AI chip lease deals rarely come bigger than this one. On 26 October 2026 SpaceX told the US Securities and Exchange Commission that it had signed a binding commitment letter with an entity managed by Apollo Global Management. Apollo’s vehicle will buy graphics processing unit pods and related equipment and lease them to SpaceX, which expects principal payments to total about $40 billion over roughly five years.
Elon Musk had already trailed the deal on X the day before, and he used it to make a point about everyone else. Debt markets, he wrote, “have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.” The 200-plus “neoclouds” that rent out GPU capacity, he added, “are going to see guppy-level death and consolidation.” Their shares fell in early trading on Monday before recovering later in the day.
This guide explains what the SpaceX AI chip lease actually commits the company to, how the finance lease structure works, where Musk’s description goes further than the filing, and why the rival GPU clouds took the warning personally. It ends with what the deal means for any business that buys AI compute. For background on the companies Musk was aiming at, see our earlier explainer, Neoclouds and the Enterprises That Need Them.
Table of contents
- What the SpaceX AI Chip Lease Covers
- How the AI Chip Lease Works as Finance
- “Closed” or Not? Musk’s Claim Versus the Filing
- Musk’s Warning to Neocloud Rivals
- How the Market Reacted to the AI Chip Lease News
- Why the AI Chip Lease Favours SpaceX
- The Risks in a Five-Year AI Chip Lease
- How Many Vera Rubin Racks Could SpaceX Get?
- What the AI Chip Lease Means for Businesses Buying Compute
- Our Verdict on the SpaceX AI Chip Lease
- AI Chip Lease FAQs
- References
What the SpaceX AI Chip Lease Covers
The AI chip lease is simple to describe and complicated to judge. SpaceX gets access to a very large quantity of AI hardware without paying for it up front. Apollo’s vehicle owns the equipment and SpaceX pays rent that covers the cost, interest and expenses over about five years.
The AI chip lease in brief
According to the AFP report carried by Tech Xplore and others, the chips are Nvidia’s Vera Rubin GPUs, which Nvidia began rolling out in the second half of 2026. Apollo will make them available gradually, starting in the first half of 2027. The contract runs for about five years and is worth roughly $40 billion, plus interest and certain fees.
AFP also notes that the contract “has yet to be finalized.” That detail matters, because it is the difference between a done deal and a firm intention. The AI chip lease is binding as a commitment, but the master lease itself still has to be signed.
What the 8-K filing says
The primary source is a short Form 8-K, filed under Item 8.01 (“Other Events”) and signed by SpaceX’s chief financial officer, Bret Johnsen. It gives 24 October 2026 as the date of the event and 26 October as the filing date. SpaceX now reports to the SEC as a Nasdaq-listed company under the ticker SPCX.
The filing says SpaceX “entered into a binding commitment letter to enter into a master finance lease” with the Apollo-managed lessor. The lessor “will acquire and lease to the Company certain graphics processing unit pods and related equipment” for use at data centers that SpaceX selects. The filing does not name Nvidia or Vera Rubin. That identification comes from the press reports.
AI chip lease deliveries in tranches from 2027
The equipment will arrive in batches, which the filing calls tranches. Each tranche gets its own lease schedule, with a term of “approximately five to five and one-quarter years” from the date that schedule starts. The schedules are expected to begin “during the first half of 2027 upon delivery of the GPUs.”
So the AI chip lease does not switch on all at once. Each delivery starts its own five-year clock, which spreads both the hardware and the repayments across time. The table below sets out the main terms as disclosed.
| Term | What SpaceX disclosed | Source |
|---|---|---|
| Lessor | An entity managed by Apollo Global Management | 8-K |
| Status | Binding commitment letter; master lease expected to close before year end | 8-K |
| Equipment | GPU pods and related equipment (Nvidia Vera Rubin, per AFP) | 8-K, AFP |
| Total principal | About $40 billion over the lease terms | 8-K |
| Term per tranche | About five to five and a quarter years | 8-K |
| Rent | Quarterly in advance: 1/20 of financed cost, plus fixed-rate interest and costs | 8-K |
| Interest rate | Fixed, but not yet set; determined at pricing | 8-K |
| First deliveries | First half of 2027 | 8-K |
How the AI Chip Lease Works as Finance
Musk called the arrangement debt. The filing calls it a finance lease. Both descriptions are fair, and the gap between them explains a lot about why Apollo was willing to commit so much money.
