On October 25, 2026 (US time), Elon Musk posted on X that SpaceX had "closed" $40 billion in financing to buy GPUs. In the same post, he predicted that more than 200 "neoclouds" — new cloud providers that own large fleets of GPUs and rent out the computing power to companies — face "guppy-level death" and consolidation. But SpaceX itself is rapidly increasing its dependence on AI infrastructure: its AI division accounted for roughly 52% of revenue in the July-September 2026 quarter.
Still, the reason Musk distinguishes SpaceX from neoclouds appears to lie in its non-AI revenue base. SpaceX has the Starlink satellite communications service and its rocket launch business, and Starlink in particular generates substantial operating profit. Companies whose main business is renting out GPUs, by contrast, depend heavily on AI demand to recoup capital spending and repay debt. Comparing SpaceX's reported financing plan with the credit costs of leading neocloud CoreWeave suggests that the difference lies less in the business models themselves than in the strength of the earnings base and credit support behind the borrowing.
Two days after reports of a stalled financing, Musk claims the $40 billion has closed
Musk's comment came in reply to a post by investor Chamath Palihapitiya. Musk claimed that in recent weeks the debt market had all but stopped lending to companies without substantial revenue outside AI data centers. He then explained that SpaceX has investment-grade credit and earns revenue from Starlink and rocket launches. In another post, he added that most of the debt raised can be repaid early and will be used to buy GPUs delivered in the first and second quarters of next year. CNBC and TeslaNorth interpret that delivery timing as January-June 2027.
Musk also said the financing would not significantly affect this quarter's results. At the Bank of Mitsubishi UFJ's October 23 midpoint rate of 156.51 yen to the dollar, $40 billion comes to about 6.26 trillion yen. That is an extremely large amount for GPU-purchase financing, but as of the October 25 post, details such as the lenders, interest rate and repayment terms had not been disclosed.
Several reports on the financing had appeared since early October.
| Date (US time) | Main developments |
|---|---|
| Oct. 6 | The Financial Times reported a $40 billion financing plan combining roughly $10 billion in bank loans and roughly $30 billion in investment-grade bonds. Business Insider Japan also covered it |
| Oct. 7 | CNBC reported that talks were at a preliminary stage and that the GPUs were expected to serve as collateral |
| Oct. 15 | CoreWeave's credit default swaps (CDS) briefly rose above 1,140 bp (CNBC) |
| Oct. 20 | SpaceX reported July-September 2026 results, with revenue of $14.6 billion |
| Oct. 23 | Bloomberg reported that a plan to move the borrowing off SpaceX's own balance sheet was struggling amid turmoil in the AI market |
| Oct. 25 | Musk posted on X that the $40 billion financing had "closed" |
Notably, Musk claimed the deal was done just two days after Bloomberg reported that the financing was struggling. According to TeslaNorth and CNBC, Musk did not give the exact date of closing, and no separate formal announcement from SpaceX has been confirmed. It is therefore unclear whether the structure reported in early October — $10 billion in bank loans and $30 billion in bonds — was completed as is or whether the terms changed. Musk's post is important information, but the financing terms cannot be treated as confirmed fact.
SpaceX also gets about 52% of revenue from AI; the difference from neoclouds is Starlink's earning power
SpaceX's July-September 2026 revenue was $14.6 billion, of which its AI division, "Super Intelligence (SI)," accounted for $7.56 billion. That is about 52% of the total, meaning AI-related businesses now make up more than half of revenue. The AI division's revenue in the same period a year earlier was only $813 million, so the business mix changed dramatically in just one year. However, according to TeslaNorth, the AI division's revenue also includes $538 million in advertising revenue, so not all of it comes from renting out GPU computing resources.
Even so, GPU cloud is clearly one of the businesses driving the AI division's growth. Morningstar assessed that the expansion in AI revenue in the July-September quarter was driven mainly by renting out computing resources as a neocloud. According to the IPO filing (S-1) reported by Business Insider Japan, Anthropic has a contract to pay SpaceX $1.25 billion per month through May 2029. Discounted rates apply in May and June 2026, and the contract can be terminated by either side with 90 days' notice.
The scale of capital spending also confirms the tilt toward AI. Of SpaceX's $19.6 billion in capital expenditure in July-September, $16.4 billion went to the AI division — about 84% of the total. That AI-related capital spending drives the company's overall investment plan is something SpaceX shares with neocloud operators whose main business is GPU cloud.
