On the morning of October 13, 2026, journalists and analysts listening to JPMorgan Chase's Q3 earnings call heard CFO Jeremy Barnum say something that would have been unusual at any other point in the AI cycle.
The bank's AI spending has been misreported and the actual figure is significantly lower than the $700 million that has circulated in media coverage — and no, he would not say what the real number is.
"I'm very reluctant to spend a lot of energy quantifying" AI benefits, Barnum said. He acknowledged that current gains are not yet visible in quarterly earnings.
He pointed to early productivity evidence in software engineering, including a test involving legacy code rewrites. He noted that some productivity gains are being offset by token costs. And then he moved on.
The same day, Goldman Sachs CEO David Solomon spoke at length about AI on his firm's earnings call. He cited studies estimating the U.S. data center buildout could require more than $10 trillion of investment through 2032.
He described Goldman's position in the AI cycle as a financier of the infrastructure those systems require, not a developer. He laid out a framework for responsible AI governance.
He did not say a single thing about what Goldman is spending on AI internally, what it is getting for that spend, or whether any of it is showing up in results.
The following day, Bank of America CEO Brian Moynihan said the bank has approximately 130 AI projects in production, each with a defined payback, and that AI-related benefits are running at $250–300 million this year.
Moynihan said benefits are expected to be two to three times the expense involved over time. He did not say what the expense involved currently is.
What Each Bank Said, and Didn't
The Q3 2026 earnings calls produced a consistent pattern across all five institutions. Banks were willing to discuss AI in terms of operational outcomes — faster code, more efficient advisers, fewer manual processes.
They were less willing to discuss what they are paying for those outcomes. And they were almost entirely unwilling to give analysts the pair of numbers — cost and benefit — that would allow an outside observer to evaluate whether the returns are real.
Citigroup CFO Gonzalo Luchetti said the bank deploys approximately $2 billion per year in technology investment, with AI layered on top.
The "layered on top" construction was precise and deliberate as it signals that AI is additive to an existing cost base, but it does not say by how much.
CEO Jane Fraser's disclosures were operational: 14,000 engineers using the firm's agentic coding tool Devon, a 30% reduction in tech vendor spend, new product launches accelerating by 30–40%, 100 manual processes being re-engineered end to end. No cost figure accompanied any of these numbers.
Wells Fargo gave the thinnest disclosure of the group. CEO Charlie Scharf described the bank's AI deployments across code writing, branch activity, marketing, and cybersecurity. He said the bank ranked number five globally and number three in the United States on the Evident AI Index.
He launched a new AI adviser tool for wealth management clients built on BlackRock's Aladdin Wealth platform. CFO Michael Santomassimo listed AI-related disruption among the external risks the firm monitors. Neither offered an AI spending figure or a quantified benefit.
JPMorgan's Barnum was the only CFO on any of the calls to directly address the accuracy of a published AI spend figure, and his message was that the number most often cited is wrong on the high side.
He said the earlier benefit quantification that accompanied those expense figures was accurate and included established machine learning applications such as fraud prevention, which suggests the $700 million figure was conflating long-standing ML infrastructure spend with newer generative AI investment. But he would not say where the line falls or what the newer spend actually is.
The Measurement Problem
The reluctance to quantify is not uniform across banks. Bank of America's Moynihan announced $700 million in expected AI benefits in 2027, based on a pipeline of projects already in place.
CFO Alastair Borthwick noted that headcount has fallen as the company automates processes, describing technology modernization and AI as explicit drivers of that reduction. These are verifiable operational outcomes, even if the underlying cost structure is not disclosed.
But Moynihan's $700 million figure is a benefit projection with no denominator. Without knowing what Bank of America is spending on AI specifically, it is not possible to assess whether $700 million in benefits represents a strong return, a marginal one, or something in between.
Goldman's Solomon offered a version of the same dynamic from the opposite direction. His call was dominated by the AI infrastructure investment cycle as a financing opportunity — the $10 trillion data center buildout, the record $3.37 billion in equities financing revenues, the firm's positioning as a capital markets and lending beneficiary of AI capital expenditure.
What Solomon did not say is what Goldman itself is spending to deploy AI internally, or what operational benefits the firm is seeing from that deployment. The firm's AI posture on the call was almost entirely external-facing.
Barnum's remark that he is "very reluctant to spend a lot of energy quantifying" AI benefits is the clearest statement of a position that appears to be shared, if not articulated, across the industry.
The argument implicit in that posture is that AI's benefits are systemic and structural. They show up as fewer people needed to do more work, faster product cycles, lower vendor costs, and that trying to isolate them in a quarterly earnings number would be both methodologically difficult and potentially misleading.
Banks asking investors to trust that AI is working, while declining to show the arithmetic, are asking for a degree of patience that will, at some point, need to be accompanied by numbers.
JPMorgan's Barnum acknowledged as much when he said current gains are not yet visible in quarterly earnings. The question implicit in that statement is when they will be.