An AI chip lease versus a loan
In a finance lease, the lender buys an asset and rents it to the user for most of its useful life. The user carries the asset on its own books and pays rent that works like loan repayments. Decrypt describes it as “a long-term rental that works like a loan,” which is a good plain-English summary.
The difference from an ordinary loan is security. Apollo’s vehicle owns the GPUs. If SpaceX stopped paying, the lessor would not be an unsecured creditor waiting in line; it would own the hardware. That makes a large AI chip lease easier to fund than a general corporate loan of the same size, as long as the hardware keeps its value.
The one-twentieth rent formula
The 8-K spells out the rent. Each quarter, paid in advance, SpaceX owes “an amount equal to one-twentieth of the financed cost of the applicable tranche.” On top of that comes interest “at a fixed rate per annum on the outstanding lease balance” and “certain ongoing costs and expenses.”
Twenty quarterly payments of one-twentieth repay the whole cost in exactly five years. Because interest is charged on the outstanding balance, the interest share falls as the principal is paid down. The chart below shows how that works for an illustrative $10 billion tranche. The tranche size is our example, not a figure from the filing.
The arithmetic: one-twentieth of $10 billion is $500 million a quarter, or $2 billion a year, so the balance drops by $2 billion each year and reaches zero after year 5. Across the full AI chip lease, one-twentieth of $40 billion would be $2 billion a quarter if every tranche were outstanding at once. Because deliveries are staggered, the real quarterly figure at any moment depends on how many tranches have started.
Why Apollo backed the AI chip lease
Apollo is not new to Musk’s AI spending. AFP reports that in January 2026 Apollo led a $3.5 billion financing package that let an investment fund buy Nvidia chips and lease them to xAI for the Colossus 2 data center in Memphis. The new AI chip lease is about 11 times that size: $40 billion divided by $3.5 billion is 11.4.
Private credit firms like Apollo lend directly to companies instead of going through banks, as Decrypt explains. Lending against hard assets with a creditworthy tenant is exactly the kind of deal these funds want, and a five-year AI chip lease with an investment-grade borrower fits the pattern.
"Closed" or Not? Musk's Claim Versus the Filing
Coverage of the AI chip lease split over a single word. Some outlets reported that SpaceX had “closed” $40 billion of financing. Others said it had agreed to a deal that was not yet final. The difference comes from what Musk wrote compared with what the company filed.
What Musk wrote on X
Musk’s post on Sunday 25 October, quoted by Decrypt and CNBC, read: “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 continued: “SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.”
CNBC reports that Musk was replying to Chamath Palihapitiya, the founder of Social Capital, who had laid out a gloomy view of the data center buildout. More on that view below.
What the filing says
The 8-K describes a binding commitment letter, not a signed lease. It says entry into the master finance lease “is subject to certain specified closing conditions and is expected to close prior to year end.” The fixed interest rate “will be determined at pricing,” which has not happened yet.
The forward-looking statements section is blunt about the risk. It warns “that the transaction may not close on the anticipated terms or timeline, or at all,” and that deliveries, terms or total payments “may differ from current expectations.” None of that is unusual for an 8-K, but it does not match the word “closed.”
| Point | Musk on X (25 Oct) | SpaceX 8-K (26 Oct) |
|---|---|---|
| Status | “closed” | Binding commitment letter; closing expected before year end |
| Structure | “debt” | Master finance lease |
| Size | $40B | About $40B in principal, plus interest and costs |
| Interest | Not mentioned | Fixed rate, set at pricing, not yet determined |
| Hardware | “GPUs” | GPU pods and related equipment, delivered in tranches |
Why the wording matters
For most readers the difference is small: Apollo has committed, and the deal is very likely to happen. For investors in rival GPU clouds it matters more. Musk’s argument was that SpaceX can raise money that others cannot. A signed commitment supports that argument, but the final cost of the AI chip lease, which depends on the rate set at pricing, is still unknown.
Musk's Warning to Neocloud Rivals
The bigger headline was not the AI chip lease but the threat. Musk did not just announce a financing; he predicted that most of SpaceX’s competitors in GPU rental would fail.
“Guppy-level death”
“That means the 200+ neoclouds are going to see guppy-level death and consolidation,” Musk wrote. “Some will make it through, but not many.” Guppies are known for breeding in huge numbers and dying in huge numbers, so the image is of a crowded field thinned out fast.
