TL;DR
Palo Alto Networks CEO Nikesh Arora told CNBC that AI token prices need to fall by as much as 90% for large-scale enterprise adoption, calling OpenAI’s 54% GPT-5.6 efficiency gain “a good start” but not enough. He argued demand is “infinite” and costs will “rationalize over time.” His plea reflects a real paradox: per-token prices have collapsed while total enterprise AI bills keep rising, driven by agentic usage.
Palo Alto Networks chief executive Nikesh Arora says the cost of running AI needs to plunge before businesses can deploy it at scale. He told CNBC on Thursday that token prices may need to fall by as much as 90%, according to CNBC.
Arora was reacting to OpenAI’s claim that its new GPT-5.6 model is 54% more token-efficient on agentic coding. “I think 54% is a good start,” he said, making clear it is nowhere near enough.
He wants the trend to continue, with efficiency improving further over the next year and dramatically more the year after. Only then, in his telling, does mass enterprise adoption become affordable.
Despite the sticker shock, Arora is not bearish on demand. “The demand continues to be infinite,” he said, arguing that with an infinite demand curve, costs “will rationalize over time”.
His logic is that the market will either grow into the spending or force prices down. Budgets should ease, he suggested, as the underlying technology becomes more efficient.
The paradox behind the plea
Arora’s complaint captures a genuine puzzle in enterprise AI. Per-token prices have collapsed, yet total bills keep climbing, so much so that prices fell 98% while enterprise AI bills tripled.
The culprit is agentic AI, which calls a model over and over to complete a task. A single ambitious project can burn through a fortune, as one developer’s agents ran up a $1.3m token bill in a month.
That is why cheaper headline prices do not automatically translate into lower costs. Usage grows faster than prices fall, and the bill goes up anyway.
Squeezed buyers and a price war
The strain is already changing behaviour, with some firms capping how much AI staff can use as costs bite. Arora is voicing, from the buyer’s seat, a frustration many enterprises share.
The good news for him is that a price war is under way, with DeepSeek making a 75% discount permanent and rivals racing to match. A wave of startups is also chasing cheaper inference to squeeze more output from every chip.
Whether that adds up to Arora’s 90% is another matter, since efficiency gains can be swallowed by ever-heavier usage. His bet is that scale eventually wins, and the economics settle.
For now, the man running a cybersecurity giant is effectively telling AI vendors their product is still too expensive to use everywhere he wants to use it. Coming from a customer of that size, it is a message the model makers will hear.