Broadcom CEO Hock Tan, former Google CEO Eric Schmidt, and Intel CEO Lip-Bu Tan each identified capital costs as the binding constraint on AI infrastructure expansion
Broadcom, SpaceX and Oracle AI chip financing
Broadcom, SpaceX, Oracle Each Seek Tens of Billions in AI Chip Debt
The scale of these transactions reflects AI hardware costs that exceed what individual companies can fund from their own balance sheets. Morgan Stanley estimates AI infrastructure will require $1.5 trillion in external financing by 2028.
The full picture
Broadcom is organizing more than $50 billion in financing for OpenAI's custom AI chip program, with Apollo and Blackstone in early discussions as lenders and a close targeted before end of 2026. The Information reported the deal hit an $18 billion financing obstacle. The chip program, called Nexus, targets deployment of 10 gigawatts of AI accelerators by late 2029; its first two chip generations are named Jalapeño and Serrano. One report describes the structure as mirroring a $60 billion facility Broadcom previously arranged for Anthropic; another source puts that facility at $42 billion, leaving the two figures unreconciled. In June, Broadcom established a separate platform with Apollo and Blackstone to finance more than 20 gigawatts of capacity for developers including Anthropic and OpenAI through 2028, starting with an initial $35 billion tranche.
SpaceX is seeking $40 billion structured as $10 billion in bank loans and $30 billion in investment-grade bonds to fund a Nvidia chip order. Apollo is expected to lead the deal and help distribute the debt; PIMCO is also in talks. The transaction targets a 2027 close and had not been publicly announced by SpaceX or Apollo as of the reporting.
Oracle is negotiating with Apollo and Goldman Sachs for chip financing tied to a 1-gigawatt data center, aiming to close by year-end. The deal may be structured so that a separate investor-funded company buys the chips, allowing Oracle to pay from cloud revenue without increasing its own debt.
Apollo's involvement in both the Broadcom and SpaceX deals follows its announcement of a $500 billion compute-financing initiative with Nvidia and five other financial institutions. Morgan Stanley estimates AI infrastructure will require $1.5 trillion in external financing by 2028.
How it developed
Apollo appears as a common lender across three AI chip financing deals reported October 14: Broadcom arranging more than $50 billion for a custom OpenAI chip under a program called Nexus with an $18 billion snag reported, SpaceX seeking $40 billion in debt for a large Nvidia order, and Oracle in talks for off-balance-sheet data center financing.
Late-September analysis argues debt-financed non-Nvidia clusters likely incur loan spreads that erase hardware savings, and Sequoia's David Cahn identified a roughly $600 billion gap between hyperscaler AI spending and ecosystem revenue.
Analysis published: non-Nvidia chip clusters likely cost more to finance than they save on hardware, with GPU loan spreads widening significantly at lower credit grades
Jensen Huang tells CNBC a 1-gigawatt Nvidia AI factory costs $50-60 billion but can generate roughly $50 billion in annual rental revenue
Sources
4 more sources
Related
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free