Palo Alto Networks CEO Demands 90% Cut in AI Token Prices

Palo Alto Networks CEO Nikesh Arora has publicly demanded that large language model providers cut token prices by as much as 90 percent. Without that, he argued in a CNBC interview on July 9, companies won't be able to deploy artificial intelligence at scale, despite surging demand for its capabilities.

Arora addressed OpenAI's claims directly, referencing the company's boast at the launch of GPT-5.6 that the model uses 54 percent fewer tokens on agentic coding tasks than its predecessors. For the Palo Alto Networks chief, that's not enough to call it a breakthrough in deployment costs.

The cheaper-token paradox

The figures Arora cited point to a paradox the whole industry is now grappling with. Prices per individual token have fallen by nearly 98 percent in recent years, yet companies' total AI bills have tripled. The reason lies in how AI agents operate: instead of a single query, they run long, multi-step sequences, generating far more model calls than a conventional chatbot.

An extreme example of this dynamic is the case of a single developer whose coding agents generated a token bill of $1.3 million in one month. Stories like this are now circulating among finance chiefs as a warning against rolling out agentic AI without budget limits.

The price of ambition

Arora argues that at many companies, IT leaders are now spending more time curbing AI usage than looking for new applications for it. That's a reversal from a year ago, when the priority was rolling AI tools into as many processes as possible, as fast as possible.

I think 54 percent is a good start - Nikesh Arora, CEO of Palo Alto Networks

At the same time, Arora maintains that demand for AI remains practically unlimited, and that the market will "rationalize itself" within the next two years, either through further price drops or by companies simply accepting higher spending as a fixed cost of doing business.

Pressure mounts on model providers

Arora isn't alone in this criticism. Palantir CEO Alex Karp has raised similar concerns about the current billing model, pointing to what he sees as an unsustainable way of pricing AI services for enterprises. A growing argument holds that model providers offer free access to consumers while charging businesses steeply for the same tokens in enterprise form.

The pricing pressure is intensifying as more open-source models reach the market, which analysts estimate cost up to four times less than offerings from leading labs, even as they remain several months behind in capability.

What it means for companies in Poland

For Polish companies still learning to manage AI budgets, the fight over token pricing has a very practical dimension. Deploying AI agents in finance, customer service, or software development can generate bills that grow far faster than original cost estimates predicted, forcing IT departments to impose hard usage limits from the start of a project.

If model providers actually start cutting prices at the pace Arora is suggesting, companies will gain more predictable deployment costs. For now, though, it's enterprise users who bear the risk of unpredictable bills for increasingly autonomous AI systems.

Sources: Palo Alto CEO Arora says AI pricing needs to fall 90% as token costs skyrocket (cnbc.com), Palo Alto CEO: AI token prices must fall up to 90% (thenextweb.com), Palo Alto Networks CEO Nikesh Arora Urges AI Companies To Cut Model Prices For Enterprises (officechai.com)