Apollo Capital Management is leading discussions with a group of banks to structure a $40 billion debt package that would fund SpaceX’s purchase of Nvidia graphics processors, according to a person familiar with the talks. The financing would lean almost entirely on the investment-grade bond market, the same window SpaceX tapped four months ago for a $25 billion multi-tranche sale that drew what CNBC described as record demand. Since that deal priced, AI-linked bonds have sold off and credit spreads have widened, meaning the new issuance will test whether fixed-income buyers still treat GPU collateral as money-good at scale.
GPUs as collateral
The proposed structure would pledge the chips themselves as security, a bet that the hardware retains value across a seven-year useful life, the industry’s working assumption, according to a credit-market source. Computing capacity remains scarce and the shortage shows no sign of easing, which underpins that valuation. But the source also noted that investors are demanding higher yields to service AI debt, a signal that selectivity is rising even as issuance pipelines swell.
A familiar issuance pattern
SpaceX would follow Meta, Amazon and Google into the market with a bond deal sized to fund AI infrastructure. The company’s June bond sale came two weeks after its IPO, a sequence that suggested the public listing was partly a ratings catalyst. Now the same investment-grade window is being asked to absorb another $40 billion backed by a single asset class that did not exist as collateral in any meaningful size two years ago.
The yield test
Both sources said more debt sales are expected across the sector. The question is not whether the market can clear the paper, the June deal proved it can, but at what spread. If GPU-backed paper prices wider than unsecured tech issuance, the collateral argument weakens. If it prices tighter, the asset class gains a new benchmark. Either way, Apollo’s role as arranger puts a private-credit firm at the center of a public-market transaction that could define the cost of capital for the next wave of AI buildout.