Oct 27, 2026 

Article

Airlines rarely own every plane they fly. They lease them from financiers who own the metal and collect rent. On October 26, SpaceX disclosed that it's doing something similar with AI chips, at a scale no one has done before.

What happened

In a regulatory filing, SpaceX said it entered a binding commitment on October 24 for a master finance lease worth about $40 billion, covering GPUs and related data-center equipment. An entity managed by Apollo Global Management will buy the GPU pods and lease them to SpaceX for data centers SpaceX chooses. Apollo says the hardware is Nvidia's next-generation Vera Rubin chips for SpaceX's terrestrial compute buildout. SpaceX calls it the largest committed GPU financing yet.

Apollo and Goldman Sachs led the financing, with Goldman as sole structuring agent. Funds managed by BlackRock, Goldman Sachs Asset Management and KKR made anchor commitments.

How the deal works

• Each equipment tranche runs about five to 5.25 years.
• Aggregate principal payments are expected to total about $40 billion, plus financing charges. The fixed rate will be set before SpaceX signs the lease itself.
• Closing is expected before the end of 2026, subject to conditions. Lease payments start when the GPUs arrive, expected in the first half of 2027.

Closed or not?

This detail is worth getting right. Elon Musk posted on X that SpaceX had closed the $40 billion financing. The SEC filing describes a binding commitment that's still subject to closing conditions. For readers, the filing is the authority: the money is committed, not yet deployed.

Why SpaceX wants it

The capital needs are enormous. SpaceX's capital spending hit nearly $20 billion last quarter, up from $6.9 billion a year earlier, and its CFO expects $35 billion to $40 billion in the fourth quarter alone. Leasing spreads that cost across the working life of the chips instead of landing on the balance sheet at once.

Musk argued that SpaceX's investment-grade status and its Starlink and launch revenue let it raise money that pure-play compute rivals can't. He also warned that more than 200 "neocloud" GPU providers could face failures and consolidation as debt financing tightens.

The customers for this compute aren't only in-house. Anthropic's S-1, filed the same day, says Anthropic will pay SpaceX about $3.1 billion a month for Nvidia-powered compute once all promised capacity is delivered, according to CNBC.

The risk nobody can price yet

Planes fly for decades. GPU generations turn over far faster. A five-year lease means paying for Vera Rubin chips long after newer ones are on the market. If demand for AI compute cools, or if efficiency gains let models run on less hardware, the lessor owns depreciating equipment and the lessee owns a fixed bill. That's the bet Apollo, Goldman, BlackRock and KKR just made, which makes this deal a useful signal for the whole AI infrastructure trade.

What it means for builders

1. Compute is becoming a financial product. Expect more ways to rent, lease and pre-buy capacity, and more price competition among providers fighting to keep their chips busy.
2. Check your provider's balance sheet. If you rely on a smaller GPU cloud, know how it's financed and keep a fallback ready.
3. Stay asset-light until usage is predictable. Rent compute by the job, and only sign long commitments once your demand is steady enough to forecast.

This article is information, not investment advice.