There are also similarities in the financing structure. Under the plan reported by Bloomberg, SpaceX was considering raising funds to buy NVIDIA GPUs through a special purpose vehicle (SPV) and keeping that debt off the parent's balance sheet. An SPV is an entity set up to hold specific assets or liabilities, and depending on control relationships, contract terms and accounting requirements, it can sometimes allow financing without recording the debt on the parent's consolidated balance sheet. The reported plan aimed to limit the impact on SpaceX's own credit rating through this kind of off-balance-sheet treatment.
CNBC also reported on October 7 that the GPUs being purchased were expected to serve as collateral. In such a structure, the asset value of the GPUs and the usage contracts with customers determine the creditworthiness of the loan. The basic mechanism — borrowing against AI computing resources, then renting that equipment to customers to earn revenue — is not very different from the GPU-backed loans neoclouds use.
One difference, however, is that SpaceX can earn profit from businesses other than GPU cloud. While the AI division's operating profit in July-September was $431 million, the Starlink division posted an operating profit of about $2.4 billion. The space division, which includes rocket launches, recorded an operating loss of $7 million, so the business generating profit outside AI is effectively Starlink.
Given this earnings structure, the investment-grade credit that Musk emphasized as the difference from neoclouds is probably not supported by the GPU cloud business alone. Because more than half of SpaceX's own revenue now comes from its AI division, what decides the success of a large financing is not whether a company is in AI, but how much stable profit and credit its non-AI businesses can supply. Whether the $40 billion is actually backed by Starlink's earnings or the parent's credit cannot be determined until the contract terms are made public.
CoreWeave's CDS briefly topped 1,140 bp as credit markets worry about repayment capacity
A key comparison for assessing the financing environment of companies heavily dependent on AI is neocloud leader CoreWeave. The market's view of its credit risk shows up in the price of its credit default swaps (CDS). A CDS is a financial contract that protects against losses if a default occurs, and a higher premium means the market sees greater credit risk. Rates are quoted in basis points (bp), where 1 bp equals 0.01%.
According to CNBC, CoreWeave's five-year CDS briefly exceeded 1,140 bp on October 15. Data from ICE Data Services reported by The Next Web showed it hit a record 1,194.5 bp that day, a level that prices in a probability of default within five years of more than 60%. Gavan Nolan of S&P Global Market Intelligence generally regards CDS spreads above 1,000 bp as a level indicating heightened credit stress. However, default probabilities implied by CDS depend on assumptions such as recovery rates, so it is not established that the company will actually fail with a probability above 60%.
CoreWeave's debt burden is also heavy. According to CNBC, the company carries more than $51 billion in debt and related obligations, and the yield on its 2031 notes, which carry a 9.75% coupon, has risen to 15.78%. Rising bond yields mean investors are demanding a higher risk premium. In the April-June 2026 quarter, revenue was $2.575 billion, while the net loss was $626 million and net interest expense reached $640 million.
What the market fears is not necessarily a decline in demand for AI computing. Evercore ISI's trading desk told CNBC that the market is no longer accepting AI-related debt that lacks operating cash flow to back it. Nolan also explained that the issue is not the size of demand but whether that demand translates into the cash flow needed to repay borrowings. Even if GPU demand is strong, credit risk rises if capital spending and interest payments swell and actual cash collection cannot keep up.
What credit markets focus on here is whether debt can be repaid from operating cash flow, not necessarily whether a company has businesses outside AI. On this point, Musk's claim and the market's yardstick do not fully match. They do share the idea, however, that a stable earnings base outside AI can diversify the sources of repayment and potentially strengthen creditworthiness.
The problem of GPU cloud businesses being vulnerable to customer contracts is also visible in Japan. At the end of July 2025, Sakura Internet announced large-scale cancellations of GPU usage contracts and twice revised down its forecast for the fiscal year ending March 2026. Although its business scale differs from that of the leading overseas neoclouds, the two share the feature that changes to a major customer's contract can swing GPU-related revenue outlooks significantly. When large capital spending is financed with borrowing, such contract risk bears directly on the assessment of repayment capacity.
Lambda secured an investment-grade GPU-backed loan, showing a path remains for AI-only firms
Still, companies without non-AI earnings have not been completely shut out of large-scale financing. The strongest counterargument to Musk's view is that even if the parent has no revenue from other businesses, long-term contracts with high-credit customers can allow it to raise funds.
For example, according to CNBC, Deutsche Bank analyzes that about two-thirds of CoreWeave's backlog of more than $100 billion comes from investment-grade customers such as Microsoft, Meta and IBM. CoreWeave CEO Michael Intrator also indicated in an internal memo that the backlog comfortably exceeds the company's total debt. The company also raised more than $10 billion in unsecured and convertible bonds in the April-June 2026 quarter. Even with concerns running high in credit markets, it has a record of continuing to raise funds on the strength of customer contracts and business scale.
