The CEO of a $248 billion cybersecurity company just told the AI industry its pricing model is broken. Nikesh Arora, who runs Palo Alto Networks, is calling for AI token prices to fall to one-tenth of current levels within three to five years, arguing that the math simply doesn’t work for enterprises trying to deploy AI at scale.
Arora made the case during a CNBC interview on July 9, building on comments he first delivered on the 20VC podcast on June 22. His core argument: rising inference costs and losses baked into consumer AI offerings have created an economic structure that actively discourages the very enterprise adoption that AI companies desperately need.
The token economics problem
Arora pointed to concrete efficiency gains already happening, noting that OpenAI’s latest model achieved 54% better token efficiency for agentic coding tasks. His roadmap is aggressive: a 20% efficiency improvement within the next 12 months, followed by a full 90% reduction in prices within 24 months.
Palantir CEO Alex Karp has also taken aim at the token-based pricing model used by major AI developers. Karp has advocated for more open alternatives, citing enterprise hesitancy as a direct consequence of the current cost structure.
Why this matters beyond Silicon Valley
Companies are pouring staggering sums into AI infrastructure. Amazon recently issued $25 billion in bonds partly to fund its AI buildout.
Arora’s perspective carries particular weight because Palo Alto Networks sits at the intersection of enterprise software and AI deployment. The company integrates AI into its cybersecurity products, meaning Arora isn’t just theorizing about enterprise AI costs. He’s living with them every quarter.
The crypto and market angle
Projects building decentralized inference networks have marketed themselves on cost efficiency relative to major cloud providers. A 90% price drop from the incumbents would narrow or eliminate that advantage, forcing decentralized compute projects to find other value propositions like censorship resistance or data privacy.
Amazon’s $25 billion bond issuance signals conviction, but conviction funded by debt requires eventual returns. If AI pricing falls 90% before those infrastructure investments generate adequate revenue, the math gets uncomfortable.