The queue for borrowed money got longer this week, and the names in it told the story. SpaceX, Broadcom and Oracle all moved within days of one another, each raising cash to feed the same machine: the construction of artificial-intelligence infrastructure.
SpaceX is arranging roughly $10 billion in loans alongside $30 billion in investment-grade bonds, according to people familiar with the matter. Broadcom is putting together financing measured in the tens of billions of dollars for OpenAI and Anthropic. Oracle is buying Nvidia chips through a structure held off its balance sheet. Add in the Broadcom-Anthropic transaction and the pending debt across the AI chain tops $150 billion.
The deals have crowded a market that is already near its limit. Public bond markets are saturated, so lenders are steering more of this paper into private credit and special-purpose vehicles, places where terms are looser and fewer eyes are on the details.
The strain shows up in the price of insurance against default. Credit-default swaps on Oracle reached 261 basis points, on SpaceX 197, and on Broadcom 136, each a record, according to market data. In the background, the 10-year Treasury yields about 5.31%, the 30-year about 5.70%, and the spread on CCC-rated corporate debt sits near 1,000 basis points.
Goldman Sachs has a name for what is unfolding: a war for capital. The bank’s framing points at the risk that has quietly built into the AI trade. The money is pouring into a single investment cycle, and it is beginning to crowd out Treasuries and other forms of financing that would otherwise find buyers.
The logic behind the borrowing is straightforward. The companies building AI need chips, data centers and power, and none of those are cheap. Equity markets funded the early years of the boom. As the sums have grown, the industry has turned to debt, lengthening its balance sheets in a way that would have looked reckless a few years ago.
The shift marks a real change in how AI is financed. For the first stretch of the boom, the money came mostly from venture rounds and share sales, and the losses sat with shareholders. Debt is different. It carries fixed payments and covenants, and it puts the lenders, not just the owners, on the hook for the outcome.
That shift has consequences for everyone else. When the biggest borrowers in the market are all spending on the same thing, the cost of capital for other businesses rises. A company that wants to build a factory or buy back stock now competes with the next AI data center for the same dollars.
The concentration is what worries analysts. A downturn in AI spending would hit not just the chipmakers and cloud providers but the lenders who financed them and the investors who bought the debt. The risk, spread across hundreds of billions of dollars, has become systemic in the narrow sense of the word: everyone is now tied to the same outcome.
There is no sign yet that the borrowing is slowing. Oracle, whose interest-bearing debt now exceeds $169 billion by some estimates, is said to be talking with Apollo and Goldman about a fresh round of financing for chip purchases. Other AI-infrastructure companies are pursuing deals in the same size range.
The lenders, too, are changing shape. Banks that once led these financings have stepped back from the largest deals, handing the business to private-credit funds and insurers hungry for yield. Those buyers can hold debt to maturity and care less about daily price moves, which makes them willing partners for borrowers who want to keep building before the economics are fully proven.
For the companies doing the borrowing, the calculus is simple enough. The value of a finished data center, if AI demand holds, dwarfs the cost of the debt used to build it. The gamble is on timing: whether the revenue arrives before the interest payments compound.
The bond market has seen waves of borrowing before, for oil, for telecoms, for housing. Each ended badly for someone. What distinguishes this one is the speed at which a single technology has come to dominate the queue for capital, and the small number of companies doing nearly all of the asking.
The spread between CCC-rated debt and Treasuries, near 1,000 basis points, is another warning. It says the market already prices a real chance of trouble among the weakest borrowers, even as the biggest names in AI keep raising money on increasingly generous terms. The two things cannot both be right forever.
For now, the lenders keep showing up. The yields are high, the borrowers are famous, and the demand story is, for the moment, intact. But a market in which a handful of companies are doing nearly all the borrowing is a market asking to be tested.
That test may not come on schedule. It may come when one of these borrowers revises its revenue, or when a lender pulls back, or when the cost of the 10-year Treasury makes the whole arithmetic stop working. The pile of debt, for now, keeps growing, and no one has agreed on where it stops.
