A $60 billion financing deal for AI chips is said to have entered the syndication distribution phase. According to media reports citing informed sources, Bank of America, Citibank, and Morgan Stanley have begun to transfer part of the debt to other banks. This is the largest chip financing transaction to date and marks a crucial step for banks in shifting from "committing to funding" to "distributing risk."

It is reported that a senior secured loan of approximately $42 billion, guaranteed by Broadcom, has already commenced syndicated distribution. With Broadcom's A-rated credit rating, this portion of the debt may in the future enter the private placement or investment-grade bond market. Additionally, there is another $18 billion in subordinated debt not guaranteed by Broadcom, which will be launched later, and Blackstone has committed to subscribing for about $9 billion of it.

This funds will be used to support Anthropic in leasing Google's TPU chips, corresponding to their chip orders for 2027. The lease payments will begin only after the chips are delivered. Reports indicate that, as a barometer of AI's debt risk preferences, the distribution progress and final pricing of this transaction will directly reflect institutional investors' true valuation of the long-term demand for AI's computing power.

As investors' doubts about whether huge capital expenditures by tech companies can be converted into long-term profits increase, banks launching this distribution are essentially shifting the highly concentrated infrastructure risks of AI to the market. Senior bonds are purchased with confidence in Broadcom's credit, while subordinated bonds represent a bet on the survival probability of Anthropic.

Layered design: Senior debt relies on Broadcom's credit, while subordinated debt bets on Anthropic

The financing structure exhibits a clear risk stratification. Approximately $42 billion in senior secured loans are guaranteed by Broadcom, which, with its A-rated credit rating, has the potential to enter the private placement or investment-grade bond markets in the future, targeting a broader range of institutional investors. The $18 billion in subordinated debt, however, is not guaranteed by Broadcom, and the risk exposure is entirely dependent on Anthropic's own cash flow and ability to fulfill its obligations.

Blackstone has committed to subscribing for approximately $9 billion of the subordinated debt and will participate in the distribution of the remaining portion.

Insiders say that since subordinated debt investors will directly face the credit risk of Anthropic, banks may choose to enter the market later this year after IPO is completed in Anthropic. By then, potential investors will be able to access its financial disclosures. The timing and pricing of the subordinated debt distribution itself become a direct test of the market's credit pricing for Anthropic.

Convertible Bond Terms: Broadcom Moves from Guarantor to Potential Shareholder

A key detail in the transaction structure is that, according to Broadcom's latest quarterly report, Anthropic can issue up to $42 billion in convertible notes to Broadcom for the payment of lease payments. This clause is also included in the list of terms presented to investors.

This arrangement has transformed Broadcom from a mere chip supplier and guarantor into a potential equity holder of Anthropic. If Anthropic is successful, Broadcom can benefit from the increased valuation through share conversions; if it fails, Broadcom's exposure will further expand from guaranteed debt to equity losses. The report also states that the custom TPU jointly developed by Broadcom and Google is also becoming a force that challenges NVIDIA's dominant position in the AI chip sector through this transaction.

Rating Review: Credit Warnings on Off-Balance-Sheet Guarantees

As banks embarked on their distribution efforts, rating agencies had already issued warnings regarding Broadcom's large-scale off-balance-sheet guarantees.

The report mentioned that Standard & Poor's classified the residual value support provided by Broadcom as a contingent debt obligation and included it in the adjusted debt calculation; Moody's warned that Broadcom might be obligated to increase this, which could limit its financial flexibility. Also in August, credit analysts at Bank of America estimated that Broadcom's maximum remaining value guarantee exposure on the AI financing platform could reach $370 billion, with a theoretical maximum loss of up to $42 billion under extreme stress tests.

These warnings point to the same issue: in order to reduce the borrowing costs for customers such as Anthropic, Broadcom has assumed obligations that far exceed its balance sheet capacity. The $60 billion financing round is the latest and also the largest link in this chain of risk transfer.

Distribution Pricing: A Benchmark for the Trillion-Dollar AI Financing Experiment

This $60 billion financing is not an isolated case.

It was reported that nearly four months ago, in June of this year, Broadcom just completed a $35 billion deal with Apollo and Blackstone, and announced the creation of a 20-gigawatt “AI XPV” platform to provide computing power for customers such as Anthropic and OpenAI. The combination of these two deals constitutes a chip financing experiment worth nearly $100 billion.

The report indicates that the final spread on these distributions will have a demonstrative effect: if the $42 billion in senior secured loans can be distributed at a spread close to that of investment-grade bonds, it suggests that the market accepts Broadcom's guarantee structure; if the $18 billion in subordinated debt requires a significant discount or higher interest rates to attract investors, it means that the market's credit pricing for AI startups is tightening. Banks have chosen to initiate these distributions at this time, which coincides with the period when there are the greatest doubts about the return on capital expenditure for AI.

Subsequent focus will be on the distribution and pricing of subordinated debt, as well as whether Anthropic can provide a benchmark for credit pricing within IPO.