SpaceX’s latest AI expansion plan is not a conventional capital raise. It is a test of whether debt investors still believe the hunger for computing power can justify ever-larger borrowing.

Mid-June: Less than two weeks after its record IPO, SpaceX raised $25 billion in bonds across several maturities, drawing strong demand from fixed-income investors. The company was already building an AI-compute business alongside its rocket and Starlink operations, with deals involving Google, Anthropic, Reflection AI and Cursor, which it later acquired.¹

September 24: Elon Musk said SpaceX’s Colossus data center in Memphis was operating 230,000 Nvidia AI chips, while Colossus 2 had 550,000. He projected a combined 1.21 million chips by late December, qualifying the target with: “If we get lucky.”¹

Now: Apollo Global Management and several banks are discussing a roughly $40 billion financing package for more Nvidia GPUs, according to people familiar with the talks. The structure could include about $30 billion in investment-grade debt and $10 billion in bank loans, with the GPUs themselves likely serving as collateral.¹

The case for the deal rests on scarcity: credit-market participants expect high-performing GPUs to retain value for roughly seven years, while compute shortages persist. But the timing is uneasy. AI-related bonds have sold off, credit spreads have widened, and investors are demanding higher yields. SpaceX already reported nearly $40 billion in debt, while its long-dated bonds have traded around 85 cents on the dollar, at yields compared with junk-rated debt.²

Bulls argue the spending cycle is rational rather than reckless. Jim Cramer framed the thesis bluntly: “The more Nvidia [that SpaceX] buys, the more money it makes.” He said Musk was following the model of buying Nvidia compute and lending it out for profit.³

That is the wager facing SpaceX: turn GPUs into fast-growing rental revenue before a market growing more selective decides the AI buildout has borrowed too much.