Oct 9, 2026 · 14 min
AI spending escapes corporate forecasts
Why Businesses Can’t Budget for AI
Unpredictable model use is turning AI budgeting into an operational problem just as investors question whether the technology can justify its costs.
- 1Businesses cannot reliably forecast AI bills when model choice, reasoning settings and token usage vary by task.
- 2A cheaper model can cost more when it takes extra steps, while agentic workflows make spending harder to control.
- 3AI-style phrasing is entering everyday speech, potentially improving efficiency while making communication more generic.
Don't miss
Stephanie Stamm’s model comparison shows Gemini Flash repeatedly encountering obstacles, consulting other systems and failing while costing far more than Gemini Pro.
The brief
Friday’s headlines range from Navi Pillai’s Nobel Peace Prize to a planned livestream of Nidal Hasan’s execution and protests after an ICE shooting.
Investor doubts are spreading across AI markets as weaker-than-signaled OpenAI revenue and Firmus’s canceled IPO raise questions about infrastructure spending and valuations.
Businesses struggle to budget for AI because employees switch models, reasoning settings and workflows; cheaper systems can consume more tokens and cost more overall.
Stephanie Stamm’s comparison of two Google models makes the problem concrete: Gemini Pro finished a complex video task cheaply, while Gemini Flash failed after generating a far larger bill.
Companies can impose model training and daily token limits, but the episode argues that rapid technical change keeps precise forecasting elusive—and is also reshaping everyday speech.