Nikesh Arora, chairman and CEO of Palo Alto Networks, said AI token costs must drop approximately 90% [1] for enterprises to adopt the technology at scale.
This pricing hurdle represents a critical bottleneck for the corporate sector. While many businesses seek to integrate artificial intelligence into their workflows, the recurring cost of processing tokens can make large-scale deployment financially unsustainable.
Arora discussed these challenges during an interview on CNBC’s “Squawk on the Street” program on July 9, 2026 [1, 4]. He said that the current cost structure prevents the technology from moving beyond limited use cases into comprehensive enterprise operations.
"We need to see the pricing of AI come down," Arora said [1].
He acknowledged that some progress has been made by major developers. Arora cited a 54% [3] efficiency gain from a recent OpenAI model as a positive step toward lowering costs. However, he said that this improvement is not yet sufficient to trigger mass adoption across the business world.
"Token costs need to fall about 90% before AI can be adopted at scale," Arora said [2].
Arora said that the gap between current efficiency and the required price point remains significant. He said that while the 54% [3] gain is a good start, the industry still requires a much deeper reduction in pricing to make AI viable for the average large-scale corporate environment [3].
The CEO's comments highlight a growing tension between the capabilities of generative AI and the economic realities of deploying those tools across thousands of employees and millions of daily transactions.
“"Token costs need to fall about 90% before AI can be adopted at scale."”
The insistence on a 90% price reduction suggests that for many enterprises, AI is currently a luxury tool rather than a utility. If token costs do not decline rapidly, the 'AI revolution' in the corporate sector may remain confined to high-margin industries or specific high-value tasks, delaying the broader productivity gains promised by the technology.