TradingKey, summarising the posts, adds that Musk also said SpaceX would not acquire neocloud companies. In other words, he does not plan to buy the survivors cheaply; he expects to compete with whoever is left.
Who the neoclouds are
Neoclouds are specialist companies that rent out GPU capacity to AI developers, as opposed to the general cloud giants such as Amazon, Microsoft and Google. The best known are CoreWeave, Nebius and IREN. Decrypt notes that CoreWeave started in 2017 mining Ethereum with graphics cards before pivoting to AI.
These companies typically fund GPU purchases with a mix of share sales and heavy borrowing, then rent the capacity out under multi-year contracts. That model works when money is cheap and demand is strong. It is exposed if either changes, which is why an AI chip lease announced alongside a warning about credit markets landed so hard.
SpaceX is now a rival cloud seller
CNBC points out the obvious conflict: “SpaceX is a direct competitor to the neoclouds which also rent compute.” According to AFP, SpaceX rents computing capacity to Anthropic, the maker of the Claude chatbot, and to Google, and those deals have become a major source of revenue.
So the warning was not neutral commentary. It came from a competitor with a fresh $40 billion of hardware financing, talking about rivals who must pay more for theirs. That does not make it wrong, but readers should weigh it accordingly.
How the Market Reacted to the AI Chip Lease News
Neocloud shares were already under pressure before Musk posted about the AI chip lease. His comments landed on a sector that investors had been selling for weeks.
A sell-off that started earlier
CNBC reports that over the month before the posts, CoreWeave and Nebius were each down more than 30% and IREN had fallen about 40%. The sell-off began after the spread on CoreWeave’s five-year credit default swaps reached levels usually seen as a sign of distress, and the yields on its bonds climbed.
Bars are scaled to IREN’s 40%: 30 is 75% of 40. Because CNBC gives CoreWeave and Nebius only as “more than 30%,” both bars use the 30% floor. Decrypt gives a firmer figure for CoreWeave over a slightly different window: from $87 on 30 September to $61 on 16 October, a fall of about 30%.
Monday’s moves
On Monday 26 October, CoreWeave, Nebius and IREN all fell in early trading. TradingKey put the pre-market falls at more than 4% each. CNBC reports that “the stocks all rebounded later in the day,” which suggests investors read Musk’s post as a sharper statement of a worry they already had, not as new information.
TradingKey also cites Reuters data showing that short sellers had made around $6 billion in paper gains on neocloud and data center stocks since the start of October. Betting against the sector was already a crowded trade before the AI chip lease was announced.
How rivals and Nvidia answered
Nebius pushed back directly. Its chief revenue officer, Marc Boroditsky, told CNBC that “we’re not carrying the kind of debt that they are,” distancing the company from CoreWeave. He said that if Nebius brought its 2027 capacity online, it would sell out.
Nvidia, which has invested in several neoclouds, had already tried to calm the market at its GTC event in Berlin the week before. Chief executive Jensen Huang told CNBC that demand for neoclouds is “off the charts,” and that the missing piece is working capital. “The demand of the capital is much greater than the availability of the capital. That’s going to get solved,” he said.
Why the AI Chip Lease Favours SpaceX
Musk’s argument rests on one claim: that SpaceX can borrow against businesses that have nothing to do with AI. The evidence for that claim is fairly strong.
Investment grade and outside revenue
Decrypt reports that Moody’s rated SpaceX Baa1 in June 2026, an investment-grade level. An investment-grade rating tells lenders a borrower is unlikely to miss payments, which lowers the cost of borrowing and widens the pool of willing lenders.
Starlink, SpaceX’s satellite internet service, and its rocket launch business both generate revenue that does not depend on AI demand. If GPU rental prices fell sharply, SpaceX could still make its lease payments from those businesses. A pure GPU cloud cannot say the same, and that is the core of Musk’s point.
| Factor | SpaceX | A typical neocloud |
|---|---|---|
| Credit rating | Baa1, investment grade (Moody’s, June) | Often below investment grade |
| Revenue outside AI | Starlink and launch services | Little or none |
| Own AI demand | Grok and other in-house models | Depends on outside customers |
| Cost of borrowing | Fixed rate, still to be set | CoreWeave bonds yielded about 15% (Decrypt) |
| Main risk | Concentration of capital in AI hardware | Refinancing before contracts pay off |
The cost of money for rivals
Decrypt reports that CoreWeave carries roughly $50 billion of debt, and that its unsecured bonds yielded about 15% in mid-October. That means investors wanted around 15% a year to lend to it without specific assets as backup. A secured AI chip lease with an investment-grade tenant will almost certainly price far below that, although the exact SpaceX rate is not yet known.