An even more significant case is GPU cloud provider Lambda. The company announced it had closed a $926 million secured Term Loan B backed by GPUs deployed for investment-grade customers. Terms were set on August 12, 2026, and it was GPU-asset-backed financing using an SPV. Moody's assigned the loan an investment-grade rating of Baa2, and the interest rate was set at SOFR (the Secured Overnight Financing Rate) plus 3.00%.
What matters here is not that Lambda as a company received an investment-grade rating, but that a loan backed by specific assets and contracts was rated investment grade. It shows that even if a borrower's business is concentrated in GPU cloud, a high-credit-quality loan can be put together by properly combining collateral assets with contracts with customers. The basic mechanism is shared with SpaceX's reported financing plan, and it shows that a path to financing remains open for companies with no non-AI business.
However, CoreWeave's CDS and Lambda's GPU-backed loan cannot be compared on identical terms. CoreWeave's CDS mainly reflects the credit risk of the company itself, whereas an SPV loan backed by specific GPUs and customer contracts has a different structure of risk borne by the lender. Moreover, Lambda's $926 million is less than one-fortieth of the $40 billion SpaceX claims. Closing a small secured loan at investment grade does not by itself prove the ability to raise tens of billions of dollars.
Musk's view is therefore hard to support in the sense that all neoclouds will lose access to financing. On the other hand, there remains a large gap between companies that can raise tens of billions of dollars with the backing of a parent's creditworthiness and those that raise funds by pledging individual GPU assets and customer contracts. Lambda's case shows the latter path is open, but whether that mechanism can be scaled up to the $40 billion SpaceX claims has not yet been confirmed.
Uncertainty over SpaceX's credit, too: what would undermine Musk's view
SpaceX is not free of financing concerns either. CNBC reported on October 7 that the company's existing debt was trading at levels close to speculative grade, and Bloomberg reported on October 23 that this financing was struggling. While Musk emphasizes SpaceX's creditworthiness, the actual market assessment and financing terms do not necessarily match his explanation. Rather than a clear boundary between SpaceX and neoclouds, it is more appropriate to think that differences in earnings base, collateral, customer contracts and borrowing size create differences in degree of creditworthiness.
The assessment would also change depending on the structure through which SpaceX completed the $40 billion financing. If it was done as an SPV loan backed by the Anthropic contract and GPU assets, without a parent guarantee, the structure itself is no different from a neocloud's GPU-backed loan. In that case, earnings from Starlink and the rocket launch business were not used, at least directly, as a source of repayment or credit support. The basis for the view that large-scale financing was possible precisely because of non-AI businesses would be weakened.
There are two conditions to watch in testing this view going forward.
First, whether a company whose core business is GPU cloud completes a financing of several billion dollars at an investment-grade rating between October 25, 2026 and year-end. If it turns out that a company can raise funds on favorable terms without non-AI revenue even at several times the $926 million Lambda raised, the view that business diversification determines large-scale financing would be significantly shaken. What matters especially is whether the scale of financing can be expanded on customer contracts and GPU collateral alone.
Whether CoreWeave's CDS falls below 1,000 bp is also one gauge of whether tension in credit markets has eased. However, its CDS was 452 bp in June 2026, so it swings widely with market conditions. A decline in CDS alone cannot be taken to mean that funding capacity has recovered; it must be assessed together with the size, interest rates and collateral terms of loans actually completed.
Second, whether the terms of the $40 billion financing SpaceX claims become public. If future disclosures, such as results for October-December 2026, reveal it was an SPV loan without a parent guarantee, backed mainly by GPU collateral and contracts with AI customers, the explanation that Starlink's earning power directly supported this financing becomes hard to sustain. Conversely, if guarantees or credit support from SpaceX itself were in place, that would support the view that having non-AI businesses was a funding advantage.
For neoclouds, the case that currently offers a concrete option is Lambda's. If loans combining contracts with investment-grade customers and GPU assets expand from the hundreds of millions of dollars to several billion, and even tens of billions, companies with no non-AI business may be able to keep funding large capital investments. One clue to its feasibility will be the size and terms of new borrowing revealed in CoreWeave's next earnings report.
The mass shakeout of the neocloud industry that Musk predicts remains, for now, only a forecast. However, CoreWeave's credit costs show that growing GPU demand alone cannot support enormous capital spending, and that cash flow and creditworthiness backing repayment capacity have become more important. The focus going forward is not only whether demand for AI infrastructure continues, but how many companies remain that can turn that demand into stable revenue and cash recovery.