AI keeps getting cheaper per token. The bills keep getting bigger, because each job uses more tokens.
AI keeps getting cheaper per token. The bills keep getting bigger, because each job uses more tokens.
Ed Barrow of Cloud Capital says token prices have dropped as the infrastructure got more efficient, but the number of tokens consumed has massively outpaced it. Products call the model more often, for more use cases. Parallel CEO Parag Agrawal says the newest models think longer, so more compute goes into every answer. Pegasystems CFO Ken Stillwell says some frontier tokens now cost more than earlier models did, and companies send work to models far bigger than the task needs.
The unit price drop is real. Volume is outrunning it. The savings come from matching the model to the task: tuned local models, smaller models like Haiku 5.5, and tools that keep those costs in control.
Sources: Ed Barrow (Cloud Capital), GrowCFO Show, "#303 The True Cost of AI: What CFOs Need to Know," Sep 29, 2026. Parag Agrawal (Parallel), 20VC, "Five Predictions for a World of Agents | The Ads Business Model Will Die," Sep 26, 2026. Ken Stillwell (Pegasystems), CFO Thought Leader, "1218: Why Pegasystems Is Betting on Work, Not Tokens," Sep 30, 2026.