News

SpaceX locks in a $40B GPU lease as Musk warns most AI neoclouds will fail

SpaceX has signed a binding commitment for an approximately $40 billion GPU lease with an Apollo managed entity, the largest committed GPU financing so far. At the same time Elon Musk says credit markets have closed for AI cloud companies without other revenue, and neocloud stocks dropped again on Monday.

SpaceX locks in a $40B GPU lease as Musk warns most AI neoclouds will fail

SpaceX has lined up what it calls the largest committed GPU financing ever: an approximately $40 billion master finance lease for Nvidia hardware, arranged with an entity managed by Apollo Global Management. The deal was disclosed in a Form 8-K filing and a press release on Monday, October 26, 2026. A day earlier, Elon Musk had already announced it on X, and he paired the news with a blunt warning for the rest of the AI infrastructure market: credit is drying up for anyone who does not have serious revenue outside of AI data centers, and most of the so called neoclouds will not survive.

Those two messages belong together. One company gets $40 billion on attractive terms because it has a satellite internet business and a rocket business behind it. Hundreds of others, whose only product is renting out GPUs, are watching their bond yields climb and their share prices fall. This article looks at what SpaceX actually signed, what Musk said, how the market reacted and why the financing of AI compute is turning into one of the most important stories of this cycle.

Diagram of the lease structure: lenders led by Apollo and Goldman Sachs, with anchor commitments from BlackRock, Goldman Sachs Asset Management and KKR advised accounts, fund an Apollo entity that buys Nvidia Vera Rubin GPU pods and leases them to SpaceX, which pays rent quarterly in advance over about five years per tranche

What SpaceX signed

According to the filing, SpaceX entered into a binding commitment letter on October 24 to sign a master finance lease with an Apollo managed entity. That entity will buy GPU pods and related equipment and lease them to SpaceX for use in data centers that SpaceX selects. The press release says the hardware is the next generation of Nvidia Vera Rubin chips, meant for what SpaceX calls its "terrestrial compute buildout".

The mechanics are more interesting than the headline number:

  • The equipment arrives in tranches, and each tranche gets its own lease schedule of about five to five and a quarter years from delivery.
  • Rent is paid quarterly in advance. Each payment is one twentieth of the financed cost of the tranche, plus a fixed interest rate on the remaining balance. Twenty quarterly payments add up to five years, which is how the principal is paid off.
  • The fixed rate has not been set yet. It will be determined at pricing, before the master lease is signed.
  • SpaceX expects total principal payments of about $40 billion. Closing is planned before the end of 2026, and the leases should start in the first half of 2027 when the GPUs are delivered.

The financing parties are led by Apollo and Goldman Sachs, with Goldman acting as sole structuring agent. Funds and accounts managed by BlackRock and Goldman Sachs Asset Management, as well as capital accounts advised by KKR, made anchor commitments. In other words, this is not one bank writing one big check. It is a structure that turns GPUs into a lease asset that large investment managers can hold in their portfolios.

There is one detail worth pointing out. On X, Musk described the deal as $40 billion of "primarily callable debt" that SpaceX had "closed". Reuters noted on Sunday that it was not clear what "closed" meant or whether pricing had been agreed. The filing answers that question: what exists today is a binding commitment, the lease itself still has closing conditions, and the rate is still open. That is normal for a deal of this size, but it is a good reminder that a social media post and a securities filing are not the same thing.

What Musk said about neoclouds

Musk's comments came in reply to a post by investor Chamath Palihapitiya, who argued that there will be much more "partially stranded compute" in 2027 than currently forecast: hardware that is bought and paid for but earns less than its cost of capital. Musk agreed and went further. "As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn't have substantial revenue independent of AI data centers," he wrote, according to CNBC. He said the "200+ neoclouds are going to see guppy-level death and consolidation" and that some would make it through, "but not many". Later he added that SpaceX would not buy any of them, calling that "pointless".

