A closer look at the financing behind the buildout
Federal Reserve minutes released November 18 reveal heightened scrutiny of the financing supporting the artificial-intelligence buildout, including leasing commitments and debt structures that officials said could amplify losses if investment returns disappoint. The account also describes an especially notable widening in credit spreads on high-yield bonds issued to finance data-center construction.
The disclosures concern the Federal Open Market Committee's October 27–28 meeting, not a new November interest-rate decision. They put financing risk alongside the strong construction and investment activity that policymakers continued to observe in the data-center sector.
Almost all meeting participants discussed the risks of AI investment failing to meet investor expectations. Most noted that related debt had expanded rapidly, much of it through complex and opaque financing structures. A few pointed to increasingly concentrated exposures among a small number of firms.
Fed staff specifically highlighted off-balance-sheet leasing commitments and circular financing arrangements among AI-related companies. Their assessment was conditional: those arrangements could amplify losses if returns disappointed, rather than demonstrating that such losses had already occurred.
Strong building activity, more skeptical markets
The market account presented to the committee described a shift during the period between meetings. AI infrastructure initially helped lift equity prices, but investor sentiment weakened after two leading AI companies reported revenue growth below expectations and a hyperscaler disclosed capital-spending plans requiring more financing than investors had anticipated.
Credit spreads widened across investment-grade and high-yield markets, with bonds financing data-center construction experiencing particularly pronounced pressure. The account also attributed a reversal in earlier Treasury-yield increases partly to uncertainty about AI companies' revenue prospects and the sustainability of the data-center building pace, alongside concerns about French sovereign debt.
That financial reassessment had not erased the sector's economic strength. Participants described AI-related investment, manufacturing and data-center construction as particularly strong, in contrast with weak housing and other interest-sensitive sectors. Many also reported that shortages of skilled workers for data-center construction and related industries had produced sizable wage premiums.
Several participants saw a risk that the investment surge could prove excessive and deliver lower-than-expected returns. A couple suggested it might be crowding out other investment. Those were policymakers' risk assessments, not findings that particular developments were unviable.
Property-sector pressure extends beyond data centers
The minutes describe broader strains across real estate. Many participants said higher borrowing costs and more selective lending had weighed somewhat on residential investment and commercial property. Staff reported that real estate investment trust share prices declined over the intermeeting period and that commercial mortgage-backed securities delinquency rates remained elevated.
Mortgage credit availability continued to ease gradually, but elevated interest rates kept origination activity subdued. Residential mortgage activity declined further through September amid sharply higher mortgage rates, underscoring the distinction between access to credit and willingness or ability to borrow at prevailing prices.
The staff nevertheless judged strong bank and household balance sheets to be important buffers. Several participants described the banking system as well capitalized, while others cautioned that supervisory data might not fully capture indirect exposures through nonbank institutions.
The October meeting ended with a 10–2 vote to retain the federal funds target range at 3.75%–4%. The newly published discussion adds detail about property-financing vulnerabilities; it does not enact new lending restrictions or establish a data-center-specific regulatory measure.
