The banks are reportedly asking some investors to take stakes of up to $500 million before formal syndication even begins

Wed 4 Nov 2026 · 21:44 EST · 3 min read

In AI

Share

Goldman Sachs and Apollo Global Management are out with a sales pitch. The product is roughly $40 billion of financing tied to AI chips for SpaceX.

Bloomberg News reported on Wednesday that the two firms have started approaching potential buyers for the multi-tranche package. Some investors are being asked to commit up to $500 million each, and the formal syndication process hasn’t even started yet.

How the deal is built

Think of it less like a loan and more like a very expensive car lease. Instead of SpaceX buying the hardware outright, an Apollo-managed entity would purchase the GPUs and server equipment and lease them to SpaceX.

That structure is designed to keep the obligations off SpaceX’s balance sheet. The company gets the compute without its books showing a $40 billion debt pile.

The financing is split into multiple tranches. Each slice carries a different level of risk and return, so different types of investors can pick the layer that suits them.

Bloomberg, citing people familiar with the matter, reported that the early outreach involves stakes of up to $500 million per investor. Asking for checks that size before syndication suggests the arrangers want anchor commitments locked in early.

Total principal payments under the leases come to approximately $40 billion. SpaceX expects leasing to begin in the first half of 2027, once the hardware is delivered.

A record-sized bet on compute

The lease arrangement was first disclosed on October 26, 2026, through regulatory filings and follow-up public statements. It is described as the largest GPU financing commitment on record.

AI, tech, and the markets they move—in one daily briefing.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

We respect your privacy. Unsubscribe anytime.

The purpose is to expand SpaceX’s terrestrial AI compute. That capacity is meant to support work including Grok models and related cloud services.

The scale lines up with what SpaceX has already been spending. The company had previously guided capital expenditures of $35-40 billion for the fourth quarter of 2026 alone.

Elon Musk also weighed in on X in November 2026 on the question of investment-grade status. Credit quality matters here because it shapes which institutional buyers can legally hold the paper.

The deal also broadens where SpaceX gets its money. Leaning on lease structures rather than conventional corporate bonds diversifies the company’s funding sources.

Why the timing is awkward

The pitch lands at a delicate moment for AI-linked debt. Markets for that paper have been volatile, and investors are demanding higher returns.

The reason is simple. Concern has been building over the enormous sums companies are pouring into compute, and whether that spending will pay off.

Political uncertainty adds another layer. A recent Democratic electoral victory has introduced fresh questions for the sector, and the financing is being marketed right in the middle of that.

What this means

For SpaceX, the structure is the point. Moving roughly $40 billion of equipment obligations off the balance sheet lets the company scale AI infrastructure while keeping its headline leverage cleaner.

For Apollo, the deal fits a broader private credit strategy: owning hard assets and collecting lease income from a high-profile tenant. The GPUs themselves serve as collateral, though their value depends on how quickly newer chips make them obsolete.

The early asks of up to $500 million per investor also hint at the buyer base. Only large institutions can write checks that size, which means this deal is being built for pension funds, insurers, and big asset managers rather than a wide pool of smaller lenders.

Then there is the delivery schedule. Leasing is expected to start in the first half of 2027, so any hardware delays would push back the cash flows that investors are signing up for.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.