Major Wall Street banks have begun the process of selling off portions of a $60 billion (approx. 81.5 trillion won) loan provided for Anthropic's Google Tensor Processing Unit (TPU) lease to other investors, the Financial Times (FT) reported on the 5th (local time), citing multiple sources.

The transaction, the largest of its kind in semiconductor procurement financing history, is considered a test bed for measuring market demand for artificial intelligence (AI) debt.

Investors have demanded higher risk premiums in recent months amid concerns that the massive capital expenditures (capex) by AI companies may not translate into long-term profits, the FT reported.

According to the report, Bank of America (BofA), Citigroup, and Morgan Stanley, which agreed to provide the loans, started the process of selling parts of their loans to other investors on this day.

The assets put up for sale this time are approximately $42 billion (approx. 57.1 trillion won) in senior secured loans guaranteed by Broadcom.

Thanks to Broadcom's credit rating (A-), the loans could potentially be sold to a broader investor base in markets such as the investment-grade bond market in the future, sources said.

The $18 billion junior debt, which lacks Broadcom's guarantee, is scheduled to be rolled out later.

Private equity firm Blackstone has already agreed to purchase about $9 billion of the $18 billion.

One source said that the junior debt, as it remains directly exposed to Anthropic's credit risk, is likely to be launched after an initial public offering (IPO) later this year so that investors can review financial disclosures.

The funds lent by the three banks—BofA, Citi, and Morgan Stanley—will be used by Anthropic to cover the costs of leasing Google TPUs.

Broadcom co-develops TPUs with Google.

In June, Broadcom also signed a $35 billion financing agreement with private equity firms Apollo and Blackstone, launching the "AI XPV" platform to supply more than 20 gigawatts (GW) of AI computing capacity by 2028.

(Photo: AP, Yonhap News)