What 60 major US financial institutions said about the cost of AI on their earnings calls, indexed from public transcripts. Updated weekly while a season is in progress.
One row per institution, ordered by when it first discussed AI cost on a call. Blank cells: no call in the corpus (coverage starts 2023). Outlined: the first "token" cost mention on record, Moody's, February 2024.
“There's a lot of talk about token spend and the cost of the technology.”
“It's very, very different lending to a chip maker that has 80% margins where we get paid back in a year and a half versus lending to someone else in the supply chain who it's going to take 15 years to get paid back or 10 years to get paid back and hope that the LLM provider who's renting that space is going to be there.”
“My longer answer though, looking to try to give you some meat on the bone, would be that AI CapEx expectations continue to move up.”
“Our associates are generating more than 400,000 prompts a day, and as of last week, we had over 300 AI use cases approved, all of which have good economics, of which 114 are live generative AI use cases.”
“Because this point about escalation of token usage, escalation of token costs, it's the same story that we've seen before with cloud.”
“Coding-related token use grew 5x quarter-over-quarter, while committed lines of AI-written code grew almost 10x.”
“In a competitive capitalist world, we always use AI to do a better job for the customers, and we can't just say, "Oh, it's going to increase our margins and we're going to keep that."”
“And of course, we have incremental cloud costs and token costs relating to large language models.”
“Our token costs and the usage that way is a fraction, a minor fraction relative to the efficiencies, and the insights, and the improvements that we gain. We measure it in hard dollars. This is not liberal arts around here.”
“If I listen to some of my peers and all the different noise around token cost explosion, we're not in that fact pattern at all here.”
“Cheaper tokens ultimately would expand usage more than they would compress price, I think.”
“Even as our AI solutions are seeing great demand, we are hearing more and more from customers that they are paying more attention to token costs and the overall expense of their own AI investments. Customers are looking for ways to minimize or manage token expenses, including building solutions in-house.”
“As it relates to token costs, I know a number of companies are talking about token costs. That's not an issue for us.”
“Critically, we carefully monitor our AI-related investments and expenses, including model usage and token costs. They're held to the same return standards we have for any other investments we make or expenses we have.”
“I think one topic which is not financially meaningful this year, but which I think is interesting and may become in the future, is the question of token expense. That is something that we're spending a bunch of time on, I think as probably pretty much everyone in corporate America is. Just for the avoidance of doubt, it is a trivial number for the first half of the year. We are forecasting some meaningful acceleration in that number for the second half of the year.”
“In terms of token costs and how you manage expenses of a large innovation and investment program, you got to remember that we've been innovating as a strategy for more than a decade now.”
Q4 full-year calls run longer and more strategic; year-over-year comparisons are the primary series.
| segment | calls | discuss AI cost | cost sentences per call | year ago | mgmt/analyst cost sentences |
|---|---|---|---|---|---|
| exchanges and data | 8 | 63% | 6.0 | 50% | 37/11 |
| asset managers | 15 | 33% | 0.7 | 6% | 9/2 |
| banks | 20 | 60% | 2.1 | 52% | 30/13 |
| insurers | 16 | 56% | 1.6 | 12% | 18/7 |
| season | calls | mention AI | discuss AI cost | cost sentences | per call | "token" AI calls |
|---|---|---|---|---|---|---|
| '22 Q4 | 61 | 7% | 3% | 2 | 0.03 | 0 |
| '23 Q1 | 62 | 19% | 3% | 7 | 0.11 | 0 |
| '23 Q2 | 62 | 40% | 8% | 28 | 0.45 | 0 |
| '23 Q3 | 62 | 34% | 10% | 17 | 0.27 | 0 |
| '23 Q4 | 62 | 36% | 13% | 27 | 0.44 | 1 |
| '24 Q1 | 62 | 34% | 16% | 20 | 0.32 | 0 |
| '24 Q2 | 62 | 47% | 16% | 23 | 0.37 | 0 |
| '24 Q3 | 62 | 44% | 16% | 22 | 0.35 | 1 |
| '24 Q4 | 62 | 44% | 23% | 31 | 0.50 | 0 |
| '25 Q1 | 61 | 44% | 7% | 7 | 0.11 | 0 |
| '25 Q2 | 62 | 66% | 29% | 40 | 0.65 | 0 |
| '25 Q3 | 61 | 67% | 38% | 69 | 1.13 | 0 |
| '25 Q4 | 60 | 85% | 50% | 68 | 1.13 | 1 |
| '26 Q1 | 59 | 80% | 56% | 90 | 1.53 | 2 |
| '26 Q2 | 59 | 80% | 53% | 127 | 2.15 | 9 |
MAXI