On Sunday, Elon Musk said SpaceX had "closed" a $40 billion deal for artificial intelligence chips. On Monday, the paperwork told a more careful story. SpaceX $40 billion AI financing is now confirmed in a regulatory filing — but as a binding commitment to a finance lease, not as a completed purchase, and not as money already spent. The distinction matters, because it changes what the deal is, when the chips arrive, and what could still change. This article explains the filing in plain English, and how it differs from what was reported on Sunday morning.
First, the timeline. In early October, the Financial Times reported that SpaceX was seeking about $40 billion to buy Nvidia chips, with Apollo Global Management leading the effort and a mix of roughly $10 billion in bank loans and $30 billion in investment-grade debt. Pacific Investment Management Company was named as being in talks. On Sunday 25 October, Musk posted on X that SpaceX had closed $40 billion in debt financing for graphics processing units (GPUs). On Monday 26 October, Reuters, TechCrunch and Dow Jones reported the filing itself — and the structure it describes is different from both the October plan and the Sunday post.
SpaceX $40 billion AI financing: what the filing actually says
According to reporting on the filing by Reuters and TechCrunch, SpaceX has entered into a binding commitment for roughly $40 billion in financing for AI chips and related data-centre equipment, which SpaceX described as the largest committed GPU financing. The counterparty is an entity managed by Apollo Global Management. That entity will buy the GPUs and related server hardware and lease them to SpaceX under a master finance lease. Goldman Sachs acted as sole structuring agent, with anchor commitments from funds and accounts managed by BlackRock, Goldman Sachs Asset Management, and capital accounts advised by KKR. Pacific Investment Management Company, named in the early-October talks, does not appear in the final anchor list reported on Monday.
The equipment is specific: the filing points to Nvidia's next-generation Vera Rubin chips for what SpaceX calls its terrestrial compute buildout — the ground-based data centres that power its AI work, as opposed to its rockets and satellites. No GPU count and no per-chip price has been disclosed, and none should be inferred from the $40 billion headline.
A lease, not a loan — and not a purchase
The most important correction in the filing is about what kind of transaction this is. SpaceX is not buying $40 billion of chips outright, and it has not spent $40 billion. Under the lease structure, the Apollo-managed entity buys the equipment and SpaceX pays to use it. Aggregate principal payments will be about $40 billion, excluding interest and other expenses, spread over the life of the leases. Payments are quarterly, generally one-twentieth of the financed cost of the equipment plus interest at a fixed annual rate on the outstanding lease balance plus certain additional expenses. Each tranche of equipment will be leased for about five to five-and-a-quarter years, beginning when that tranche is delivered.
Two things are still undetermined. The fixed interest rate had not yet been set when the filing was reported, so the all-in cost of the deal cannot yet be calculated. And the arrangement itself is a binding commitment subject to specified closing conditions, expected to be finalised before the end of 2026. Leases are expected to commence in the first half of 2027, when the GPUs and equipment are delivered. Musk's Sunday statement that funding mainly targets chips arriving in the first and second quarters of 2027 is consistent with that timetable, and he also said the financing would not be a major factor in this quarter's results.
Why lease rather than borrow to buy? For hardware that improves every year, leasing spreads the cost over the equipment's useful life, preserves cash for other spending, and avoids owning a depreciating asset outright. That logic cuts both ways: SpaceX gains flexibility, but it also takes on years of quarterly payments for hardware that may be superseded before the lease ends. The filing's structure is therefore best read as a financing choice, not simply as a shopping receipt.
Why SpaceX needs the chips
The scale of the deal only makes sense against the scale of SpaceX's AI business. TechCrunch, citing Axios, reports that SpaceX's AI segment — the Grok chatbot, the Cursor coding tool, and cloud compute services — brought in $7.6 billion of $14.6 billion in total revenue in the third quarter of 2026, more than Starlink and launches combined. Capital expenditure surged to nearly $20 billion in the quarter, up from $6.9 billion a year earlier, and chief financial officer Bret Johnsen has said SpaceX expects to spend $35–40 billion in the fourth quarter of 2026 alone. The $40 billion lease sits inside an even larger spending programme, not on top of a small one.
Johnsen said the lease diversifies SpaceX's funding sources and addresses a notable portion of the funding required for its growth plans, and that the company is seeing payback periods of less than a year on AI infrastructure investments as it scales compute to power Grok and cloud services. That payback claim is SpaceX's own assessment, and it should be read alongside the scepticism elsewhere in the market about whether AI revenue can keep pace with AI spending.
