Will AI Token Costs Rise This Year?

JPMorgan Chase disclosed during its second-quarter 2026 earnings call on July 14, 2026, that it expects a meaningful acceleration in artificial intelligence (AI) token expenses in the second half of the year. Chief Financial Officer Jeremy Barnum said the cost remains trivial for the first half of the year, and that the full-year contribution is still expected to be trivial despite the anticipated increase.

Barnum said the company is working to route AI tasks to appropriately sized models rather than defaulting to the most expensive options available. As an example, he said summarizing lengthy analyst reports does not require the latest, most costly model. The company said it has built infrastructure over the past couple of years to support this kind of model selection, including the use of open source tools where appropriate.

The AI commentary came alongside a standout quarter. JPMorgan reported net income of $16.9 billion and EPS of $6.14, with return on tangible common equity of 23%. Revenue rose 15% year-over-year, driven by markets, investment banking and asset management, while expenses climbed a similar 15% to $27.3 billion on higher volume-related costs and continued investment. Investment banking fees were up 30% year-over-year, and equities trading revenue jumped 86%. 

The bank raised its full-year net interest income guidance to approximately $105.5 billion and now expects full-year adjusted expenses of about $107.5 billion, alongside a lower forecast for card net charge-offs of roughly 3.2%.

Nearly 1,000 Use Cases Across the Firm

Chairman and CEO Jamie Dimon said the company has close to 1,000 AI use cases in development, with about 50 considered the most important.These span risk, fraud, marketing, hedging, prospecting, note-taking, idea generation and document reading. He said the firm has reduced jobs by 30% to 40% in select discrete areas as a result of AI adoption, adding that most affected employees were offered roles elsewhere in the company.

An analyst on the call referenced Block's reported 40% workforce reduction, which the analyst attributed to AI-driven efficiency gains, and asked whether a similar restructuring could apply to banks. Dimon responded that JPMorgan expects AI to generate efficiency but pushed back on the idea that this would translate directly into higher margins for the company.

"We are going to use AI to do a better job for our clients," Dimon said, adding that over time "the benefit accrues to the customer, not to JPMorgan in this case," since competitors are adopting the same tools. To illustrate why he doesn't expect AI to permanently expand margins, Dimon offered a hypothetical: if JPMorgan sustained a 50% return on equity growing 10% a year, the company would, within 50 to 60 years, approach the size of the entire U.S. economy — an outcome he called implausible.

AI Capital Expenditure Growing, Hard to Isolate

Dimon said industry-wide AI capital expenditure rose from roughly $400 billion last year to about $700 billion this year, against total U.S. capital expenditure of roughly $4 trillion annually, with AI spend projected to exceed $1 trillion next year. 

He cautioned that isolating AI-specific spending is becoming harder, since AI infrastructure projects such as data centers also drive demand in adjacent trades like plumbing and electrical work, and non-AI capital expenditure could see a modest decline next year as spending shifts toward AI.

Dimon also said he expects expense growth broadly to slow, "maybe... in 2027 or 2028." AI will be a central topic at the firm's leadership offsite in July, spanning front office, middle office, back office, marketing and risk functions.

>