TMT Daily "TOTVS: Enterprise AI costs stay high even as token prices fal..."

Research evidence excerpt

TMT Daily "TOTVS: Enterprise AI costs stay high even as token prices fal..."

Global Research

25 June 2026ab

TMT Daily Equities

Emerging MarketsTOTVS: Enterprise AI costs stay high even as

token prices fall; LatAm Telcos: Cisco sees AI Telecommunications

tripling corporate network traffic by 2029 Leonardo Olmos,AnalystCFA

leonardo.olmos@ubs.com

+55-11-9658 7533

TOTVS: Enterprise AI costs stay high even as token prices fall Gustavo Farias

Associate Analyst

What happened: According to Bain & Company, AI models keep getting cheaper, but gustavo.farias@ubs.com

the cost for a company to finish a task with AI often does not fall. Key takeaways +55-11-2767 6154

were:

Cheaper AI models do not translate into lower costs. The cost of tokens, the

small units used to measure how much AI is used, fell by about half between

December 2024 and December 2025. Even so, tokens consumed grew by 4.5

times over the same period. In some cases, AI agents already cost more than

hiring people abroad to do the same work, though Bain says this is still rare.

Overall, while AI models are getting cheaper, the effective cost per task is

often staying flat.

Three reasons the cost per task stays flat. (i) Companies tend to use the

newest models as soon as they come out, which usually stay expensive vs. the

older ones they were using. (ii) Each task uses more tokens, as AI agents take

on more complex work. (iii) Once a team discovers what agents can do, they

find 10 more workflows to throw at them.

AT&T's case: fix was to send work to smaller, specialized models. The

article cited the case of AT&T, which uses about 8 billion tokens a day, and the

company said it changed the way its system works. Big "super agents" now pass

tasks to smaller models built for specific jobs, instead of pushing everything