Why SpaceX Finds It Easier to Borrow Money Than CoreWeave Despite Buying the Same GPUs
Through a leasing structure in which entities such as Apollo purchase GPUs and then lease them back to SpaceX itself, SpaceX obtained a $40 billion GPU financing commitment, with equipment expected to be delivered in batches in the first half of 2027, each batch with a lease term of about five years. The article's comparison points out that SpaceX has Starlink operating profit and IPO fundraising to support the group's credit, while CoreWeave has about $50 billion in debt, with some bond yields rising to 15%–16%, and Microsoft accounts for about 67% of its 2025 revenue. Nebius, meanwhile, obtained a $3.5 billion senior secured loan from Citi. The differences in financing costs among the three may affect the pace of their data center construction and their computing power pricing.
Why read this
AI score 68
The article compares the financing structures of three computing power companies, illustrating how non-AI business revenue affects the financing costs of GPU leasing.
Topics
Body · AI translation
[[**SpaceX secures $40 billion in GPU lease financing backed by Starlink profits, IPO cash, and group credit, while yields on some CoreWeave bonds rise to 15%–16%, accelerating the divergence in AI compute financing.**
**Key points**
**1. SpaceX finances $40 billion GPU purchase through a lease structure**
Entities such as Apollo will first purchase the GPUs and then lease them to SpaceX, with deliveries in batches in the first half of 2027, each batch under a lease of about five years and rent prepaid quarterly; Goldman Sachs designed the deal, with BlackRock, KKR, and others participating.
**2. Starlink profits and IPO cash support group credit**
Q3 Superintelligence revenue was $7.56 billion, exceeding the $5.248 billion from the connectivity business, but connectivity business profit of $2.367 billion far exceeded Superintelligence's $431 million; the IPO raised $85.675 billion net, with cash and short-term securities of $89.855 billion.
**3. CoreWeave has major customers but still faces bond repricing**
Its customers include Microsoft, OpenAI, and others, and Nvidia has committed to purchase $6.3 billion of idle compute, but its debt is about $50 billion, yields on some bonds are 15%–16%, and CDS is about 1,200 basis points; Microsoft accounted for about 67% of its 2025 revenue.
**4. Nebius still obtains loans, but ultimately GPU returns will decide**
Nebius obtained a $3.5 billion senior secured loan from Citi, after a previous $775 million financing backed by cash flow from deployed GPUs and investment-grade customer contracts; the interest rate on the SpaceX lease has not been disclosed, and the new GPUs will not be delivered until 2027, so it must prove that revenue covers costs. 2026-11-06 17:40 Buying GPUs the same way, why is it easier for SpaceX to borrow money? Zhixin Quick Overview
This article is from the WeChat public account: Zhixin pro, author: Burke
On October 25, Musk announced on X that SpaceX had obtained $40 billion in GPU financing.
Immediately afterward, he began to worry about his peers' businesses. In Musk's view, companies lacking stable revenue outside AI data centers are finding it increasingly difficult to obtain funds from debt markets, and among the more than 200 Neocloud companies worldwide that specialize in renting out AI compute, a large number may go bankrupt or be acquired in the future.
The next day, SpaceX disclosed the specific arrangements for this financing. An entity managed by Apollo will fund the purchase of GPUs and related equipment, and then lease them to SpaceX for use. The equipment is expected to be delivered in batches in the first half of 2027, with each batch under a lease of about five years. The transaction has currently obtained financing commitments and is expected to complete formal closing before the end of the year.
Musk attributed the financing success to the revenue brought by Starlink and the rocket launch business. In his view, SpaceX does not have to rely entirely on the AI compute business to make money, so it can continue to obtain large-scale funding support when financial institutions tighten lending.
But just a few days earlier, another Neocloud company, Nebius, had also obtained a $3.5 billion bank loan, while CoreWeave was heading in another direction: this company holds long-term orders from tech giants such as Microsoft and OpenAI, yet yields on some of its bonds have risen to 15%–16%. All are buying GPUs, and all have customers, so why are creditors willing to continue providing funds to some companies while beginning to worry about the debt repayment capacity of others?
1. The $40 billion did not go directly into SpaceX's hands
This time, SpaceX did not directly take the $40 billion; instead, it had an entity managed by Apollo first buy the GPUs and servers, and then lease the equipment to itself.
On October 26, SpaceX submitted an 8-K filing to the U.S. Securities and Exchange Commission, disclosing the financing commitments signed two days earlier. An entity managed by Apollo will purchase complete GPU systems and related equipment and deploy them to data centers designated by SpaceX. The equipment is expected to be delivered in batches in the first half of 2027, with each batch corresponding to a lease arrangement with a term of about five years to five years and three months.
Rent is prepaid quarterly. Each payment includes one-twentieth of the financing cost of that batch of equipment, as well as fixed interest calculated on the outstanding balance and other fees. In other words, SpaceX can first obtain computing equipment worth tens of billions of dollars, and then pay principal and interest over the next five years or so.
