---
title: "SpaceX’s GPU Bondholders Would Like Their Collateral to Stay on the Ground"
description: "The chip-backed debt is rated above SpaceX itself, yields more than SpaceX’s own bonds, and its lenders reportedly want the chips kept on Earth"
author: "Nate Ledger"
published: 2026-11-20T14:15:21Z
modified: 2026-11-21T00:19:16Z
url: https://rews.cc/a/spacex-s-gpu-bondholders-would-like-their-collateral-to-stay-9de3b9
language: en
tags: ["space", "debt", "elon-musk", "ai", "bonds", "business"]
publisher: "Rews (https://rews.cc)"
---

# SpaceX’s GPU Bondholders Would Like Their Collateral to Stay on the Ground

*The chip-backed debt is rated above SpaceX itself, yields more than SpaceX’s own bonds, and its lenders reportedly want the chips kept on Earth*

By Nate Ledger · November 20, 2026 · https://rews.cc/a/spacex-s-gpu-bondholders-would-like-their-collateral-to-stay-9de3b9

## In brief

- Apollo and Goldman Sachs led roughly $44 billion of debt to fund SpaceX’s purchase of Nvidia chips, Reuters reported
- The new bonds priced 3.45 points over Treasuries, about 8.3% yield, versus about 7.2% on SpaceX’s 2031 bonds
- Moody’s rated the debt A3, one notch above SpaceX, citing hell-or-high-water lease clauses; KBRA rated it A-minus
- Lenders want the chips kept on Earth; Moody’s says they will sit in US data centers majority-owned by SpaceX
- The bonds mature in 2032, but the principal will be repaid in less than four years to reduce credit risk

Here is a sentence you don’t often see in a credit agreement: please do not fire the collateral into space. As [Michelle Chan reported for the *Financial Times*](https://arstechnica.com/space/2026/11/spacex-lenders-dont-want-elon-musk-launching-their-collateral-into-space/), in a story republished by Ars Technica, the lenders behind SpaceX’s enormous Nvidia chip financing want the chips that secure their debt to stay on Earth. That is a reasonable thing for a lender to want, and also a very funny one, and the reason it’s both is that the borrower is a rocket company run by Elon Musk, who would like to put data centers in orbit.

Start with the deal. SpaceX [announced a financing commitment](https://rews.cc/a/musk-says-spacex-got-40-billion-of-gpu-debt-that-lenders-now-084384) of about $40 billion in October to buy Nvidia graphics chips, and Apollo and Goldman Sachs ended up leading roughly [$44 billion of GPU debt](https://www.tradingview.com/news/reuters.com,2026:newsml_L3N4711LM:0-apollo-goldman-lead-44-billion-spacex-gpu-debt-financing/), according to Reuters. Before Thursday, more than $19 billion of the bonds already had commitments, $4 billion of that from Pimco. SpaceX says funds managed and advised by BlackRock, Goldman Sachs Asset Management and KKR also put in anchor orders. Apollo, Goldman and Pimco declined to comment, and SpaceX didn’t respond to the FT.

The newest bonds priced at 3.45 percentage points over US Treasuries, which works out to a total yield of about 8.3 percent, according to people familiar with the deal. Subtract and you get an implied Treasury yield somewhere around 4.85 percent. Then compare: SpaceX’s regular corporate bonds maturing in 2031 traded at roughly 7.2 percent on Thursday. So the chip debt pays about 1.1 percentage points more than SpaceX’s own debt.

![A rocket pulls a rope tied around a computer chip and an anchor against a blue background.](https://rews.cc/img/faada284040bd4969dfe64711a135e060cfab13f.webp)

Here’s the odd part. Moody’s rated the chip debt A3, which is one notch *above* SpaceX’s own credit rating. KBRA gave it A-minus. In other words this debt is supposed to be safer than lending to SpaceX directly, and it pays more than lending to SpaceX directly. I can think of a few reasons for that gap: it’s a huge and new sort of deal, the collateral is a pile of chips that lose value fast, and buyers who are being asked to absorb tens of billions of dollars of something unfamiliar tend to want paying for the trouble. Some of that spread is the price of novelty. Novelty is a cost, and if you can be the one who gets paid for it, nice.

