All About AI?: Goldman vs Citigroup

January 19, 2026 | Updated | This is the 800th post since March 2017, when we resumed publication of The Institutional Risk Analyst. The IRA was created in 2003 by Dennis Santiago and Christopher Whalen at Lord Whalen LLC (dba “Institutional Risk Analytics”). The original blog engine for The IRA was a custom application in MSFT SQL built by Dennis and hosted on a cluster of terrestrial servers in back of the offices in Hawthorne, CA, next to the wet bench.

Today we ponder the latest results from two banking giants, Goldman Sachs (GS) and Citigroup (C), both of which outperformed the rest of the industry in stock price appreciation during 2025. Citi was the best performer among the top-five money center depositories. GS was the leader among larger banks generally, but Lending Club (LC) passed them both in the second half of 2025.

Both of these banks are in the midst of sweeping reorganizations, something that seems to define normal for Goldman and Citi. We continue to believe that merging these two global universal banks would make a lot of sense. Let the Citi bankers run the commercial bank and the Goldman investment bankers run the broker-dealer. Add together the deposits, consumer finance book and investment assets, and you have a dominant financial institution and a real competitor to JPMorgan (JPM). GS has the better market value so their shareholders would get the biggest slice of the post-close equity pie.

In Q4 Goldman Sachs beat Street profit expectations, driven by a surge in investment banking and trading revenue, despite a modest revenue hit from the Apple Card portfolio transition. In fact, there was actually a $2 billion benefit from the deal from release of reserves, this in addition to the considerable operational benefit of ending the AAPL relationship. GS blew past earnings estimates by more than 20%, illustrating that good things happen when the firm focuses on its core strength, namely doing deals and asset gathering.

GS has the least impressive corporate disclosure of any large bank, especially compared to Citi which actually provides copies of its supplement in MS Excel. The amount of information provided by GS is adequate in a legal sense, but nothing like the heaps of data that comes from Citi or the other top-five money center banks.  The table below shows total assets and LTM stock performance for the top five money center depositories plus GS.

Source: FFIEC, Google Finance

Of note, U.S. Bancorp (USB) is the smallest money center bank, not the largest regional, contrary to what you may read in the media. This is due to its huge payments platform and equally large custodial business. USB just announced the acquisition of BTIG, one of the top investment banking and research shops covering financials.

Prior to BTIG, USB acquired MUFG Union Bank's core banking franchise in late 2022, including over a million customers and three hundred retail branches in California, Washington and Oregon. The Union Bank transaction made USB the second largest national player in residential mortgages after JPM and they are an important custodian in the secondary mortgage market. 

Citigroup's top headline was a strong beat on profit expectations, driven by a rebound in investment banking and robust performance in its services and wealth management divisions. But non-interest revenue was almost cut in half in Q4, as shown in the table below from the Citi presentation.

Higher expenses and a $1.2 billion hit from selling its Russian operations to Renaissance Group combined with ongoing challenges in areas like credit cards where delinquency is rising. A big drop in principal transactions did not help nor did the large impairment charge or the $1 billion charge for a change in recognition for foreign currency exposures. Net revenue was off 10% sequentially in Q4 2025. Citi is in a very complex business.

Regarding exposures to Nonbank Financial Institutions (NBFIs), Citi CFO Mark Mason opined on the risk. Most banks talk about NBFI exposures being “investment grade,” but we are skeptical of such claims for the reasons we outlined last week ("Does Private Credit Hurt Bank Stocks?"):

“Overall, I would say the NBFI exposure is predominantly investment grade. So that’s a consistent theme for us as firm, certainly is the case as it relates to how it’s reflected in this disclosure. That means we’re working with top-tier asset managers that are sponsors of private credit or established consumer platforms. We’re maintaining collateral pools that are well diversified with concentration limits. We’re ensuring that there are structural protections, including ample subordination that helps to result in the high investment-grade attachment point. And we’re monitoring all the underlying collateral, and we have transparency at the loan-by-loan level. And so, when I kind of take a step back and look at that, we’re very selective from a risk perspective as to how we play across all of these subcategories, but particularly as it relates to private credit. And I think the key takeaway is that, that category is very broad.”

GS had a remarkable year in 2025, as shown in the table below. It is good to see the firm back on track after the disastrous adventure in retail banking and credit cards.  The table below is really one of the few pages in the GS public disclosure that has any useful information, which is a major reason that the Fed's Y-9C is must reading for analysts of Goldman Sachs. Compare the GS disclosure to Citi and the conclusion is obvious.

Unfortunately, the leadership team at GS still does not seem to have any solid ideas about the future except falling back on the traditional business lines such as investment banking and capital markets, with a secondary emphasis on asset management. Perhaps more concerning is the focus on “AI” as the leading value driver for the future.

When corporate managers start waffling about “AI” you know that they have nothing else to say.  The folks at GS led by David Solomon are really smart people, but "AI" is their battle cry for the future of Goldman Sachs 3.0? Really?

