AI borrowing slows as investors grow wary of debt binge

The amount of new debt that tech companies raised to help fund costly artificial intelligence developments fell sharply last month as investors reassessed the industry’s risks following months of rampant borrowing.

AI-related debt issuance globally dropped to $23bn in September, nearly half the amount raised in the previous month, according to data compiled by Morgan Stanley, which covers both public bonds and private placement transactions. New financing has steadily fallen since the market hit a peak in June, when companies raised a record $113bn to help pay for infrastructure critical to the technology, including data centres and chips.

In the US investment-grade market, where blue-chip tech groups borrowed around $306bn between January and August, AI-related bond issuance ground to a halt last month.

Morgan Stanley said the decline in new issuance was primarily caused by the fact that so much debt had already been raised earlier in the year. But there are also growing concerns among investors over whether such heavy investments can pay off in the long run, while data centre construction faces increasing political opposition.

The pullback comes at a delicate time for the industry as investors evaluate how much exposure they already have to AI and scrutinise some companies’ growing debt loads and uncertain capital expenditure needs.

“Actual day-to-day financing is getting a little bit trickier,” John Aylward, the founder of London-based credit specialist Sona Asset Management, said at the FT and Latham & Watkins’ Private Capital Summit in London this week. “Everybody’s being tapped and everybody’s getting a bit tighter.”

The pace of AI-driven debt issuance has rocketed this year. An estimated $466bn of AI-linked debt has been raised by companies in 2026 so far, according to Morgan Stanley, up from $101bn for the same period last year. The volume of new debt supply has allowed credit investors to demand higher yields to finance certain projects, which has in turn driven up hyperscalers’ borrowing costs.

The shift in investor sentiment is also set to test upcoming blockbuster deals to fund the purchase of advanced AI chips. Wall Street banks are assembling a $60bn financing package for Broadcom and Anthropic, while SpaceX is in talks to raise $40bn to purchase Nvidia chips, the FT has reported. Both deals are expected to be syndicated to a broad base of investors in the coming months.

Even as the world’s biggest tech companies have clamoured to raise additional funds, data centres across the US have begun to run up against local opposition. Residents are concerned about how the energy-intensive sites will affect local water supply and air quality.

Oracle’s massive data centre campus in New Mexico, called “Project Jupiter”, has become one flashpoint. About $18bn of loans tied to the project are under further strain after Oracle issued a force majeure notice when the site struggled to gain access to electricity. The loans were privately quoted at around 85 cents on the dollar in recent days, people familiar with the transaction said.

“Are we going to learn quickly enough from those mistakes?” Jordan Rieger, managing principal and head of European credit at Monarch Alternative Capital, said, referring to broader risks in data centre financing.

He added that the recent slowdown in new financings could be a welcome opportunity for debt investors to reassess credit risk after they had been inundated by new issuance in recent months. “It’s a healthy breathing opportunity to say: have we structured these facilities right?”

Rob Dafforn, the chief investment officer of distressed debt specialist Polus Capital, said the rampant “speculation” and “very highly levered” deals in the sector mean some investors are already paying more than they should.

“The cost of financing some of these transactions is high, and I’m not sure with this uncertainty over timing [and] pay-off profiles where the value is really accruing in the system,” Dafforn said.

Distressed-debt funds like Polus have largely stayed on the sidelines of the AI trade as company valuations have soared. While Dafforn’s firm has begun to map out the web of AI-related financing transactions that have already been agreed, he has not yet made any investments.

The amount of debt raised for the AI industry has little precedent in modern history as the sheer volume has transformed how both other companies and governments issue debt. However, the speed at which the relatively nascent technology has taken over credit markets has also raised questions.

“Often we find ourselves in new sectors or new industries laying the train tracks as we’re driving the train,” said Rieger. “Hopefully it doesn’t go off the rails before it’s too late.”

Additional reporting by Ramsay Hodgson