The underlying technology explains the urgency. Training and running large AI models requires thousands of graphics processors working in parallel, connected by very fast networking, with power and cooling far beyond ordinary computer rooms. A single training run can need tens of megawatts of electricity, and idle processors still cost money while they wait for data, power, or network capacity. That is why the bottleneck for AI companies is not just buying chips, but financing, powering, and cooling entire data centres — and why the financing structure matters as much as the hardware.
What the deal means for neoclouds
Musk used the announcement to warn about neoclouds — specialist cloud providers that rent out GPU computing for AI, usually from data centres packed with graphics processors, whose revenue depends almost entirely on AI workloads. He said debt markets had almost entirely shut off money for companies without substantial revenue independent of AI data centres, that more than 200 neoclouds face what he called "guppy-level death and consolidation," and that SpaceX is not buying any neoclouds. Those counts and characterisations are Musk's assertions, not independently verified figures.
Markets reacted on Monday, but not in a straight line. In premarket trading, shares of neocloud companies including CoreWeave, Nebius, and IREN fell, with CoreWeave down about 4.5% and Nebius about 3.4%, according to Reuters. Later on Monday morning, some of those moves partly reversed, with CoreWeave reported up more than 2% while Nebius remained down about 2%. A single day's trading should not be read as a settled verdict on the sector.
Independent analysis points in a similar direction but with a longer horizon. Gartner has forecast that more than 80% of neocloud providers will not survive independently, a forecast that reflects the same problem Musk identified: a pure-play GPU rental business has no second income stream when AI demand cools or financing tightens, while SpaceX has Starlink and launch revenue and the large cloud providers have diversified businesses. That distinction — diversified revenue versus dependence on a single AI workload — is the real story behind the neocloud warning, and it applies whether or not any individual company fails.
What could still change
Three things could still move this deal before a single chip is delivered. The interest rate has not been set, so the total cost is unknown. The closing conditions have not yet been satisfied, so the commitment could still be delayed or altered before the expected year-end finalisation. And delivery itself is a 2027 event: leases begin only when equipment arrives in the first half of 2027, so any slip in Nvidia's Vera Rubin production or in data-centre readiness would push the timetable. Readers should also note what this article does not claim: no GPU count, no per-chip price, no all-in total including interest, and no final role for any lender beyond those named in Monday's reporting.
The honest summary is this: SpaceX has locked in the structure and the scale of its AI chip financing, with named partners and a delivery timetable, but it has not yet closed the lease, priced the debt, or received the hardware. The $40 billion is a commitment to pay over about five years, not a purchase already made. That is still a very large bet on AI computing — and, given SpaceX's AI revenue and spending plans, a bet the company clearly believes it can afford. Whether the neocloud shakeout Musk predicts follows is a separate question, and Monday's mixed market reaction suggests investors are not yet sure of the answer.
Sources: Reuters (6 Oct 2026 and 26 Oct 2026), TechCrunch (26 Oct 2026), Dow Jones via Morningstar (26 Oct 2026), BusinessWorld Online (7 Oct 2026), MarketWatch (26 Oct 2026), CoinDesk (26 Oct 2026), Cointelegraph (26 Oct 2026). Full links: https://www.reuters.com/business/media-telecom/spacex-seeks-40-billion-buy-nvidia-chips-ft-reports-2026-10-06/ · https://www.reuters.com/business/media-telecom/spacex-enters-40-billion-deal-with-apollo-ai-chip-financing-2026-10-26/ · https://techcrunch.com/2026/10/26/spacex-locks-in-40b-gpu-financing-deal-with-apollo/ · https://www.morningstar.com/news/dow-jones/202610265002/spacex-reaches-40-billion-financing-agreement-with-apollo-managed-entity-update · https://bworldonline.com/technology/2026/10/07/785188/spacex-seeks-40-billion-financing-led-by-apollo-to-buy-nvidia-chips-ft-reports/ · https://www.marketwatch.com/story/musk-says-spacex-landed-a-40-billion-debt-deal-and-forecasts-doom-for-some-ai-rivals-a0a4991e · https://www.coindesk.com/markets/2026/10/26/elon-musk-confirms-usd40-billion-spacex-debt-deal-for-ai-chips-warns-of-neocloud-shakeout · https://cointelegraph.com/news/musk-warns-over-200-ai-neoclouds-face-guppy-level-death
Related reading: SpaceX $40 Billion GPU Deal 2026: Musk's Neocloud Warning · Elon Musk Says a Tesla–SpaceX Merger Could Work. Here's What It Would Mean · Tesla's Flying Roadster: What the SpaceX Thruster Demo Really Proved