Elon Musk has confirmed that SpaceX has closed a SpaceX $40 billion deal to buy graphics processing units (GPUs) for artificial intelligence, and in the same breath he warned that most of the smaller AI cloud companies known as neoclouds are heading for failure or consolidation. The announcement, made in a post on X on Sunday 25 October 2026, is trending per news coverage across Reuters, SiliconANGLE, TipRanks, TradingView and other outlets — because it links two of the biggest questions in technology right now: how much the AI boom really costs, and who can still afford to pay for it.
SpaceX says the money will buy Nvidia chips and other GPUs that will arrive in the first and second quarters of 2027. Musk described the borrowing as primarily callable debt, meaning much of it can be repaid early, and said it would not be a major factor in the company's results this quarter. At the same time, he argued that debt markets have almost entirely stopped lending to companies that do not already have substantial revenue outside AI data centres. That contrast — SpaceX can borrow, most neoclouds cannot — is the heart of the story.
What the SpaceX $40 Billion Deal Actually Covers
The SpaceX $40 billion deal is debt financing, not a share sale and not a government grant. According to Reuters, Musk said on Sunday that SpaceX had closed $40 billion in debt financing for graphics processing units, underscoring how expensive it has become to build AI computing capacity. An earlier Financial Times report on 7 October had said SpaceX was in talks to raise $40 billion for Nvidia chips, citing Apollo Global Management as the lead investor and Pacific Investment Management Company among the institutions involved, with roughly $10 billion in bank loans and $30 billion in investment-grade debt. Musk's post did not name the lenders or confirm pricing, and Reuters noted it was not immediately clear from his post what he meant by the deal being "closed" or whether investors had agreed pricing for the debt.
What is clear is the purpose and the timetable. The GPUs will be delivered in the first half of 2027 and will fill SpaceX's AI data centres. Musk said the debt is primarily callable, which gives the company flexibility to repay it early if cash flow allows. He also said it would not be a major factor this quarter — a notable claim given that SpaceX expects capital expenditure of $35 billion to $40 billion in the fourth quarter of 2026 alone, according to chief financial officer Bret Johnsen. In other words, the $40 billion is large in absolute terms, but it sits inside an even larger spending programme.
Why does a rocket company need so many AI chips? SpaceX is no longer only a launch business. Its renamed Super Intelligence segment — covering cloud services, AI models and applications such as Grok — has become its largest revenue line, while Starlink provides a second, very different income stream from satellite internet. The chips are the raw material for that AI business: without GPUs, there is no capacity to rent to AI developers, no training for large models, and no inference capacity for everyday users. Buying at this scale is therefore a bet that demand for computing will keep growing into 2027, and that owning the hardware outright is cheaper than renting it from someone else.
What Is a Neocloud — and Why Musk Says Most Will Fail?
A neocloud is a company that builds or leases data centres, fills them with GPUs — usually from Nvidia — and rents that computing power to AI developers and other technology customers. Unlike the big hyperscalers such as Microsoft, Amazon and Google, which have many different businesses, neoclouds mostly focus on AI compute. Well-known examples include CoreWeave and Nebius. Their pitch is simple: AI labs need chips urgently, and a specialist provider can move faster than a giant cloud company.
The problem, as Musk framed it, is how that pitch is financed. Neoclouds typically build first and collect revenue later. They borrow heavily to buy chips and power, then hope contracted customers pay enough, for long enough, to service the debt. Musk wrote on X: "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 centres." He then added: "That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many."
He also ruled SpaceX out as a rescuer, writing: "SpaceX is not buying any neoclouds. Pointless." Early Uber investor Jason Calacanis echoed the consolidation view on X, writing simply: "Consolidation is coming." Analyst Dan Ives framed the wider race differently, predicting another 800 to 1,000 data centres in the United States over the next few years and calling the fight for computing power a "Game of Thrones" battle. In that framing, SpaceX is not a bystander — it is aggressively building its own hyperscaler-style infrastructure to compete with both neoclouds and traditional cloud providers.
Why the SpaceX $40 Billion Deal Matters for AI Chips
The SpaceX $40 billion deal matters because it shows the AI chip market splitting into two tiers. In the top tier are companies with an investment-grade credit rating and revenue from other businesses — SpaceX points to Starlink and rocket launches; the hyperscalers point to cloud, advertising, e-commerce and software. Those companies can still borrow at manageable rates. In the lower tier are pure-play neoclouds, whose only asset is the data centre itself and whose only revenue is rented compute. When lenders doubt that model, borrowing costs spike, and the whole business case wobbles.
Evidence of that split was already visible before Musk's post. Weiss Ratings flagged CoreWeave as the most exposed neocloud, citing $35.1 billion of recourse and non-recourse debt at 30 June, later borrowing lifting bonds and loans to roughly $48 billion to $50 billion, against $5.5 billion in cash at midyear, with $640 million in second-quarter net interest expense and a $626 million quarterly loss. It also noted yields on some CoreWeave bonds moving above 15 percent — a level associated with distressed debt. Nebius was described as more complex, with $8.55 billion in debt at 30 June, later increased by secured financing and convertible notes, but also holding $8 billion in cash and expecting customer prepayments. Applied Digital was ranked next, with about $6.4 billion in debt and $2.95 billion in cash at the end of August.
