Goldman Sachs and Apollo Global Management have started selling pieces of SpaceX’s roughly $40 billion financing for Nvidia’s newest AI chips to outside investors, turning the largest GPU financing on record into a live test of Wall Street’s appetite for AI hardware debt. The package backs about $47.3 billion of Nvidia Rubin equipment due in the first half of 2027, and SpaceX has agreed to keep paying rent even if the hardware fails or disaster strikes. The sale comes days after the midterm elections, as voters and lawmakers sharpen their scrutiny of the AI buildout.
The deal, in numbers
Goldman and Apollo are approaching investors before formal syndication begins, offering some buyers stakes of up to $500 million each, according to Bloomberg’s November 4 report. The package includes an $8.3 billion senior secured loan led by Goldman for bank syndication and several bond tranches, with Apollo a principal bond investor and Pacific Investment Management Co. also taking part.
Together with a $7 billion lease prepayment, the financing supports about $47.3 billion of Nvidia Rubin-generation equipment due for delivery in the first half of 2027. SpaceX called the arrangement the largest committed GPU financing ever, ahead of a $36 billion structured-notes deal for GPUs leased to Anthropic.
The scale is the point: a single tenant, a single chip generation, and a five-year credit product built on top of both. How fast Goldman and Apollo place the pieces will show how much appetite investors still have for financing AI hardware at this scale.
How the lease actually works
SpaceX does not buy the chips. Under a binding commitment letter signed on October 24 and announced two days later, an entity managed by Apollo acquires the GPU pods and related equipment and leases them to SpaceX for the data centers it selects.
The equipment arrives in tranches, each with its own lease schedule of about five to five and one-quarter years, with rent paid quarterly in advance. SpaceX expects principal payments across the schedules to total about $40 billion, and the lease is expected to close before the end of the year, subject to closing conditions.
Apollo’s October 26 release names the chips as Nvidia Vera Rubin parts for SpaceX’s compute buildout, and says they will help power Grok and SpaceX’s cloud services. Financing parties led by Apollo and Goldman Sachs, sole structuring agent, committed to the deal, with anchor commitments from funds managed by BlackRock and Goldman Sachs Asset Management and accounts advised by KKR.
What investors are being asked to swallow
Bloomberg, citing documents it reviewed, reported two terms that matter most to buyers. SpaceX must keep paying rent regardless of equipment condition, casualty or force majeure. And SpaceX gives no residual value guarantee, so investors carry the risk of what used AI chips are worth when the leases end.
Some things remain undisclosed. No company statement on the syndication has appeared, and the Bloomberg account rests on unnamed sources. The fixed lease rate, the number of buyers and the final tranche sizes have not been disclosed, and the $40 billion figure is the expected lease principal, not money already spent.
Our read: the bill for the AI boom is coming due in public
This deal makes the AI buildout’s financing model visible. Rent, don’t own. Keep the debt off the balance sheet. Let investors carry the hardware risk. Nvidia has backed a pattern of AI financing, guaranteeing billions in credit for Nscale’s Anthropic debt, and the industry keeps finding new ways to fund growth without putting the bill on its own books.
The timing is awkward for the sellers. The syndication begins as voters in 38 local governments across ten states backed limits on data center construction, and a Congress takes a harder look at the industry. If AI profits lag the spending, the clause that forces SpaceX to pay no matter what becomes the story, not the size of the deal.
Why It Matters
The sell-down is the first real market test of $40 billion-scale AI hardware credit. If Goldman and Apollo place it easily, the buildout’s debt machine keeps running and every lab’s financing stays cheap. If buyers balk, financing costs rise across the industry.
And the Rubin chips at stake are the next generation, so whoever controls them controls the next round of model training. AMD’s own billion-dollar chip financing shows the same debt engineering is now standard across the entire chip business today.
Background
The financing sits inside a wider wobble in AI infrastructure deals. Microsoft walked away from Nscale’s Texas data center deal this week, and the industry’s growth plans are being repriced by voters, regulators and lenders at the same time. SpaceX’s lease shows the money is still moving, but the terms are getting harder and the scrutiny sharper.
Source: Aju Press on Bloomberg’s report and Voltgrid’s syndication report
0 Comments