Building AI data centers is turning into one of the largest debt exercises in corporate history, and this week the pipeline got visibly fuller.

The Wall Street Journal reported that Broadcom has been seeking more than $50 billion to fund the custom AI chip it is developing with OpenAI, in a package that could include "several gigawatts of OpenAI chip capacity" and is slated to close before the end of the year. Broadcom had already assembled a roughly $60 billion debt package to finance Anthropic's own custom-chip build-out.

Oracle, meanwhile, is reported to be in talks with Apollo and Goldman Sachs, with the financing meant to bridge the gap between when it pays for hardware and when cloud revenue arrives. And SpaceX is seeking about $40 billion to buy Nvidia hardware — roughly $10 billion in bank loans and $30 billion in investment-grade debt, according to the Financial Times, with Apollo expected to lead the deal and the transaction due to close in 2027.

The pattern is the same across all three: instead of paying cash for accelerators, the buyers borrow against future compute revenue, using the chips themselves as the anchor for very large, largely non-public credit deals. Analysts and investors have started asking how much of the AI build-out is now financed with debt and what happens if the revenue curve bends — a question sharpened by reports that OpenAI's revenue run rate is closing in on $50 billion a year but sits below some earlier figures.

Corporate bond and private-credit markets have absorbed the issuance so far, but the scale is concentrating AI exposure in lenders as well as in a handful of technology companies. SpaceX, for its part, is also pursuing its own $16.8 billion TeraFab project in Grimes County, Texas, and has confirmed "just discussions" with TSMC about possible involvement.

Sources