Based on our work with a number of large consumer lenders and vendors, we see AI as mostly a throw-away tool to enhance some consumer interactions, but within very strict limits. The false and spurious results from AI are simply too significant to support more comprehensive uses in customer facing applications. Internal use cases for AI offer far better risk-adjusted returns for expense control, but little in the way of obvious revenue opportunities.

"Everyone knows that AI still makes mistakes," writes Edd Gent in IEEE Spectrum. "But a more pernicious problem may be flaws in how it reaches conclusions. As generative AI is increasingly used as an assistant rather than just a tool, two new studies suggest that how models reason could have serious implications in critical areas like health care, law, and education."

Naturally Mike Mayo of Wells Fargo asked about AI during the GS earnings call:

“This is a new era for Goldman Sachs, Goldman Sachs 3.0, and you’re redesigning the whole firm around AI, so that could be very exciting, but I’m looking for the output that you’re looking for from this. I know it’s early days, but whenever I ask about AI, it’s always answers at the 10,000-foot level. It’s transformational. It’s a game changer. It’s a superpower. We all get that, but what are you hoping to achieve? So this decade, your revenues are up two-thirds. Your headcount’s up one-fourth, so that’s one way maybe you could frame the output that you’d like to achieve, but how much more in revenues? How much more in efficiency? Just can you put some meat on the bones? Thank you.” 

GS CEO David Solomon responded and illustrated how little actual substance there is in future claims about the value creation potential of AI:

“I appreciate the question, Mike, and I appreciate the way you frame it, and I understand why there’s a strong desire to get more from us. What I promise you is you’re going to get more over time as we’re in a position to give you metrics, to give you targets, and to really explain it. I want to step back at a high level. Just the one thing that I’d say, and I’d frame it slightly differently than you’d frame it. This is not a new era for Goldman Sachs, One GS 3.0. We’re not going to transform the whole firm with AI. We are focused on our two core businesses, driving growth in our two core businesses. And both, I think, we’re incredibly well-positioned and positioned to win. AI in this technology is an opportunity for us to drive productivity and efficiency in the organization, and we are very, very focused on it because it will add to our capacity to invest in growth in the business.”

As you can see, the conversation around AI at GS is mostly around expense control, a common theme shared by GS and Citi. When it comes to AI driving revenue growth, the corporate happy talk at GS fails. At Citi, by contrast, the conversation remains focused on reducing expenses and corporate divestitures. This includes an agreement to sell the consumer business in Poland, selling Citi’s remaining operations in Russia, and the sale of a 25% stake of Banamex to one of Mexico’s most prominent investors. 

The graphic below shows the expenses of Citi in detail. CEO Jane Fraser mentioned AI six times in her Q4 comments, but provided explicit guidance about operating leverage: "We expect our disciplined expense management, combined with top-line revenue momentum, will drive another year of positive operating leverage as we target an efficiency ratio of around 60% for the full year." Getting efficiency to 60% implies significant headcount reductions in 2026.

At a little over 1.1x book value and $210 billion in market cap, Citi arguably has more room for improvement than does Goldman at 2.5x book and $280 billion in market cap. Goldman has set the bar high with their performance in 2025 and the bank will have a tough time exceeding those levels of growth and earnings. Citi, on the other hand, has thrown down the gauntlet on operating leverage and has something to prove. Fraser has already announced headcount reductions in January and will need to do more to hit the 60% efficiency target.

The Institutional Risk Analyst (ISSN 2692-1812) is published by Whalen Global Advisors LLC and is provided for general informational purposes only and is not intended for trading purposes or financial advice. By making use of The Institutional Risk Analyst web site and content, the recipient thereof acknowledges and agrees to our copyright and the matters set forth below in this disclaimer. Whalen Global Advisors LLC makes no representation or warranty (express or implied) regarding the adequacy, accuracy or completeness of any information in The Institutional Risk Analyst. Information contained herein is obtained from public and private sources deemed reliable. Any analysis or statements contained in The Institutional Risk Analyst are preliminary and are not intended to be complete, and such information is qualified in its entirety. Any opinions or estimates contained in The Institutional Risk Analyst represent the judgment of Whalen Global Advisors LLC at this time, and is subject to change without notice. The Institutional Risk Analyst is not an offer to sell, or a solicitation of an offer to buy, any securities or instruments named or described herein. The Institutional Risk Analyst is not intended to provide, and must not be relied on for, accounting, legal, regulatory, tax, business, financial or related advice or investment recommendations. Whalen Global Advisors LLC is not acting as fiduciary or advisor with respect to the information contained herein. You must consult with your own advisors as to the legal, regulatory, tax, business, financial, investment and other aspects of the subjects addressed in The Institutional Risk Analyst. Interested parties are advised to contact Whalen Global Advisors LLC for more information.