Overview
- Palo Alto CEO Nikesh Arora told CNBC on Thursday that token prices should fall about 20% in the next 12 months and roughly 90% by the following year to make large‑scale enterprise AI viable.
- Companies have moved from flat subscriptions to metered, per‑inference billing, which multiplies costs for agentic systems that run many steps and has made forecasting and budgeting much harder.
- Major firms are cutting use and adding controls, including monthly caps, routing apps to cheaper internal models, and creating AI FinOps teams to track token spend and tie projects to measurable ROI.
- Employee behaviors such as 'tokenmaxxing' and internal leaderboards have inflated consumption at some firms, forcing firms to remove incentives and tighten governance to stop wasteful usage.
- Rising inference demand is driving a large compute and power buildout while also pushing buyers toward older, open‑weight and non‑U.S. models; this could pressure vendor margins and reshape the market unless costs and efficiency change.