News from the IT Home on October 8th: The funding needs for AI infrastructure construction are driving a series of large-scale private debt financing transactions.

According to reports by The Wall Street Journal and Bloomberg today, Broadcom has been arranging financing of over $50 billion (Note from the editor: current exchange rate is approximately 335.597 billion yuan) for the custom AI chips of OpenAI in recent weeks, and has already engaged in preliminary discussions with private credit institutions such as Apollo Global Management and Blackstone Group. Sources familiar with the matter said that the negotiations are still in their initial stages, and the scale of the financing may change.

Broadcom and OpenAI previously announced a collaboration to develop 10-gigawatt custom AI chip systems, with plans to deploy them from the second half of 2026 until the end of 2029. OpenAI internally refers to this chip project as “Nexus”, with the first and second generation chips codenamed Jalape ño and Serrano respectively. According to people familiar with the matter, this financing may cover several gigawatts of OpenAI chip production capacity, with completion expected before the end of this year. Representatives from Broadcom and OpenAI both declined to comment.

Bloomberg also reported on the same day that Broadcom has held preliminary discussions regarding financing for the OpenAI chip project. Early consultations indicate that both parties may seek to raise approximately $30 billion (about 201.358 billion yuan at current exchange rates) in debt, but no formal process has yet been initiated, and the plans may still change. This new financing will build upon a large-scale debt financing that Broadcom previously arranged for Anthropic.

According to Bloomberg, Broadcom is raising $60 billion (approximately 402.716 billion yuan at current exchange rates) in new AI chip financing for Anthropic and other companies. The banks participating in the transaction have begun issuing syndicate distribution letters for $42 billion (about 281.901 billion yuan at current exchange rates) in Class A preferred secured loans. Blackstone Group is leading the $18 billion (about 120.815 billion yuan at current exchange rates) in Class B subordinated debt portion and has committed to contributing $9 billion (about 60.407 billion yuan) from its various funds.

Meanwhile, Oracle is in talks with Apollo and Goldman Sachs for a specialized financing arrangement to purchase chips for a 1-gigawatt data center. Sources say Oracle hopes to complete this transaction within this year and is still discussing it with multiple potential financing partners. The financing structure may involve private investors contributing to establish an off-balance-sheet entity that will purchase the chips and then lease them to Oracle for use. This arrangement is aimed at helping Oracle bridge the funding gap between the cost of hardware purchases and the revenue generated by its cloud computing business, while also avoiding the company from taking on more debt, thereby allowing it to control its debt costs when competing with more financially robust tech giants. It is not yet clear how many chips Oracle plans to finance and purchase, but the cost of buying NVIDIA chips for a 1-gigawatt data center could reach several billion dollars.

The financing activities of SpaceX were first reported by The Financial Times. Sources familiar with the matter said that SpaceX is in talks with lenders led by Apollo for a financing deal of approximately $40 billion (about 268.477 billion yuan at current exchange rates) to purchase NVIDIA chips. Of this amount, about $10 billion (about 67.119 billion yuan) will come from bank loans, and $30 billion (about 201.358 billion yuan) will be in the form of investment-grade bonds. Bond fund Pimco participated in the discussions among the relevant lenders. The transaction is expected to be completed in 2027, rather than within this year. Both Apollo and Pimco declined to comment, while SpaceX and NVIDIA did not immediately respond to media inquiries.

SpaceX CEO Elon Musk previously stated during the company's financial report conference call in August that the company decided to build everything entirely on NVIDIA technology, considering its Vera Rubin architecture to be the best choice.