Credit markets are starting to flash warning signs around the companies building out artificial intelligence infrastructure. On a recent episode of The 8:30, hosts Julie Hyman, Jake Conley, and Pras Subramanian examined how SpaceX, Oracle, and Broadcom are among the firms turning to debt financing to cover the enormous costs of chips, data centers, and computing capacity - and why the secondary market reception of that debt is drawing scrutiny.
Cracks Showing in the Secondary Market
The panel pointed to specific signals that bond investors are growing cautious. Credit default swap spreads on SpaceX debt have widened, meaning the cost of insuring against a default has gone up. Oracle's bond yields are climbing and its CDS levels have reportedly hit records. Bonds tied to Meta's Hyperion data center project have reportedly traded at record lows. None of this means these companies are in immediate distress - once financing is secured, the money is in hand regardless of how the bonds later trade. But as the hosts noted, weak secondary-market performance is a poor signal to send if a company intends to return to debt markets again soon, which is reportedly the case for SpaceX following its IPO.
Demand Isn't the Question - Duration Is
As Subramanian framed it, nobody is disputing that demand for this debt exists; buyers keep showing up. The more pressing question is how long that appetite can be sustained without some kind of correction or slowdown. Capital that might otherwise flow into other parts of the market is instead concentrating in the AI trade, a dynamic sometimes described as crowding out. The hosts suggested the real issue isn't whether investors will buy AI-linked debt today, but whether the underlying businesses will generate enough cash flow to service that debt when it comes due - and which companies might be left exposed if the build-out takes longer than expected to pay off.
The Push-Versus-Pull Debate Over AI Spending
Hyman revisited an argument she has made since the AI investment wave began: that much of the spending has felt driven by sellers insisting the technology is necessary, rather than by organic demand pulling it into the market. She cited commentary from Box CEO Aaron Levie, who has argued that the computing needs of this phase of AI - from personal agents to systems that monitor other AI systems for security risks - will be substantial. The panel raised a pointed counterpoint: much of the justification for scaling up compute involves building tools to manage risks created by other AI tools, a somewhat circular rationale. At the same time, the hosts acknowledged that companies actively using AI report genuine productivity gains, suggesting adoption will likely be real but uneven - arriving in a slower, "lumpier" pattern than the current financing pace implies.
Why This Matters Beyond Tech Balance Sheets
The stakes extend past any single company's bond offering. Heavy reliance on debt to fund physical infrastructure - chips, servers, power, facilities - ties the AI sector's fortunes more tightly to interest rates, credit conditions, and investor confidence than during earlier software-driven booms. If revenue from AI products lags behind the pace of borrowing, the companies most exposed are those that raised the most debt against future, not current, earnings. For a broader market watching AI as a primary growth driver, the health of this debt - not just equity valuations - has become a meaningful indicator of whether the build-out is sustainable or simply front-loaded.