When two companies buy the same chips and one pays much less to finance them, the cheaper one can undercut on price and still make money. That is the practical meaning of Musk’s warning.
Colossus, Grok and in-house demand
SpaceX absorbed xAI in February 2026, according to AFP and Decrypt. xAI had built two large data centers in Memphis, Tennessee, Colossus 1 and Colossus 2, in about two years. Together they house several hundred thousand GPUs. Musk has said Colossus 2’s computing power should more than double by the end of 2026 compared with the end of September.
IBTimes reports that SpaceX said the new hardware would support GPU deployments for Grok and for existing and future cloud customers. Having its own large AI workload means SpaceX is less exposed to a fall in outside demand than a rental-only provider. If customers leave, Grok can still use the chips. For context on SpaceX’s other big AI move this year, see our piece on the SpaceX Cursor acquisition.
The Risks in a Five-Year AI Chip Lease
None of this makes the deal risk-free. A five-year lease on fast-moving hardware is a bet on how long that hardware stays useful, and several well-known investors think 2027 will be rough.
Chips age faster than an AI chip lease
GPUs lose value quickly, because each new Nvidia generation makes the last one less competitive. Decrypt points to CoreWeave’s second quarter of 2026, when it booked nearly $1.4 billion in depreciation and amortisation against about $2.6 billion in revenue. That is a charge of about 54 cents for every dollar of revenue ($1.4 billion divided by $2.6 billion).
An AI chip lease does not avoid this. SpaceX will pay for Vera Rubin hardware until about 2032, while Nvidia is likely to release at least one or two newer generations in that time. The question is whether the chips still earn enough in years four and five to justify the rent.
GPUs as airline seats
Ken Griffin, the founder of Citadel, offered a sharp analogy at his firm’s markets conference on 20 October, as reported by Decrypt. GPUs, he said, resemble airline seats: once they exist, adding one more costs almost nothing, so prices can collapse when supply outruns demand.
If that happens, the owner with the lowest financing cost survives longest. That supports Musk’s view of who wins, but it also means SpaceX’s rental revenue could fall, even if it can still afford the lease.
Stranded compute in 2027
Palihapitiya, whose post Musk was replying to, expects “a lot of partially stranded compute and compute sitting in inventory in 2027 than is currently forecasted.” That hardware, he wrote, would be “bought and paid for but effectively underearning (below total cost of capital) or collecting dust.”
“2027 will be a turbulent year because it will clear out a bunch of wrong/bad bets,” he added. “It will also be a consolidation year for a few.” The first tranches of the SpaceX AI chip lease start in exactly that year.
The rate is still open
The final cost of the AI chip lease depends on the fixed rate set at pricing. If credit markets tighten further before then, as Musk himself says they have, the rate could be higher than SpaceX would like. Being investment grade protects SpaceX from the worst of that, but it does not make it immune.
How Many Vera Rubin Racks Could SpaceX Get?
The filing says nothing about how many GPUs the $40 billion AI chip lease buys. One analyst estimate has filled that gap in the coverage, but it needs careful handling.
The GF Securities estimate
IBTimes reports that Jeff Pu of GF Securities estimates SpaceX could receive about 20,000 of Nvidia’s Vera Rubin NVL72 racks in 2027, out of projected global supply of 96,000. Pu reportedly said SpaceX had originally sought 30,000. He also estimates that six leading neocloud providers will buy a combined 34,000 racks.
Nvidia describes the NVL72 as a rack-scale system that combines 72 Rubin GPUs with Vera CPUs, networking, power and liquid cooling, installed as a complete unit. The chart below shows Pu’s estimates side by side.
Bars are scaled to global supply: 34,000 is 35% of 96,000, and 20,000 is 21%. SpaceX’s estimated share would equal about 59% of the six neoclouds’ combined total (20,000 divided by 34,000). If the 30,000-rack figure Pu says SpaceX first sought were met, it would be about 31% of supply.
What is and is not confirmed
IBTimes is careful to say these are analyst estimates. Neither SpaceX nor Nvidia has confirmed them as orders, allocations or delivery schedules, and it is unclear whether Pu’s six neoclouds include SpaceX. The financing has not been publicly linked to the rack numbers, and no rack price has been disclosed.