Apollo was not the only one putting up the money. SpaceX announced that day that Goldman Sachs was responsible for designing the transaction structure, and that BlackRock, Goldman Sachs Asset Management, and KKR-related funds had also made investment commitments. The participants span asset management institutions, private credit, and large investment banks.
This financing method benefits both sides: institutions such as Apollo can collect principal and interest under the lease, while SpaceX does not have to pay the full purchase price at once when the equipment is delivered.
Musk attributed this financing to SpaceX's group credit. The financial report disclosed on October 22 provided an explanation: SpaceX's AI business has become its largest source of revenue, but the main operating profit still comes from the connectivity business in which Starlink sits. According to the company's disclosed third-quarter results, SpaceX's revenue for the quarter reached $14.575 billion. Among that, Superintelligence business revenue was $7.56 billion, already exceeding the $5.248 billion from the connectivity business including Starlink.
The Superintelligence business sells the most, but profit mainly comes from Starlink. In the third quarter, the Superintelligence business contributed $431 million in operating profit, while the connectivity business reached $2.367 billion. SpaceX's AI business has already started making money, but the group's main source of operating profit is still the division that operates satellite communications. Moreover, SpaceX's Superintelligence business also includes products such as Grok, Cursor, and X, so the $7.56 billion in revenue cannot all be regarded as GPU computing power sales revenue.
However, SpaceX itself is also bearing increasingly large capital expenditures. In Q3, the company's capital expenditures reached $19.593 billion, of which $16.444 billion came from the Superintelligence business. Chief Financial Officer Bret Johnsen also expects fourth-quarter capital expenditures to reach $35 billion to $40 billion. This scale of investment is far higher than the group's operating profit for the quarter.
SpaceX also has a funding advantage. In June this year, the company completed its IPO, raising approximately $85.675 billion net. By the end of September, its cash and short-term marketable securities totaled $89.855 billion. Starlink provides operating profit, while the IPO supplemented SpaceX with a large amount of equity funding. These conditions help the company bear long-term computing power investment and also give it more options when negotiating with financial institutions.
But secured financing does not mean secured investment returns. SpaceX still needs to wait until equipment delivery in 2027 before it can gradually put the new computing power into use. Moreover, having non-AI businesses is not a necessary condition for obtaining financing. Just three days before Musk's remarks, another company that specializes in renting out AI computing power, Nebius, also announced that it had obtained a $3.5 billion bank loan facility.
II. CoreWeave has major customers, but its bonds fell
Compared with SpaceX, CoreWeave's business is much simpler. It rents computing resources to AI companies; the more customers need to train and run models, the more GPUs it needs to buy and the more data centers it needs to build.
The problem is that customers signing orders does not mean CoreWeave has already received enough money to pay for these investments. On October 15, Reuters reviewed CoreWeave's customer contracts and financing situation. Its customer list includes Microsoft, OpenAI, Meta, and Anthropic. Nvidia not only holds shares in it but has also committed to purchasing about $6.3 billion worth of idle computing power by 2032.
But these contracts failed to stop bond prices from falling. According to LSEG data, CoreWeave's debt at the time was about $50 billion. Yields on some of its bonds rose to 15%–16%, and credit default swap prices reached about 1,200 basis points.
Rising bond yields do not mean that all of CoreWeave's borrowing must be repriced at that rate. Fixed-rate debt already signed is still executed under the original contracts. But it shows that if the company needs to issue new debt under similar conditions, its financing costs could be much higher than before.
CoreWeave has not lacked money in recent years. In 2023, the company obtained $2.3 billion in financing backed by GPUs; in 2024 it obtained another $7.5 billion in financing. After that, private credit institutions such as Blackstone and Magnetar and several banks continued to participate. In March 2026, the company added an $8.5 billion financing.
This money was used to purchase GPUs and build computing facilities. CoreWeave also secured a large number of long-term customer contracts, creating a mutually reinforcing relationship: major customers commit to buying computing power, the company uses the contracts to raise financing, and then uses the financing to build the data centers customers need. When customer demand grows, this approach can support rapid expansion.
CoreWeave needs to purchase chips, build computer rooms, and connect power first, while customers pay computing power fees according to contract schedules. There is a time gap between equipment investment and customer payments. Once a project is delayed, the equipment may already have arrived while the loan has begun accruing interest. Even if the customer has not canceled the order, the company still has to bear the financing cost during the delay.
Customer concentration also amplifies this risk. According to company documents cited by Reuters, Microsoft contributed about 67% of CoreWeave's 2025 revenue, up from 62% in 2024. Microsoft is a customer capable of bringing large, long-term demand, but it also means that a large part of CoreWeave's revenue depends on the procurement and delivery arrangements of the same company.
