Financial system liquidity is generally stable, but cracks are emerging as we head into year-end. In the Treasury repo market, for example, the daily effective Fed funds rate spiked negative, below the Secured Overnight Financing Rate (SOFR) (Slide 1), suggesting that collateral supply tightened relative to cash supply. On the other hand, some of the change in the spread could be seasonal; it is certainly not as severe as the negative spike in the spread at this time last year. In addition, primary dealers report an upward trend in fails to receive and deliver (Slide 2) in Treasurys over the last year. There also seems to be more stress in the bank treasury world. On paper, for example, banks have more deposits than they can deploy in loans, even as loan growth continues to chug along. Yet borrowings are up over the last year to levels last seen after the regional bank failures in 2023 and early 2024 (Slide 3), suggesting liquidity on bank balance sheets is not as abundant as it seems.
A key question for bank treasurers is where interest rates are headed and, if they rise, how much. For example, r*, the theoretical real, risk-free overnight interest rate, has trended higher over the past 2 years (Slide 4). On the bright side, bank treasurers can earn over 500 basis points on a jumbo mortgage loan compared with the average rate they pay for a 1-year CD, the highest spread they have seen in a decade (Slide 5).
Artificial intelligence (AI) remains top of mind for bank treasurers, and for good reason. According to a 2026 McKinsey survey, companies reported that their workers used AI in at least one business function over the past year (Slide 6), with banks leading the way. For example, McKinsey’s survey found that banks are among the top users of AI coding tools to build in-house functionality without buying outside vendor software (Slide 7). However, this year’s survey respondents at banks and insurance companies did not report a material improvement in the time saved using AI compared with last year's survey (Slide 8). The survey also found that the use of generative AI at work may be peaking (Slide 9).
Data centers, which support AI’s vast computing power needs, were a flashpoint in the Midterms, no more so than in Virginia, where the number of data centers exceeds even Texas, including those already in operation and those planned or under construction. According to Pew Research, published this month (Slide 10), Virginia had 398 data centers in operation, compared with Texas, which has 296. But Virginia also has another 287 data centers in the works, compared with Texas, which has only 170 in the planning stage or still under construction.
Collateral Supply in Treasury Repo Tightens
Fails to Receive/Deliver Trend Higher
Banks Tap Their Wholesale Funding Sources
R* Rising
Good Time To Buy MBS If You Are A Bank
AI Adoption Rate Climbs Higher
Businesses Can Do Their Own Coding Now
Gen AI Saves Time
Gen AI Adoption Rate Flattened
AI Data Center Plans To Expand
The Bank Treasury Newsletter is an independent publication that welcomes comments, suggestions, and constructive criticisms from our readers in lieu of payment. Please refer this letter to members of your staff or your peers who would benefit from receiving it, and if you haven’t yet, subscribe here.
Copyright 2026, The Bank Treasury Newsletter, All Rights Reserved.
Ethan M. Heisler, CFA
Editor-in-Chief

