SpaceX is reportedly preparing one of the biggest financing bets yet on the AI buildout: $40 billion in borrowing to acquire Nvidia GPUs, with Apollo Global Management and banks discussing a structure centered on roughly $30 billion of investment-grade bonds and $10 billion in loans.¹

The push follows SpaceX’s record $25 billion bond sale in mid-June, just weeks after its IPO. That deal drew strong demand, but the market backdrop has shifted since: AI-linked bonds have sold off and credit spreads have widened, making investors more demanding on yield and more cautious about exposure.²

For the bullish case, the logic is straightforward. SpaceX already rents compute to customers including Google, Anthropic, Reflection AI and Cursor, and more capacity could lift revenue quickly. The proposed financing may use the GPUs as collateral; lenders are still assuming high-end chips retain value for about seven years because computing power remains scarce.²

That is the argument championed by Jim Cramer, who says the purchase reinforces Nvidia’s appeal: “The more Nvidia [that SpaceX] buys, the more money it makes.”³ He points to the prospect that compute can be rented at returns rapid enough to rebut fears of runaway AI spending.

But the debt market is delivering the counterargument. SpaceX already reported nearly $40 billion in debt in its second-quarter financials, while its outstanding debt has traded near junk-bond levels. Bonds due in 2056 were recently priced around 85 cents on the dollar, yielding about 2.27 percentage points above Treasuries — a spread comparable to junk-rated debt despite SpaceX’s BBB rating.¹

The result is a stark test: if AI-compute revenue arrives as quickly as bulls expect, the GPUs may finance themselves. If it does not, SpaceX’s Nvidia wager could turn a growth story into a leverage reckoning.