Enterprise AI Costs Are Exploding, And Nobody Knows How to Manage Them

TL;DR

• 1

  Annual AI budgets are getting wiped out almost immediately: The Wall Street Journal reported some companies exhausted their full AI budget in just 3 months, while Uber's CTO said the company burned through its 2026 Claude Code budget in 4 months.

• 2

  Agents are the real cost accelerant: Paul says agentic workflows consume far more tokens than normal chatbot use, citing Goldman Sachs' forecast that token demand could rise 24x to 120 quadrillion tokens per month between 2026 and 2030.

• 3

  Even small teams cannot tell where the spend is going: The hosts describe hitting Claude limits at SmarterX and seeing roughly $200 in overage credits disappear instantly, with no clear explanation of what work caused the spike.

• 4

  Token usage is a terrible KPI: Christopher Penn's Substack post, "18 ways to save AI token budgets," is highlighted for arguing that organizations should focus on planning, templates, smaller models, and better prompting, not rewarding people for burning tokens.

• 5

  The current pricing model feels unmanageable for knowledge workers: Paul argues that while token metering may fit software development, marketing, sales, HR, and customer success need flat annual pricing because nobody can realistically micromanage every employee's AI usage.

• 6

  Enterprises may accept much higher fixed prices if the value is clear: Paul says he would gladly pay something like $15,000 per employee per year for unlimited AI usage if it reliably replaces assistant, SDR, or reporting work worth $80,000 to $120,000.