Overview
- Palo Alto Networks CEO Nikesh Arora said Thursday that token prices must fall about 20 percent within a year and roughly 90 percent within two years to make large‑scale enterprise AI affordable.
- SpaceXAI launched Grok 4.5 this week with published pricing that undercuts some rivals, positioning lower per‑million‑token rates as a competitive lever rather than top benchmark scores.
- Large customers have moved from unconstrained testing to strict controls by imposing monthly caps, routing workloads to cheaper models and creating AI FinOps teams to manage runaway bills.
- OpenAI and other incumbents are reportedly weighing significant per‑token price cuts under intense customer and market pressure, a shift that would compress vendor margins and complicate IPO timing.
- Startups and teams sometimes accept high token spend for speed—one founder said his firm hit about $30,000 in a month—so analysts urge tighter governance, model‑portfolio routing and attention to agentic workflows that multiply inference calls and costs.