Story
October 8, 2026
SpaceX’s $40 Billion Nvidia Bet Puts AI Debt to the Test
SpaceX’s backers see scarce Nvidia computing power as an asset that can be rented out at a profit, while wary credit investors see another test of whether the AI buildout is outrunning its financing base.
SpaceX’s proposed $40 billion GPU purchase is the latest escalation in an AI infrastructure race increasingly financed with debt. The company is seeking roughly $30 billion of investment-grade borrowing and $10 billion in bank loans, a structure that would put its BBB rating—and investor appetite for the paper—under a brighter spotlight.1
The borrowing push follows SpaceX’s record June IPO and a $25 billion bond sale less than two weeks later. That deal drew heavy fixed-income demand, but the market has changed since: AI-linked bonds have sold off and credit spreads have widened. Apollo and several banks are now discussing financing for the Nvidia purchase, potentially using the GPUs themselves as collateral.2
The commercial case rests on scarcity. SpaceX already rents computing capacity to customers including Google, Anthropic, Reflection AI and Cursor, which it later acquired. Musk said in a Sept. 24 X post that the company’s Colossus data center was running 230,000 Nvidia AI chips and Colossus 2 had 550,000; he said the total could reach 1.21 million by late December, “If we get lucky.”2
That expansion has made Nvidia bulls more confident. Jim Cramer argued that greater chip purchases translate directly into earnings for the buyer: “The more Nvidia [that SpaceX] buys, the more money it makes.” He framed the model as buying compute and lending it onward, a bet he says can be highly profitable.3
But the financing is not a cost-free vote of confidence. SpaceX already reported nearly $40 billion in debt in its second-quarter financials, while its existing bonds trade near junk-rated levels. Investors are demanding higher yields for fresh AI exposure, suggesting they still believe in the compute boom—but are less willing to fund it blindly.2
The wager, then, is straightforward and immense: that Nvidia chips retain their value for years and that customers keep paying handsomely to use them. If that holds, SpaceX can turn leverage into an AI-rental engine. If it does not, the debt market will have funded one of the sector’s most expensive assumptions.