Musk says SpaceX closed $40 billion GPU debt, warns of neocloud shakeout

Elon Musk said SpaceX had closed $40 billion of debt for GPUs in an X post displayed October 26 in Korea. He predicted widespread failures and consolidation among more than 200 neoclouds, saying debt markets had largely closed to companies without substantial revenue outside AI data centers.

For businesses renting GPUs to operate AI services, the issue concerns their suppliers’ financing and continuity. Musk cited Starlink and launch revenue, alongside investment-grade creditworthiness, as SpaceX’s advantages. His post supplied no individual company names, borrowing rate or maturity.

SpaceX had outlined its funding approach at its October 20 earnings webcast. Chief Financial Officer Bret Johnsen described a combination of operating cash flow, debt markets, and arrangements with vendors and customers. At that point, he said the company was talking to available pools of debt capital to support its expansion.

The company reported third-quarter capital expenditure of $19.6 billion. Johnsen projected $35 billion to $40 billion for the fourth quarter. That quarterly spending forecast, made on October 20, covers a different measure and period from the borrowing figure Musk subsequently announced.

The September-quarter regulatory filing records $23.745 billion of construction in progress, primarily servers, networking equipment and data-center infrastructure not yet placed in service. It lists $51.294 billion in cash and equivalents and $38.561 billion in short-term marketable securities. The filing says future capital needs may vary significantly with expansion plans and market conditions. It also describes $25 billion of senior unsecured notes issued in June, maturing from 2031 through 2056. Proceeds repaid a bridge loan and associated costs, with the remainder for general corporate purposes. These are the June notes; Musk separately announced the GPU borrowing.

From power access to revenue

Musk was responding to investor Chamath Palihapitiya, who disclosed owning energized land assets. Palihapitiya criticized differences in neocloud capitalization and organization, alongside slow contracting at hyperscalers. He also said some neoclouds were better prepared than others.

He predicted delays in supplying outside customers as large operators prioritize internal needs. In 2027, he suggested, some facilities could remain idle or earn below their cost of capital. This is the assessment of an investor with property interests. Musk’s shakeout warning is also a forecast.

Expansion announcements are continuing. On October 26, SWI Group launched SWI Compute, saying it had ordered approximately $1 billion of NVIDIA systems and finalized initial customer contracts. The parent has secured 3.6 GW of power capacity across Europe and the United States; its release describes power as contracted and systems as on order.

SWI Compute plans dedicated and on-demand capacity for European AI companies, laboratories and enterprises. Its first large UK facility would start at 30 MW with about 7,000 Rubin GPUs and 3,500 Vera CPUs. The company targets operation by the end of 2027.

Public accounts illustrate the financing burden. Musk did not name individual companies. CoreWeave’s August 11 release reported second-quarter revenue of $2.575 billion, net interest expense of $640 million and a GAAP operating loss of $49 million. These figures cover the quarter ended June 30. Neoclouds specialize in renting GPU-centered computing infrastructure.

Its roughly $104 billion revenue backlog at that date includes remaining performance obligations and other estimated future revenue under committed contracts. CoreWeave says recognition depends on delivery and service availability requirements. Contracted demand therefore retains an infrastructure-delivery condition. The backlog excludes over $25 billion in commitments added early in the third quarter.