SpaceX Closes $40 Billion GPU Financing Deal with Apollo
The largest artificial intelligence chip financing deal ever recorded took on even more dramatic contours on Monday when SpaceX announced it had closed a $40 billion agreement with Apollo Global Management to acquire next-generation NVIDIA Vera Rubin chips for its terrestrial compute infrastructure. Structured as an operating lease spanning five to 5.22 years, the deal does nothing less than redraw the boundaries of what is possible to raise for AI hardware, while revealing an increasingly clear pattern: companies with genuine cash flow streams outside the AI ecosystem are positioned to devour assets that pure-play neoclouds without diversified revenue will barely manage to lease.
What makes this arrangement so unusual is its financial architecture. Apollo will not simply lend money to SpaceX; it will purchase the chips outright, hold them on its own balance sheet, and lease them back to the company. This is a true lease, not conventional debt, meaning the liability stays off SpaceX's books and each equipment tranche carries an independent lease term. If the aggregate principal payments total approximately $40 billion, additional financing charges -- interest, structuring fees, insurance costs -- will push the total cost well beyond that figure. The fixed interest rate will be determined at pricing, which has not yet occurred, with closing expected before the end of 2026.
A deal that separates those with revenue from those who only have GPUs
Elon Musk's response on X, posted on the Sunday preceding the formal announcement, was arguably the most politically charged aspect of the whole episode. Musk used the moment to take aim at more than 200 neocloud companies -- startups that exist exclusively to rent GPU capacity for training AI models. "That means the 200+ neoclouds are going to see guppy-level death and consolidation," he wrote, arguing that debt markets had largely shut the door on companies without substantial revenue streams outside of chip leasing.
The attack is not just personal -- it is structural. The financing crisis in the specialized AI cloud sector has been intensifying over recent months. Companies like CoreWeave, which reached valuations of $35 billion in 2024, are already facing pressure on GPU rental prices and competing for capital with the big tech firms. Apollo, by contrast, manages $2.5 trillion and has built its reputation on crises in mature sectors (energy, real estate, logistics). For it, leasing AI hardware to a company with investment-grade rating, Starlink generating billions annually, and rocket launches as a second revenue stream is exactly the kind of transaction that appears in annual reports as "financing the global industrial renaissance" -- the very term Apollo chose for its capital strategy.
The cost of scale in the AI race
SpaceX's own numbers tell part of the story. Capital expenditure surged to nearly $20 billion -- up from $6.9 billion a year earlier -- and the company expects to spend between $35 billion and $40 billion in the fourth quarter alone, according to its October 20 earnings call. To put that in perspective: this amount is equivalent to the annual capital expenditure of a telecom operator in a mid-sized emerging market. The difference is that SpaceX is not building cell towers; it is building data centers with NVIDIA's next-generation chips to power Grok, the AI model competing directly with OpenAI's GPT and Google's Gemini.
CFO Bret Johnsen does not hide the ambition. In the joint statement with Apollo, he described the deal as "the largest committed GPU financing" and emphasized that the company is seeing "less than one-year payback periods" on its AI infrastructure investments. If sustained at scale, this claim carries profound implications: it means SpaceX believes demand for compute will monetize so rapidly that the depreciation schedule of the chips (typically three to four years) will be more than offset by revenue from AI services running on them.
What comes next
The deal remains subject to closing conditions -- nothing is effectively concluded until the closing occurs. Bloomberg reported that, at the time of its initial reporting on October 25, pricing negotiations were still underway. Even so, the essential terms are already firm: Apollo buys the chips, SpaceX leases them, payments total $40 billion over five years, and delivery of the first batches begins in the first half of 2027.
What this $40 billion figure reveals above all else is how quickly the AI economy is moving from "ambitious project" to "national-scale infrastructure." The largest data center construction project in history requires no government bond -- it requires a private company with cash flow, creditworthiness, and a vision of artificial intelligence justifying the outlay. If the bet holds, the question to watch is not whether others will follow this path, but whether the global debt market is prepared for the next $40 billion cycle.
Sources: CNBC, TechCrunch, Apollo Global Management
✓ Independent sources cross-checked and verified before publishing