Reuters reported on 22 October that short sellers betting against US neocloud and data centre companies had made about $6 billion in paper profits so far in October, with about $17 billion of stock across CoreWeave, Nebius, IREN, Applied Digital and Core Scientific sold short. Short interest in CoreWeave stood at 23.7 percent of free float, up from 19.6 percent at the start of October, while Nebius had climbed to 20 percent from 15.4 percent at the start of 2026. Those figures do not prove any company will fail — short sellers can be wrong, and share prices can rebound — but they show how sceptical credit and equity markets had already become before Musk weighed in.
Gartner added to the gloom on 14 October, saying more than 80 percent of neocloud providers could stop operating independently by 2030, and that more than $3.5 trillion in AI infrastructure spending could face impairment by then. Its reasoning was blunt: enterprise and consumer AI revenue may not be enough to cover the capital being poured into chips before the current generation of silicon becomes obsolete. Musk's warning therefore landed on ground that was already nervous.
What SpaceX's Own Numbers Show
SpaceX's case for being in the top tier rests on its latest results. In the third quarter of 2026, reported on 20 October, revenue rose 161 percent year on year to $14.58 billion, beating expectations of about $12.7 billion. Net income was $4.4 billion, compared with a $1.1 billion loss a year earlier, and adjusted EBITDA reached $8.8 billion. Super Intelligence revenue surged to $7.56 billion, up from $813 million a year earlier, overtaking Connectivity revenue of $5.25 billion. Starlink consumer subscribers reached 13.6 million, up from 12 million in June and 7.2 million a year earlier.
At the same time, SpaceX's spending is enormous by any standard. Capital expenditure in the third quarter was $19.6 billion, including $16.4 billion for Super Intelligence infrastructure, and nameplate computing capacity reached 1.7 gigawatts, up from 1.4 gigawatts at the end of June. The company completed its first revenue-generating Starship mission in the quarter, deploying 26 Starlink V3 satellites. The lesson of those numbers is nuanced: SpaceX is growing fast and generating real revenue from AI and connectivity, but it is still spending more than it earns in cash terms, and its fourth-quarter capital expenditure guidance of $35 billion to $40 billion shows the spending is accelerating, not slowing.
That nuance matters for readers trying to judge Musk's warning. He is not saying AI demand is fake. He is saying the financing model for many of the companies chasing that demand is fragile. A company can be right about AI and still fail if it cannot refinance its debt, if its customers delay payments, or if power and facilities arrive late. Indeed, The Information reported that AI data centre tenants are now demanding penalties of more than $5 million per day if developers fail to deliver power and facilities on time, up from about $1 million earlier in the year — a sign that delays are already a live commercial risk.
What Happens Next in 2027
Three things are worth watching as the SpaceX $40 billion deal plays out. First, delivery: do the GPUs actually arrive in the first and second quarters of 2027, and does SpaceX rent that capacity quickly, as it did with its Colossus II build? Second, refinancing: do neocloud bond yields fall back, stabilise, or rise further, and do any providers announce mergers, asset sales or restructurings? Third, customer behaviour: do AI labs keep signing long-term compute contracts, or do they pause, renegotiate, or bring more computing in-house?
For ordinary users, the most visible effect may be indirect. If consolidation reduces the number of independent GPU providers, prices for rented AI computing could become less competitive in the short term, even as larger providers gain scale. If, on the other hand, a shakeout clears out over-borrowed capacity, the surviving providers may be more stable and better able to honour long-term contracts. Either way, the cost of AI — for app developers, researchers, startups and ultimately consumers — is tied to who owns the chips and how they paid for them.
There is also a wider market context. The week before Musk's post, Wall Street struggled as rising oil prices broadened a selloff beyond the AI trade, with the S&P 500 and Nasdaq recording a second weekly loss and the world's richest losing large sums on paper as AI spending fears sank technology stocks. That backdrop helps explain why a single debt announcement can move sentiment so sharply: investors are no longer asking only whether AI will grow, but whether the growth is being funded sustainably.
None of this means the neocloud story is over. Some providers have genuine contracted backlogs — CoreWeave, for example, reported a roughly $104 billion revenue backlog in its second quarter — and customer prepayments and secured financing can change the picture for individual companies. Musk's phrase "guppy-level death" is deliberately provocative, and predictions of mass failure should be treated as one prominent voice in a debate, not as a settled outcome. What is settled is narrower and more important: SpaceX has now publicly tied its own borrowing power to revenue outside AI data centres, and it has drawn a line between companies that have that cushion and companies that do not. The next few quarters will test whether that line holds.
Sources: Reuters (25 October 2026, via syndication); SiliconANGLE; TipRanks; TradingView; WCCFTech; SpaceX third-quarter 2026 results reporting by Quiver Quantitative, Motley Fool and BeInCrypto; Weiss Ratings neocloud risk analysis; Reuters neocloud short-selling report (22 October 2026); Gartner neocloud outlook (14 October 2026).
Related reading: Elon Musk Says a Tesla–SpaceX Merger Could Work. Here's What It Would Mean · Elon Musk Pledges a SpaceX Share for Every US Child: How the Trump Accounts Giveaway Would Work · ASML Posts Record Sales as AI Demand Outruns Supply