Barclays sees the six biggest US banks selling $41 billion of senior debt in the fourth quarter

The word "record" depends on the comparison. Barclays puts the $41 billion at 30% above the long-term fourth-quarter average ^[1], so that average is near $31.5 billion ^[10]. Against the third quarter's roughly $50 billion ^[2], the same estimate is about 18% lower ^[11]. The annual figures reconcile: a $233 billion full-year projection less the $192 billion already issued leaves $41 billion ^[12], the same as the fourth-quarter estimate. That match suggests both totals cover the six banks. Crypto Briefing gives a separate projection of $294 billion for US banks overall ^[4].

The 40% rise in year-to-date issuance ^[3] puts last year's comparable total near $137 billion ^[13]. So the banks have sold about $55 billion more debt than at this point a year ago ^[14]. Crypto Briefing ties the borrowing to hyperscaler spending by Alphabet, Amazon, Meta and Microsoft ^[18]. The figures, as reported, do not show how much of the money is lent to AI borrowers.

That link can run three ways. Banks could fund data center loans and hold them, putting AI credit risk on their own balance sheets. They could lend for a few months and then sell the loans on. Or most of the extra funding could have little to do with AI, with the hyperscaler connection arriving only as fees.

The business Crypto Briefing describes fits the second and third readings best. The banks it names earn fees on each bond they underwrite and each loan they syndicate, plus trading revenue on the paper afterwards ^[7]. Underwriting and syndication both end with the debt in other investors' hands. Scale points the same way. Against more than $570 billion of AI-related debt projected for 2026 ^[6], the $233 billion full-year bank total would cover about 41% at most, even if every dollar became an AI loan ^[15].

In my view, the six banks carry AI risk mainly through their fee lines and their own funding costs, and retained loans are the smaller part. The counter-case is that they are holding large bridge or term loans for data center builders, and nothing in the reported figures rules that out. The view is wrong if the lending the banks keep on their books for tech and data center borrowers grows roughly in step with the extra $55 billion of issuance ^[14].

The AI projection also implies a slowdown. If the year-to-date AI total runs through September, the remaining $70 billion to $81 billion ^[16] is at most about half the average quarterly pace of $163 billion to $167 billion so far ^[17].