Corporate Debt Costs Rise for AI Infrastructure Builders as Credit Spreads Widen

Credit spreads for the technology companies financing the AI buildout are widening, and analysts expect further expansion later this year — raising the cost of debt for both heavily leveraged neoclouds and the larger hyperscalers that anchor the sector.

A credit spread measures the yield difference between a corporate bond and a Treasury security of comparable maturity. Widening spreads signal that investors perceive greater default risk and demand higher compensation to hold that debt.

Apollo Global Management chief economist Torsten Slok flagged the trend in stark terms. "As we speak, spreads are widening out for hyperscalers and also [credit default swaps are] widening out quite substantially," Slok told CNBC. "CDS for Oracle is at the same level it was at in 2008. At the same time, CDS also for the hyperscalers has gone off the bottom … the trend is certainly not your friend."

UBS head of credit strategy Matthew Mish issued a similar caution in a June 24 note to clients. "We expect US credit spreads to remain broadly rangebound in Q3 before widening in Q4 and decompressing into 2027," he wrote, adding that "credit returns are unlikely to compensate investors for the risk backdrop in [the second half of the year]."

The concern is sharpest for neoclouds — smaller, highly indebted infrastructure operators that power much of the AI computing buildout. Neocloud CoreWeave carries total debt of approximately 739 times its equity, while Applied Digital has a debt-to-equity ratio of 172 and Nebius stands at 131, according to FactSet data.

By contrast, the major hyperscalers carry far smaller relative debt loads. Alphabet's total debt-to-equity ratio stood at approximately 18 as of the end of the June quarter, while Microsoft's was about 30 and Amazon's was 51, as of their most recent reporting periods, FactSet data shows.

Mizuho analyst Vijay Rakesh highlighted the neocloud vulnerability in a recent client note. "We are also hearing some concerns on … neoclouds generating negative free cash flow, widening credit spreads and capital raise concerns," Rakesh wrote.

Even as spreads widen, a web of circular financing within the sector complicates the picture. Nvidia has moved to backstop some of its neocloud customers directly, and neoclouds have structured specialized debt contracts tied to the credit ratings of their hyperscaler customers to access cheaper financing. "CoreWeave did just that [for] their last data center capex," Paul Meeks, head of tech research at Freedom Capital Markets, told CNBC.

Analyst Jay Goldberg at Seaport noted the stakes in a July 15 note. "[Neoclouds] are already struggling to finance investment at the current bargain-basement price of $50B/GW," he wrote. "To this end, Nvidia is now becoming more directly involved in financing."

The Bank for International Settlements warned this month that the same circular arrangements could amplify any downturn. "The race to commit early through debt and circular financing also makes a bust more likely," Phurichai Rungcharoenkitkul wrote for BIS on July 7, with the bank's analysis suggesting overinvestment of 1.5 times the necessary level and warning that "stress in one firm could cascade to others through chains of financial exposures."

Goldman Sachs chief credit strategist Amanda Lynam, writing on July 9, anticipated that "a range of financing markets will ultimately be required to satisfy the funding need," including syndicated credit markets, private markets, joint venture funding structures, and international finance capital.

The broader debt market has already shown signs of indigestion. Bonds issued recently by Nvidia and SpaceX priced at relatively low rates that subsequently fell in the secondary market, while Amazon accepted rates that were high by its own historical standards, according to reporting published July 12.

Dan Alpert, founding partner of Westwood Capital, distilled the concern plainly. "My biggest concern is the credit quality of these companies that find themselves competing," Alpert told CNBC.

With hyperscalers continuing to expand capital expenditure — Google has raised its capex projections for both 2026 and 2027 — how debt markets absorb the next wave of issuance will be a central test for the multi-year AI infrastructure cycle.

Disclaimer