A new Deutsche Bank analysis finds AI borrowing costs mostly manageable, but some issuers are starting to show strain
Deutsche Bank has charted yield against duration across the US corporate bond market. The picture it found in AI credit is not entirely comfortable.
A thematic research note dated October 15, 2026 concluded that financing costs tied to AI remain manageable overall. Some corporate borrowers, however, are beginning to show signs of financial distress, flagged by widening spreads and high-risk borrowing.
The note plotted yield against duration for both US investment-grade and high-yield bonds.
Yield tells you what a company pays to borrow. Duration tells you how sensitive that debt is to interest rate moves over time.
The split inside the AI trade is stark. AA-rated AI names such as Meta and Nvidia were borrowing at 5–5.5%, according to the analysis.
BBB-rated Oracle , by contrast, showed yields of 8–8.5% on debt with 10 to 12 years of maturity.
The research note followed client meetings Deutsche Bank held on October 9–10, 2026. George Saravelos, the bank’s strategist, came away with a view that cut against the prevailing mood.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
At those meetings, sentiment toward bonds had turned bearish. The concern was that the sheer volume of AI-related debt being created would push long-term yields higher.
Saravelos warned that this bearish outlook had overshot reality. In his view, investors were significantly underpricing the risk of a disruption inside the AI ecosystem itself.
He cited examples such as a cybersecurity breach or a revenue miss. Either could trigger a rush toward safe-haven assets like US Treasuries, the opposite of what bond bears were positioning for.
Saravelos also identified disruptions in the AI landscape as a significant systemic risk for 2027.
Market conditions in mid-October 2026 suggest some investors were already leaning his way. Safe-haven buying picked up, and 10-year Treasury yields declined significantly amid anxiety over AI spending.
There was also market speculation about a possible suspension of 20-year Treasury issuance. The idea would be to ease supply pressure on long-term bonds.
The first takeaway is that AI credit is not one trade. The roughly 3 percentage point gap between the AA names and Oracle’s longer-dated debt is the market telling you where it sees risk.
The second takeaway concerns Treasuries. Saravelos argues that the bigger, underpriced scenario is a shock that sends money flooding into government debt. A cybersecurity incident or a disappointing revenue print from a major AI player could be enough to change the direction of the trade quickly.
What to watch next: spread movements among lower-rated AI borrowers, the direction of 10-year Treasury yields, and any official word on the future of 20-year issuance.