

CoreWeave’s five-year swap spread hit a record of nearly 1,195 basis points intraday on Thursday, October 15, before settling near 1,040. That is more than two and a half times its level a year ago. A reading above 1,000 is generally considered distressed territory. Standard pricing models translate it into roughly a 20% chance of default in any given year, or about 60% over five years. Meanwhile, yields on some of the company’s bonds briefly rose above 15%.

That is a market price, not a forecast, and it can move quickly in either direction. CoreWeave has not missed a payment. Still, the numbers show that lenders, often the most cautious money in the room, are demanding far more protection than they did a few months ago.

Why borrowing got more expensive

Part of the problem has nothing to do with CoreWeave specifically. Long-term U.S. Treasury yields have climbed above 5% as bond investors demand more compensation for inflation and other long-run risks. When the “risk-free” rate goes up, every riskier borrower has to pay more as well. Yields on high-yield corporate debt have risen in step, and CoreWeave’s own borrowing costs have climbed faster than the junk-bond average.

The timing is awkward. CoreWeave carried about $35.6 billion in debt as of June 30, plus roughly $16 billion in lease obligations, for a total of nearly $52 billion. Its unsecured bonds carry interest rates of 8.5% to 9.75%. Interest expense consumed about 22 cents of every revenue dollar last quarter. And the company expects to spend $35 billion to $39 billion on new capacity this year, much of which will need fresh financing.

Its revenue, meanwhile, is mostly fixed. The long-term contracts that reassure lenders also prevent the company from raising prices when its costs go up. So as borrowing costs climb, the cushion inside each box shrinks.

Oracle: a giant betting big on AI capacity

Oracle shows that the funding strain isn’t limited to younger, debt-heavy upstarts. The company has spent decades as a profitable software business. Now it is reinventing itself as a major supplier of AI computing, and that shift has changed its financial profile.

Oracle has signed very large, multiyear agreements to provide cloud capacity to AI developers, including OpenAI. Its backlog of contracted future revenue now stands at about $664 billion. But fulfilling those contracts means building data centers at a pace Oracle has never tried before.

The spending is enormous. Capital expenditures reached $28.5 billion in the quarter that ended in August, up from $8.5 billion a year earlier. Free cash flow came in at negative $5.4 billion. For the full fiscal year, Oracle expects to spend $90 billion to $95 billion and has not said when free cash flow will turn positive.

To close the gap, Oracle is tapping both debt and stock markets. It has said it expects to raise $45 billion to $50 billion in 2026, and it sold about $20 billion in new shares last quarter alone. Its debt now totals roughly $125 billion. Selling shares dilutes existing shareholders, and adding debt raises interest costs at a time when borrowing is already getting more expensive.

Credit markets have noticed. The cost of insuring Oracle’s debt hit a record 288 basis points on October 15. That is far below CoreWeave’s level, but it reflects growing concern about how much Oracle is borrowing and how much of its future depends on a small number of large AI customers.

That concentration is the core risk. If a major customer slows its spending, changes plans or struggles to raise its own money, Oracle could be left with costly capacity that takes longer to fill. The company’s long record of steady profits gives it more room than CoreWeave, but the strategy still depends on AI demand holding up for years.

For both Oracle and CoreWeave, success comes down to turning new capacity into steady, paying usage. Data centers must be financed, built, powered and filled with customers. Each delay adds cost while revenue waits.

Politics and policy add another layer of uncertainty

The financing squeeze is happening as the political calendar heats up. The U.S. midterm elections on November 3 will decide control of Congress, and markets often grow more cautious in the weeks before a vote that could shift policy direction.

Several issues on the ballot matter directly to AI infrastructure:

• Federal spending and deficits. Concerns about government borrowing are one reason long-term Treasury yields have stayed high. Any change in how Congress handles taxes and spending could move rates, and borrowing costs for companies like CoreWeave and Oracle would move with them.
• Energy and permitting. AI data centers need huge amounts of electricity. Rules on power plant construction, grid connections and permitting timelines could speed up or slow down how fast new capacity comes online.
• Trade and chip policy. Tariffs on equipment and export controls on advanced chips affect both what it costs to build data centers and how many customers chipmakers and cloud providers can serve.
• AI regulation. Lawmakers in both parties have debated how closely to oversee AI. New rules could change what customers build and how much computing they buy.

The election outcome itself may matter less than the uncertainty leading up to it. Investors tend to pull back from riskier, debt-funded bets when policy direction is unclear. For a sector that depends on cheap, plentiful financing, even a few weeks of hesitation can show up in credit spreads and stock prices.

The Federal Reserve remains the biggest wildcard. After this week’s inflation data, traders are weighing the possibility of rate hikes rather than cuts. Higher rates would make it more expensive to refinance existing debt and fund new projects, putting the most pressure on companies whose contracts lock in revenue while their borrowing costs keep rising.

Spending is outrunning proven returns

The bigger picture is daunting. One estimate cited by Reuters suggests U.S. AI investment could reach $9 trillion between 2025 and 2032. To earn a 10% return on that spending, the sector would need roughly $3.55 trillion in yearly revenue by 2032. These are projections, but they explain why investors want proof that the spending will pay off.

The real question isn’t whether companies want AI. It is whether AI revenue can cover the full cost of chips, power, facilities and financing.

Adoption is broad, but payoffs are uneven

Businesses are certainly using AI. In McKinsey’s 2026 survey, nearly nine in ten respondents said their organizations use it regularly in at least one function. But only 37% saw an effect on earnings, and about one in five said operating costs were holding back wider use. Sixty percent still expect their companies to spend more on AI next year.

That gap may push customers toward projects with clear, measurable benefits. It may also make them more careful about locking into large, long-term computing contracts, which are exactly the deals that CoreWeave’s box model and Oracle’s backlog depend on.

A more selective market

The AI trade isn’t over, but it is changing. Demand for computing power remains strong, and both CoreWeave and Oracle have large backlogs of signed contracts. What has changed is the price of the money needed to deliver on them.

The early phase of the boom rewarded speed and ambition. The next may reward companies that borrow carefully, finish projects on time and show that customers will keep paying.

CoreWeave’s distressed credit signals, Oracle’s record spending and fundraising, an uncertain political backdrop and the threat of higher rates all point to the same test. Can the industry turn its enormous investments into reliable profits before financing costs catch up?

Original article written for AI World Journal. Market and company facts are current as of October 17, 2026.