However, the fall in CoreWeave's bond prices does not mean that all Neocloud companies are shut out of the debt market. Just before Musk warned that industry financing was tightening, Nebius had just announced a new bank loan, saying it had obtained a $3.5 billion senior secured loan from Citi for GPU procurement, data center construction, and other operating expenses.
Three months earlier, Nebius had also obtained about $775 million in financing, maturing in October 2030, at an interest rate of SOFR plus 2.50 percentage points. This loan was supported by deployed GPUs and cash flow from investment-grade customer contracts. The company said the loan and contract receivables can cover the full capital expenditures of the corresponding projects. The same company used existing equipment and customer contracts to obtain multi-year financing, and also used a one-year loan to support new projects.
As James Martin, senior US credit strategist at SMBC, told Reuters, investors are more willing to finance data centers that are close to completion and have few obstacles to going into operation.
III. Being able to afford GPUs also means being able to sustain them
For the same batch of GPUs, the higher the financing cost, the more money the company needs to recover from customers.
Buying chips, building computer rooms, and connecting to power all require upfront investment. Once the equipment begins to be leased out, the company also has to pay for operations and maintenance, depreciation, and loan interest. Even if the GPU purchase price is the same, companies with lower financing costs have an opportunity to offer more competitive leasing quotes.
But how much computing power customers use and whether the computer rooms can be put into operation on schedule also affect returns. SpaceX has already secured $40 billion in financing commitments, and the next step is to turn these GPUs into revenue.
On October 26, SpaceX said the newly obtained financing will be used to deploy next-generation NVIDIA Vera Rubin chips to provide computing power for Grok and future cloud service customers. Chief Financial Officer Bret Johnsen also revealed that the payback period for some of the company's recent superintelligence infrastructure investments is less than one year.
However, Johnsen was referring to some recent infrastructure investments, and SpaceX has not disclosed the expected utilization rate, revenue, and payback time of the equipment corresponding to this $40 billion financing. The relatively fast payback of existing infrastructure projects does not mean that the new batch of GPUs delivered in 2027 will be able to achieve the same returns.
In particular, SpaceX's AI investment is still increasing rapidly. In the third quarter of this year, the superintelligence business achieved quarterly operating profit for the first time, but it still had a cumulative loss of $3.295 billion in the first three quarters. During the same period, capital expenditure for this business reached $39.995 billion, while revenue was only $10.939 billion.
SpaceX has Starlink's profits and ample cash reserves to support this construction period. But as the scale of investment expands, it still needs to prove that after more GPUs are put into operation, the revenue generated will be enough to cover equipment, operations, and financing costs.
SpaceX can use the group's other businesses and cash reserves to bridge this gap, while independent Neoclouds often need to continue borrowing.
On October 27, Reuters reported that Neocloud company Argentum AI, backed by Super Micro, was in talks for about $5 billion in GPU financing. The chips are planned for a data center in Poland, and the computing power will be leased to customers without investment-grade credit ratings.
The transaction is still under negotiation and financing has not yet been completed. But even if some Neocloud companies face pressure in the bond market, there are still financial institutions willing to study new GPU financing projects.
On the same day, Moody's said AI infrastructure is driving the expansion of the private credit market, but related financing also increases refinancing risk. In the coming years, as some private credit loans mature one after another, borrowers with weaker credit quality may find it difficult to obtain new funds on the original terms.
This risk is especially important for companies that need to continuously build data centers. There is a time lag between chip procurement, computer room construction, and customer payments, and when some loans mature, new GPUs may not yet have generated enough operating cash flow. The company then needs to rely on subsequent financing to repay old debt.
SpaceX has already used the group's credit to sign a five-year equipment lease, while Nebius uses both multi-year asset financing and short-term loans. CoreWeave has long-term customer contracts, but it still needs to deal with the bond market repricing its existing debt.
All three companies are expanding computing power supply, and how much it will cost them to obtain funds next may affect the speed at which they build new data centers and the computing power prices they can ultimately offer customers.
Final thoughts
Musk secured a $40 billion GPU financing commitment for SpaceX, while predicting that the Neocloud industry will see a large number of bankruptcies and acquisitions. His concerns are not unfounded: yields on some CoreWeave bonds have risen to 15%–16%, and investors have begun demanding higher risk compensation.
But financial institutions have not stopped providing funds for AI computing power. This year Nebius obtained both multi-year financing backed by existing GPUs and customer contracts and a $3.5 billion short-term secured loan. CoreWeave has large orders from technology companies, yet bond investors have begun demanding higher returns. SpaceX has Starlink's operating profit and cash reserves, but the final interest rate on the $40 billion financing lease has not yet been announced, and the new GPUs will not be delivered until 2027.
Several companies are raising funds for GPUs in advance, but they need to repay the money on different schedules. SpaceX has already found a financing channel for the next batch of GPUs, and next it still has to prove whether these devices can earn back their costs over the coming five-year lease term. AI industry signal channel: Frontier Technology. This content is published with the author's authorization. The views represent only the author and do not represent Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]