Neoclouds are the specialized cloud providers that grew up around the AI boom, such as CoreWeave, Nebius and IREN. Unlike Amazon, Microsoft or Google, they do not have a large business outside of AI to fall back on. They borrow money, buy Nvidia chips, build or rent data centers and then sell computing time, mostly to AI labs and tech companies. When demand is strong and credit is cheap, that model grows very fast. When lenders get nervous, it becomes fragile, because the hardware loses value quickly and the debt does not.

It is fair to note the obvious conflict here. SpaceX also sells computing capacity through its own data centers, so it competes directly with the companies Musk says will fail. A warning from a competitor who just secured $40 billion of his own financing is not a neutral market analysis. Several investors on X said as much. Still, Musk is not the only one saying it, and the market data so far points in the same direction.

How the market reacted

On Monday, neocloud shares fell again. In premarket trading, Reuters reported that CoreWeave dropped 4.5%, Nebius 3.4% and IREN 3.3%. Applied Digital and TeraWulf lost more than 2% each, and Core Scientific fell 1.7%.

Bar chart of premarket share moves on October 26, 2026: CoreWeave minus 4.5 percent, Nebius minus 3.4 percent, IREN minus 3.3 percent, Applied Digital and TeraWulf more than minus 2 percent, Core Scientific minus 1.7 percent

These drops come on top of a brutal month. CoreWeave, Nebius and IREN had each lost more than 30% in October before Musk posted anything. On October 21, CNBC reported that CoreWeave was down 36% for the month, valued at roughly $31 billion compared with a peak of about $90 billion, with its unsecured debt yielding about 15% and the spread on its five year credit default swaps at 1,145 basis points. Those are levels that credit markets usually associate with distress. In Australia, data center operator NextDC plunged 40% in one day after slashing its annual earnings forecast, and Firmus cancelled its planned IPO.

Reuters also cited research firm Gartner, which named CoreWeave as the neocloud most exposed to rising leverage, and said that 80% of these companies could cease independent operations by 2032. According to Reuters, bonds of Nebius maturing in 2034 and of CoreWeave maturing in 2033 hit record lows, and investors betting against these stocks have made about $6 billion in paper profits.

The companies themselves push back. Nebius chief revenue officer Marc Boroditsky told CNBC that Nebius is "not carrying the kind of debt" CoreWeave has, and that if it brought its 2027 capacity online, it would sell out. The message from the neoclouds is consistent: demand from customers has not changed. The problem, at least for now, is not demand. It is the price and availability of money.

Why SpaceX can borrow when others cannot

Musk's argument for why SpaceX got its deal is simple: it is investment grade, and it has Starlink and launch revenue. The latest quarterly numbers explain why lenders are comfortable. SpaceX reported revenue of $14.58 billion for the third quarter, up 161% from a year earlier. Its AI business, which the company calls "Super Intelligence" or SI, has become its largest unit, with revenue up almost tenfold to $7.56 billion from $813 million a year earlier, according to CNBC.

Bar chart in billions of dollars: SpaceX SI revenue of 0.81 in Q3 2025 and 7.56 in Q3 2026, total revenue of 14.58 in Q3 2026, and capital expenditure guidance of 35 to 40 for Q4 2026

The spending is just as striking. Chief financial officer Bret Johnsen expects capital expenditure of $35 billion to $40 billion in the fourth quarter alone. In the press release he said SpaceX is seeing "less than one year payback periods" on its SI infrastructure, and that the lease "diversifies our funding sources" for getting more compute online to power Grok for businesses and consumers and to serve cloud customers. A payback period under a year would be an extraordinary return on hardware, and it is a company claim, not an audited figure. But it is exactly the kind of story lenders want to hear, and it is backed by a diversified revenue base that a pure neocloud does not have.

