Banks Offload AI Debt as Investor Appetite Fades

Key points

• JPMorgan, Morgan Stanley, SMBC, and MUFG are using significant risk transfers to move AI-linked debt off their books, hitting internal exposure limits.
• Global AI-related debt issuance dropped to $23 billion in September, less than half of the previous month's volume, following a June peak of $113 billion.
• Oracle's credit rating was cut to BBB- by S&P, and its five-year credit default swaps reached 151 basis points, a level not seen since 2009.
• The 10-year Treasury yield hit 5.304% in October, its highest since 2002, increasing borrowing costs for all sectors.
• Investor demand for AI bonds has thinned, with order books shrinking from five times the amount on sale in February to under two times by July.

Background

In late September, Oracle issued a force majeure notice regarding its New Mexico data center, Project Jupiter, highlighting tensions between AI growth and debt risks. Earlier in the month, AI infrastructure borrowing had already pushed the 10-year Treasury yield to 5.17%, raising costs for data center developers. NVIDIA had recently rebranded as an AI infrastructure platform, signaling a shift in how the industry underwrites large data center demand.

How outlets are covering it

Startup Fortune emphasizes that banks are hedging their bets by offloading debt due to internal limits, citing a six-month struggle to distribute $38 billion in Oracle-related construction debt. The Financial Times highlights a broader slowdown in AI borrowing, noting that global issuance fell sharply in September as investors reassessed risks and faced political opposition to data centers. Paul Krugman argues that the AI boom is crowding out other investment by driving up interest rates, contrasting this with the 1990s tech boom which did not crowd out other sectors due to foreign capital inflows. While Startup Fortune focuses on bank balance sheet mechanics, the Financial Times points to investor sentiment and political pushback, and Krugman frames the issue as a macroeconomic crowding-out effect.

Why it matters

The shift from aggressive underwriting to risk offloading signals a potential inflection point in the AI infrastructure financing cycle. If banks cannot place debt at reasonable terms, it could constrain the capital available for data center construction, potentially slowing the pace of AI deployment. The rising cost of borrowing also affects non-AI sectors, as higher interest rates crowd out other investments, impacting housing, factories, and other commercial real estate deals.

What to watch

Investors will closely monitor upcoming large financing deals, including a $60 billion package for Broadcom and Anthropic and a $40 billion raise by SpaceX for Nvidia chips. The ability of these deals to syndicate to a broad base of investors will test the current market sentiment. Additionally, the resolution of Oracle's legal dispute with Wisconsin over a $7 billion collateral bill will be a key indicator of how regulatory and financial risks are being priced into AI infrastructure projects.