## Why it’s rated above SpaceX

The structure works like this. The bonds are paid out of SpaceX’s lease payments on the chips, and the lease includes what the rating agencies call “hell-or-high-water” clauses, which stop SpaceX from walking away. Translated: SpaceX pays the rent whether the chips work or don’t, whether the AI boom continues or doesn’t, whether it still wants the chips or not. So a bondholder’s first protection is SpaceX’s promise to pay, about as strong as a direct SpaceX bond. The second protection is the chips. If SpaceX stops paying, the lenders can in theory take the GPUs and sell them to someone else. A claim on SpaceX plus a claim on a mountain of Nvidia hardware beats a claim on SpaceX alone, and that’s how you get a rating a notch higher than the company’s.

The second protection only counts if you can actually get hold of the chips.

![A hand reaches through a circular chamber toward a microchip, illustrated in blue, yellow, and cream.](https://rews.cc/img/8e1abd89674ecbe758e529e89c34a7f25eeae564.webp)

Picture the repo man’s problem. (A hypothetical, but an obvious one.) A borrower defaults and you’re owed tens of billions of dollars backed by GPUs sitting in a data center in Texas. You send people, you get a court order, you take the racks, you sell them to some hyperscaler. Now picture the same GPUs in low Earth orbit. You do not have a rocket. The only company with enough rockets to fetch them is the one that just defaulted on you. This is a security interest in name only.

Hence the lenders’ demand, as [Benzinga summarised it](https://www.benzinga.com/markets/tech/26/11/63464192/elon-musks-dream-meets-wall-street-reality-spacex-lenders-dont-want-nvidia-chips-leaving-earth): no Nvidia chips leaving Earth. It also shows up in the Moody’s rating report, which says SpaceX plans to put the chips in US data centers that it majority-owns, controls and operates. Each of those words does work. “US” means a US court can reach them. “Majority-owned, controlled and operated” means they aren’t hidden away in somebody else’s facility. And nowhere does it say “in orbit.”

## The other trick

There’s one more piece of the structure that tells you what lenders actually think of GPUs. The bonds mature in 2032, but the principal is paid down in less than four years, an arrangement meant to reduce credit risk. That’s a frank acknowledgement of how quickly AI chips go out of date. A chip that’s state of the art in 2027 may be a lot less exciting by 2031, so the lenders want their money back while the collateral still fetches a good price. You amortise the loan faster than the asset loses value, and so the collateral is always worth more than what you’re still owed, provided it stays where you can get at it.

![A melting computer chip burns like a candle beside an hourglass against a yellow background.](https://rews.cc/img/25543a8e3873d1b785d906f79a5727a54f9878a5.webp)

Plenty of investors seem to like this. “The SpaceX GPU deal appears cleverly structured and should appeal to a panoply of buyers, potentially including deep pools of insurance capital seeking attractive yield on investment-grade debt,” said Grant Nachman, chief investment officer at Shorecliff Asset Management. “This transaction could prove a harbinger of future issuances.”

He’s probably right. Insurers have a lot of money they’re required to keep in investment-grade debt, and they’d love that debt to pay more. An A-rated bond yielding about 8.3 percent is precisely what they’re looking for. The market has been [fretting about SpaceX’s chip debt](https://rews.cc/a/spacex-s-nvidia-chip-debt-pushes-its-default-insurance-to-a--b36249) for weeks, but the fretting appears to have been priced, and buyers came in anyway.

Still, the template has a limit built into it. It works because SpaceX is a strong credit and the chips are easy to reach and easy to sell. Musk’s plan for orbital data centers asks for the reverse: chips that are very hard to reach, owned by a company that has every reason to put them there. Financing those will take a different kind of lender, perhaps one with a rocket of its own. At the moment only one company fits that description, and it’s the borrower.
