JPMorgan Discourages Use of Expensive Models for Simple Tasks

BofA’s 200,000 Employees Submit 400,000 AI Prompts Daily

"AI Boosted Productivity, But Increased Spending as Well"

Wall Street has begun managing artificial intelligence (AI) costs. The consensus is that expensive state-of-the-art models should not be applied to simple processes such as report summaries.


Wall Street in New York, USA. Photo by Reuters Yonhap News

Wall Street in New York, USA. Photo by Reuters Yonhap News

View original image

According to Yonhap News Agency, citing Bloomberg News on July 14 (local time), JPMorgan Chase is adopting a “strategic approach” in deciding where to deploy its high-performance AI models. This comes as AI utilization proliferates throughout the financial industry, driving concerns over rising costs.


Jeremy Barnum, Chief Financial Officer (CFO) of JPMorgan, stated in an analyst call following the company’s second-quarter earnings announcement that, “There is no need to use the most advanced and expensive models just to summarize analyst reports.” He added, “The key is to use the right AI model for the right purpose.”

"Why Use Costly State-of-the-Art Models for Simple Summaries?"...Wall Street Reviews Its AI Bill View original image

The focus on cost control is due to the sheer scale of recent expenditures. JPMorgan’s planned technology spending for this year is approximately $19.8 billion (about 29.4 trillion won), up nearly $2 billion (about 3 trillion won) from last year—the largest budget among all banks. Of this total, $1.2 billion (about 1.8 trillion won) is allocated to new investments, including AI-related projects in customer services and software engineering.


Barnum noted, “(The company’s token expenses) were minor in the first half of the year, but they could increase significantly in the second half.” However, he projected that annual costs are still expected to remain relatively low. “Tokens” refer to the smallest processing units into which AI models break down sentences; the more tokens exchanged, the higher the fee. Advanced models also command higher per-token fees. Feeding lengthy documents into an expensive model quickly increases the total bill.


The appearance of generative AI tools ChatGPT, Claude, and Gemini apps. Photo by Reuters and Yonhap News Agency

The appearance of generative AI tools ChatGPT, Claude, and Gemini apps. Photo by Reuters and Yonhap News Agency

View original image

Other major banks share similar concerns. Brian Moynihan, CEO of Bank of America, said on a conference call that the company is making “significant expenditures” in technology and is devoting much time to considering which AI projects should be introduced across bank operations.


Bank of America has allocated $14 billion (about 20.8 trillion won) for technology this year. It is reported that more than 200,000 employees at the company input over 400,000 AI prompts daily. CEO Moynihan acknowledged, “There are productivity improvements (from AI adoption), but expenses have also increased accordingly.”


David Solomon, CEO of Goldman Sachs, remarked, “There is a lot of discussion around token usage and technology costs.” He argued, “While it is still early days, some trial and error and course corrections will be inevitable as people try to understand the ultimate demand for AI technology and for the AI companies themselves.”



Doubts persist over whether the money spent on AI translates into tangible results. According to a report last year from researchers at the Massachusetts Institute of Technology (MIT) Media Lab, 95% of the generative AI pilot projects undertaken by companies failed to yield noticeable profit or loss impact.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing