Apollo will own the GPUs

SpaceX has secured a binding commitment for about $40 billion of financing for AI chips and related data-center equipment, using a structure that says as much about capital markets as it does about compute demand.

According to a regulatory filing, an Apollo Global Management-managed entity will buy the GPUs and associated server hardware, then lease that equipment to SpaceX for roughly five to five-and-a-quarter years from delivery of each tranche. SpaceX said it expects the deal to close by the end of 2026, subject to closing conditions, with leases beginning in the first half of 2027 as equipment arrives.

That is not the same thing as Apollo handing SpaceX $40 billion in cash and walking away. Apollo is financing specific assets, then collecting lease payments against them. For SpaceX, the appeal is plain enough: more compute without asking traditional debt investors to swallow all of the capex at once.

The company called it the largest committed GPU financing. Large boasts are cheap; committed financing is not. What matters here is that a private-capital giant was willing to underwrite the hardware in size, and to do it against a business that can point to revenue beyond selling AI capacity to other AI companies.

SpaceX’s own numbers help explain the lender appetite. Its AI segment, which includes Grok, Cursor and cloud-compute operations, generated $7.6 billion of the company’s $14.6 billion in third-quarter revenue, according to the filing and related company statements. Capital spending, meanwhile, rose to nearly $20 billion from $6.9 billion a year earlier, and CFO Bret Johnsen said on the Oct. 20 earnings call that fourth-quarter capex was expected to reach $35 billion to $40 billion.

That denominator matters. The financing is enormous, but so is the spending it is meant to support. This is less a victory lap over thrift than a sign that the AI buildout has become too large to fund neatly with one instrument or one investor class.

Apollo said the deal would support delivery of Nvidia’s Vera Rubin chips for SpaceX’s terrestrial compute buildout. Johnsen said SpaceX is seeing payback periods of less than one year on its AI infrastructure investments. Useful, if true, though “payback” can be a very flattering metric when depreciation and competitive decay are left offstage.

Still, the transaction draws a line through a market that has grown allergic to AI capex without cash flow. SpaceX did not prove every GPU bet is wise. It proved that one borrower, with multiple businesses and steady revenue, can still get size.