Note also the structure. A finance lease where an Apollo entity owns the GPUs gives lenders a direct claim on the hardware, while SpaceX gets the chips without paying for them up front. That is the same basic idea that made chip backed debt popular in the first place. Bloomberg reported this weekend, as summarized on Techmeme, that loans backed by AI chips are on track to become a trillion dollar market, with chipmakers guaranteeing a significant share of them. A senior director at Fitch Ratings quoted in that report warned that vendor financing is happening "at a scale that we've never seen before" and that the risk is concentrated.

The circular money problem

The SpaceX deal is clean in one sense: the lenders are traditional asset managers, not other AI companies. That is not always the case in this industry. A BIS Bulletin published on October 1 by the Bank for International Settlements looked at investment deals involving AI firms between 2021 and 2025, and the numbers are worth keeping in mind as the financing story develops.

Four bars from BIS Bulletin 137: 28.7 percent of AI firms' outgoing investment by deal value went into other AI firms, 55.2 percent of the money invested into AI firms came from other AI firms, 46.4 percent of AI to AI deal value also involved a supplier or customer relationship, and 16.1 percent of AI to AI deals by count had such a supply chain tie

According to the BIS, 28.7% of AI firms' investment deals by value went into companies that were also AI firms, and 55.2% of the money invested into AI firms came from other AI firms. Among those AI to AI deals, 46.4% by value also involved a commercial supply relationship between investor and target. That is the pattern where a chipmaker or cloud provider invests in a customer that then spends the money on the investor's products. The BIS says these relationships have economic logic, such as securing critical inputs, but warns that they bring macroeconomic risks and make the system harder to see through.

Put that next to the current credit squeeze and the risk becomes clearer. If a large part of the money in AI comes from other AI companies, and a large part of the hardware is financed with debt backed by the hardware itself, then a drop in GPU rental prices or a slowdown in demand can travel quickly through the chain: from neoclouds to their lenders, to the chipmakers that guaranteed some of those loans, and back to the AI labs that depend on all of them for compute.

What to watch next

There are a few concrete things to follow in the coming weeks.

  • Pricing of the SpaceX lease. The fixed rate is still open and will be set before the master lease is signed. Where it lands will show what even an investment grade AI borrower has to pay right now.
  • Big Tech earnings. The next quarterly reports from the large cloud providers will show whether the huge AI capital spending turns into revenue. Strong cloud numbers would help the whole sector, weak ones would add pressure.
  • Neocloud refinancing. Watch for delayed IPOs, emergency equity raises, asset sales or mergers. Musk predicted consolidation, and the first deals would confirm that the shakeout has begun.
  • Stranded compute. If Palihapitiya is right, some GPUs delivered in 2027 will sit idle or rent below cost. That would show up first in GPU rental prices and in the utilization numbers that neoclouds report.

My take

This is a story about who gets to keep building. The AI boom has always been capital intensive, but for the last two years almost anyone with a credible pitch could borrow money to buy Nvidia chips. That window seems to be closing, and the companies that can still raise $40 billion on reasonable terms are the ones with cash flow from somewhere else: hyperscalers, Meta, and now SpaceX with Starlink. That is good news for them and bad news for competition, because fewer and larger owners of compute means fewer independent suppliers for the AI labs and startups that rent it.

At the same time, I would not treat Musk's "guppy-level death" line as a neutral forecast. He runs a direct competitor, he has every reason to make lenders nervous about his rivals, and his own post overstated how final the deal was. The more reliable signals are the boring ones: credit default swap spreads, bond yields, cancelled IPOs and the BIS data on circular financing. Those already pointed to trouble before Sunday. Musk simply said out loud what the credit market had been pricing in for weeks.

Sources: SpaceX Form 8-K, SpaceX press release, Reuters via CNA, Reuters via Investing.com, CNBC on the deal, CNBC on neocloud stocks, CNBC on CoreWeave credit, BIS Bulletin 137, Techmeme.

Source: reuters.com

Newsletter

The AI news that matters, in your inbox.