SpaceX locks in the largest GPU financing ever: a binding $40 billion deal with Apollo on a five-year buy-and-lease structure

The largest single chip-collateral financing has landed: in a Monday regulatory filing, SpaceX disclosed a binding commitment of roughly $40 billion with Apollo Global Management for AI chips and related datacenter equipment — what the company calls "the largest committed GPU financing." Structure: an Apollo-managed entity buys the chips and server hardware, then leases the equipment to SpaceX, with lease terms of about five to five-and-a-quarter years beginning as each tranche is delivered^[1][2].

Structure: a five-year buy-and-lease ledger

The deal pushes last week's "chip-collateral financing" paradigm to a new scale. Structurally it replicates the CoreWeave template: the financial institution (Apollo) holds the hardware asset's title, SpaceX gains use through leases, and rental payments match the chips' five-to-six-year depreciation cycle — hardware-devaluation risk locked into a predictable band, while SpaceX avoids recognizing tens of billions of equipment purchases on its balance sheet at once^[1]. Previously leaked details indicate the financing prices above SpaceX's last bond, with lower-rated tranches around 8.3% — the market doesn't price Musk-complex credit cheaply, but versus equity dilution, a cost SpaceX is evidently willing to pay^[2].

Use of proceeds: the chip arms race with Nvidia

The destination is clear: purchasing GPUs from Nvidia. Musk's earnings-call vision of SpaceX AI's compute expansion (next year's ARR potentially $1 trillion, exploring a Tesla merger) now has its first landed ammunition — $40 billion buys hundreds of thousands of top-tier GPUs, making SpaceX one of Nvidia's largest single customers^[2]. Note the timing coincidence: Anthropic's S-1, filed the same day, discloses compute contracts with SpaceX AI worth up to $120.5 billion — SpaceX is simultaneously the buyer of Nvidia's chips (financing $40 billion to purchase) and the seller of compute (leasing to Anthropic) — the company is turning itself into the AI infrastructure's central node: locking chip supply upstream and model-company compute demand downstream^[1][3].

Market backdrop: alternative financing under tight credit

The macro backdrop is the broad tightening of AI credit markets. Global IG corporate spreads just hit 18-month highs, Meta's CDS printed records and CoreWeave junk bonds fell to 80 cents — traditional bond markets' appetite for AI financing is shrinking. In that window, Apollo's private-credit model becomes the alternative channel: no claim on public-market supply capacity, flexible pricing (8.3% is normal for private credit), and hardware collateral reducing credit risk. This is also Apollo's next AI-infrastructure financing move after last month's discussed $50 billion framework with Nvidia — private credit is becoming the arms race's "shadow bank"^[1][2].

What to watch

First, the deal's staged execution pace — $40 billion commitments typically land in tranches. Second, whether Apollo securitizes such GPU-lease assets into the secondary market. Third, actual execution and compute delivery under the SpaceX AI-Anthropic $120.5 billion contract. Fourth, Musk's "SpaceX-Tesla merger" exploration — if it lands, this financing's balance-sheet home becomes a technical question^[1][2][3].

Historical coordinates: from $2.3 billion to $40 billion in three years

Placed in chip-collateral financing's chronicle, the leap is stark: in 2023, when Blackstone extended $2.3 billion of private credit to CoreWeave to pioneer the format, its executives were still asking "What is a GPU? Is AI real?"; in March 2026, CoreWeave completed an $8.5 billion financing backed by a Meta contract at about 6%; now a single SpaceX-Apollo deal jumps straight to $40 billion — the scale multiplied seventeen-fold in three years, and the asker of "What is a GPU?" has become the market-maker of "the largest GPU financing ever." The curve's slope is the most direct measure of the arms race's intensity: capital isn't tiptoeing in — it's charging. The only suspense is the curve's end: when financing scale exceeds what hardware-depreciation cycles can bear, does the curve bend naturally or stop dead.