That means nobody can yet say how much of the $40 billion AI chip lease goes on Nvidia systems rather than other equipment. Anyone quoting a GPU count for the deal is guessing. For a comparable Vera Rubin commitment on the buyer’s side, see our report on Anthropic’s $45 billion Nscale deal.
What the AI Chip Lease Means for Businesses Buying Compute
Most businesses will never sign a $40 billion lease. But many now rent GPU capacity, directly or through AI services built on it, and the forces behind this deal affect them too.
Check your provider’s balance sheet
If you rent GPUs from a specialist cloud, its financial health is now part of your risk. Ask how its hardware is financed, how much debt it carries and when that debt falls due. A provider that has to refinance in a tight market could raise prices, cut service or, in the worst case, fail.
This applies to managed AI services too. Many AI tools resell capacity from neoclouds underneath. Knowing which provider sits beneath your supplier is now a sensible part of due diligence for cloud computing contracts. Ask about cybersecurity and data handling at the same time, because a provider under financial strain may also cut spending on security.
Avoid single-provider lock-in
Musk’s warning points to consolidation. If your workloads run on one smaller provider, plan for what happens if it is bought or closes. Keep models, data and deployment scripts portable, and test that you can move a workload to a second provider in days, not months.
Read the exit terms
Long GPU contracts can lock you into today’s prices for years. If Griffin is right and prices fall when supply catches up, a three-year commitment signed now could look expensive by 2028. Look for break clauses, price reviews and the right to move to newer hardware.
Expect more price swings
Very large buyers financing hardware cheaply, like SpaceX with this AI chip lease, can push rental prices down. That is good for buyers in the short run. It may also squeeze smaller providers, which could reduce choice later. Budget for volatility in both directions.
Our Verdict on the SpaceX AI Chip Lease
The SpaceX AI chip lease is real, very large and well structured for SpaceX. A secured finance lease with an investment-grade tenant and revenue from Starlink and launches is about as safe as a $40 billion bet on AI hardware can be. The one-twentieth rent formula keeps repayments predictable and spreads them across staggered deliveries.
It is not yet “closed.” The master lease still has to be signed, the rate is not set, and the hardware will not arrive until 2027. Musk’s description runs ahead of his own company’s filing.
His warning to rivals is self-interested, but it reflects a real gap in the cost of capital that the market had already priced in. The most useful lesson for buyers of AI compute is the least dramatic one: know who finances the hardware you depend on.
AI Chip Lease FAQs
What is the SpaceX AI chip lease?
It is a planned master finance lease in which an Apollo Global Management vehicle will buy GPU pods and related equipment and lease them to SpaceX. SpaceX expects principal payments of about $40 billion over terms of roughly five years.
Has the AI chip lease closed?
Not yet. SpaceX’s 8-K describes a binding commitment letter. The master lease is subject to closing conditions and is expected to close before the end of 2026. Musk wrote on X that SpaceX “closed” the debt, but the filing is more cautious.
Which chips are involved?
AFP reports that the chips are Nvidia Vera Rubin GPUs, delivered gradually from the first half of 2027. The filing itself refers only to “graphics processing unit pods and related equipment.”
How much will SpaceX pay each quarter?
Each tranche costs one-twentieth of its financed cost per quarter in principal, plus fixed-rate interest on the outstanding balance and some ongoing costs. The rate has not been set, so the total cost is not yet known.
What did Musk say about neoclouds?
He said debt markets had nearly shut to companies without substantial non-AI revenue, and that “the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.”
Should businesses worry about their GPU provider?
It is worth checking. Ask how your provider finances its hardware, plan for portability between providers, and read contract exit terms carefully before committing to long GPU deals.
References
Space Exploration Technologies Corp. Form 8-K, filed 26 October 2026 (SEC EDGAR)
SpaceX strikes AI chip lease deal as Musk warns rivals (AFP via Tech Xplore)
SpaceX Locks Up $40 Billion for AI Chips as Musk Warns Most Neoclouds Will Die (Decrypt)
Elon Musk says more than 200 neoclouds face ‘guppy-level death’ (CNBC)
SpaceX Wanted 30,000 Nvidia Racks but May Get 20,000 (IBTimes UK)
Neocloud Sector Falls Pre-Market as Musk Warns of Industry Consolidation (TradingKey)
SpaceX strikes AI chip lease deal as Musk warns rivals (AFP via Macau Business)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
