Palo Alto Networks’ Nikesh Arora says AI token prices need a 90% cut to make enterprise rollouts viable

Palo Alto Networks CEO Nikesh Arora warned that AI token prices must fall by as much as 90% for broad enterprise deployments to pencil out, setting a one-year target for costs to drop to a fifth of today’s levels and to a tenth the year after. His remarks to CNBC followed OpenAI’s claim of a 54% gain in token efficiency on agentic coding tasks—progress Arora called “a good start” but insufficient for large-scale corporate use. Arora’s comments echo growing pushback from executives, including Palantir’s Alex Karp, against per‑token pricing as companies wrestle with bills inflated by chained model calls, even as headline token rates have plunged. Firms such as Uber and Microsoft have restricted employee access to pricey AI tools, while some enterprises pivot to cheaper open‑weight alternatives, including Chinese models. Despite cost friction, capital spending on AI infrastructure is accelerating, with SpaceX and Amazon each raising $25 billion in recent debt offerings.

Read more