Palo Alto Networks CEO Nikesh Arora has issued a stark warning: current artificial intelligence token pricing is too high to support broad enterprise adoption. In a recent interview, Arora argued that token costs need to fall by as much as 90% within the next two years before companies can deploy AI across daily operations at scale.
Arora acknowledged that recent improvements in token efficiency—such as a 54% reduction in tokens used for agentic coding by OpenAI’s latest model—are “a good start.” However, he emphasized that the industry still requires another major reduction. He expects token costs to drop to about 20% of current levels within 12 months, and potentially to just 10% the following year. Lower prices would allow businesses to run more automated tasks without straining technology budgets and would enable longer-term AI project planning.
Arora is not alone in questioning present AI pricing. Palantir CEO Alex Karp has also criticized token-based pricing from major AI developers, suggesting many companies may delay adoption while they evaluate value. Karp pointed to open-weight models as a cheaper alternative, giving firms more control over costs and infrastructure. Meanwhile, several Chinese developers have released lower-cost models that compete with U.S. products, expanding enterprise options for balancing performance, security, and operating expenses.
Despite these pricing concerns, massive infrastructure investments continue. Companies like SpaceX and Amazon have raised billions in debt to fund data centers, chips, and energy capacity. Arora noted that strong demand for AI could sustain this spending while the market adjusts. “The demand continues to be infinite,” he said, adding that costs and investment levels may eventually balance as models become more efficient.
For enterprises, the next wave of AI adoption hinges on how quickly providers reduce token costs. Companies are testing more tools, but predictable and lower pricing is essential before they can expand use across daily workflows.