Token Costs Must Drop 90% for Enterprise AI Adoption, Says CEO

The high cost of AI inference tokens has emerged as a critical barrier to enterprise adoption, according to Palo Alto Networks CEO Nikesh Arora, who outlined an aggressive timeline for necessary price reductions during a Thursday interview.

Speaking on CNBC's "Squawk on the Street," Arora said token costs need to decline 20% over the next 12 months and 90% by the following year for enterprises to deploy AI at scale. When asked about OpenAI CEO Sam Altman's recent claim that the company's latest model delivers 54% greater efficiency for coding tasks, Arora responded that while the improvement represents "a good start," the industry "probably need[s] another turn at it."

Why it matters

Token pricing directly determines whether AI deployment makes financial sense compared to traditional solutions like hiring engineers. As enterprises move from experimentation to production-scale AI, cost predictability becomes essential for budget planning and ROI calculations. The pressure for lower prices could accelerate competition among model providers and shift adoption patterns toward more efficient alternatives.

Budget exhaustion forces strategy shifts

The cost challenge has already forced major companies to recalibrate their AI strategies. Uber exhausted its entire 2026 AI budget by April, according to reporting from PYMNTS. Chief Technology Officer Praveen Neppalli Naga said the company was "back to the drawing board," while Chief Operating Officer Andrew Macdonald indicated Uber would directly compare token costs against engineering hiring expenses.

The phenomenon, dubbed "token shock" in May reporting, stems partly from agentic coding tools that generate multiple inference calls per session rather than the single call typical of standard chatbot interactions. This architectural difference compounds cost exposure as enterprises scale their AI deployments.

Cost management tactics emerge

Companies that initially encouraged broad AI tool usage when costs were lower have implemented various cost-control measures. These include usage caps, guidance on selecting appropriate tools for specific tasks, migration to older and cheaper models, and adoption of open-source alternatives.

The pricing pressure has created opportunities for Chinese AI labs, which can charge less than U.S. competitors due to more efficient models and China's lower energy costs, according to June reporting.

Investment continues despite constraints

Despite cost concerns, enterprise AI investment remains robust. A PYMNTS Intelligence report titled "The Enterprise AI Benchmark Report: Financial Services Pulls Ahead in the Enterprise AI Race" found that companies across financial services, insurance, healthcare, and media sectors continue increasing AI budgets. However, these enterprises are becoming more selective, distinguishing between projects that warrant immediate capital and those requiring further proof of concept.

The tension between growing AI ambitions and token economics suggests the next phase of enterprise adoption will hinge on whether model providers can deliver the dramatic cost reductions Arora outlined.

These details were first reported by PYMNTS.