Goldman and Apollo Shop Slices of SpaceX's $40B GPU Financing

Goldman Sachs and Apollo Global Management are in talks with investors to sell portions of a roughly $40 billion financing package for SpaceX's AI chips, Bloomberg News reports, with Reuters and The Globe and Mail carrying the same account. The package funds about $47.3 billion of GPU equipment, and SpaceX calls it the largest arrangement of its kind for GPUs on record.

Here's the precise version behind the "seeking buyers" shorthand. No completed sale to named buyers has been confirmed — what's on record is a detailed set of terms. An Apollo affiliate buys the hardware, SpaceX rents it for a little over five years, the equipment arrives in the first half of 2027, and the deal should close before the end of 2026.

What's actually being sold?

Goldman and Apollo aren't raising fresh money the way a startup sells shares. They already structured the financing and are now distributing slices of it — syndication, in credit-market terms. Goldman led the loan portion; Apollo was the key investor in the bonds. Pimco also took part.

Two tranche amounts are named:

• An $8.3 billion senior-secured loan, intended for syndication to banks.
• A $7 billion lease prepayment.

The roughly $40 billion headline refers to principal payments under the arrangement. The $47.3 billion figure refers to the equipment that money funds. They measure different things, and no report reconciles them line by line.

Deal documents identify the hardware as Nvidia's latest Rubin chips. A presentation Bloomberg reviewed describes a five-year rental agreement, consistent with the "a little over five years" language in the deal description.

How does the lease structure work?

An Apollo affiliate buys the chips and servers, then leases them to SpaceX. SpaceX pays rent, and those payments add up to roughly $40 billion in principal. Investors who buy pieces of the financing are buying a claim on that rent stream.

Ownership sits with the Apollo affiliate, not SpaceX — that's the core of the off-balance-sheet design. The final accounting treatment isn't spelled out in the reports, but the stated intent, confirmed by multiple outlets, is to keep the debt off SpaceX's balance sheet.

Timing matters for practitioners. Lenders commit capital in 2026, months before a single Rubin chip arrives in 2027, so they're underwriting delivery risk as well as SpaceX's ability to pay. The fixed rate — still undetermined when SpaceX described the deal — will be the first signal of how investors price that risk.

What does the reported $500 million ticket tell us?

Some investors were reportedly asked to buy stakes of up to $500 million before formal syndication, per people familiar with the matter. That's unconfirmed, and it's worth being careful with. A $500 million ceiling is far beyond retail or small-fund territory; it points to large institutions. But "up to" is a ceiling, not an average, and an ask isn't a commitment.

The split in the record is telling. Everything structural comes from deal documents or SpaceX's own description. Everything about demand comes from unnamed sources. Treat the structure as firm and the demand picture as provisional.

How does it compare with other AI financings?

The deal echoes patterns already visible in the AI buildout:

• TeraWulf's $19B Anthropic leases, where a Google credit backstop propped up financing for another party's hardware.
• Nvidia at $25B and Oracle at $45B to $50B, where the companies tapped capital markets directly.

SpaceX's deal stands out for combining very large size with a clean split between owner and user. In the TeraWulf case the support was a guarantee; in the Oracle and Nvidia cases the borrower was the company itself. Here, an asset-manager affiliate owns the machines and the tenant simply rents them — the same playbook real estate, aircraft and shipping used for decades.

What's at stake for the market?

For Nvidia, the deal is a demand signal with funding attached: a customer that arranged tens of billions for Rubin hardware has effectively pre-paid its place in the delivery queue. For lenders, the distribution is a test of appetite for GPU-backed leases to a marquee tenant — the rate it clears at becomes a benchmark.

For SpaceX, the structure will feature in every debate about its AI compute spend, including Morgan Stanley's $289B valuation forecast for the company.

Open questions remain: the fixed rate, the buyer list, the accounting treatment, delivery risk, technology risk over a five-year term on fast-turning chips, and concentration — a single tenant and a single hardware family carry the whole $40 billion. Syndication is where the market's real view of those risks becomes visible.