AI token costs risk choking enterprise adoption, Palo Alto's Arora warns
The economics of artificial intelligence deployment are outpacing what most enterprise budgets can absorb. Against that backdrop, Palo Alto Networks chief executive Nikesh Arora made a pointed call: AI pricing needs to fall 90…
Key takeaways
- Palo Alto Networks CEO Nikesh Arora says AI pricing must fall 90 percent before enterprise adoption can scale beyond early experiments.
- Arora argues that skyrocketing token costs are passed to enterprises through token-based pricing, making the math unworkable for most use cases outside narrow, high-value applications.
- He warns that if inference costs do not compress, companies will evaluate AI on paper but not deploy it in volume.
- As a buyer and integrator of AI capabilities, Palo Alto Networks and its AI-powered products face unit-economics pressure from elevated token costs before facing it from competition.
- Arora frames 90 percent as a concrete threshold that either gets crossed or enterprise AI adoption stalls at the pilot stage.
The economics of artificial intelligence deployment are outpacing what most enterprise budgets can absorb. Against that backdrop, Palo Alto Networks chief executive Nikesh Arora made a pointed call: AI pricing needs to fall 90 percent before commercial adoption can move from early experiments to anything resembling scale across corporate infrastructure.
The cost barrier Arora named
Arora's argument centers on token costs, which he described as skyrocketing at a pace that prices out the broad business market. Costs that rise steeply on the supply side get passed to enterprises through token-based pricing, and at current rates the math does not work for most use cases outside of narrow, high-value applications. The implication is direct. If inference costs do not compress, AI becomes a capability that companies evaluate on paper but do not deploy in volume. That is a material distinction for a sector where the investment thesis rests on enterprise spending converting to recurring, high-margin revenue.
Where this sits in the AI capex cycle
The observation lands at an uncomfortable point in the broader cycle. Hyperscalers and model providers have committed heavily to compute, and those costs flow through to API pricing. Arora's 90 percent threshold implies the current cost curve has not bent far enough to reach the demand environment that technology investors have been pricing into equities. The gap between what it costs to run a model and what a mid-size enterprise will pay per query remains the sector's central tension.
The read-through for enterprise security
Palo Alto Networks is itself a buyer and integrator of AI capabilities, which gives Arora's comments more than theoretical weight. A chief executive who ships AI-powered products has a direct stake in where token costs land. If token economics stay elevated, the unit economics of those products face pressure from the cost side before they face it from competition. That is the more specific risk the market should read through from his remarks.
The macro caveat
The capex cycle from large cloud providers shows no sign of reversing, and sector-wide AI spending commitments remain high. But Arora's 90 percent figure is a concrete threshold: it either gets crossed or enterprise AI adoption stalls at the pilot stage.
Source · 來源