Palo Alto CEO: AI Pricing Needs to Drop Another 90%
0xBroomberg
Palo Alto Networks CEO Nikesh Arora warns AI token costs must fall 90% within two years — or enterprise AI adoption will stall at the pilot stage and never reach scale.
Where does the "90% drop" figure come from?
Arora laid out a two-step roadmap: cut token costs to 20% of current levels in year one, then to 10% in year two.
This means → stacked together, the cumulative reduction hits 90% — not one leap, but two consecutive deep cuts.
He cited OpenAI's latest model gaining 54% token efficiency on agentic coding tasks, calling it "a good start, but we probably need another round."
Why can't enterprises afford current pricing?
Arora said current token pricing makes AI tools "increasingly difficult to deploy" at the enterprise level — high token costs are now the top pressure point in corporate AI budgets.
In plain terms = companies want to use AI, but at today's prices, scaling it up simply doesn't pencil out.
This reflects a deeper shift: the bottleneck in AI commercialization has moved from "can it work" to "can we afford it."
Is he the only CEO calling it too expensive?
Last week Palantir CEO Alex Karp publicly criticized the token-billing models of Anthropic and OpenAI, saying the enterprise consensus is "I'm just wasting time on tokens."
This means → this isn't one executive's complaint — it's systemic pushback from the buyer side against closed-source model pricing.
High costs are already pushing enterprises toward cheaper open-source alternatives, including Chinese models closing the gap with top U.S. labs — if closed-source vendors don't cut prices, customers will vote with their feet.
Will infrastructure spending keep climbing anyway?
Arora noted AI infrastructure spending is accelerating, and predicted the market will eventually find equilibrium between demand and cost.
His exact words: "As long as you're facing an infinite demand curve, these issues will rationalize over time."
In plain terms = he's betting demand is large enough to force costs down — but the premise is that price cuts actually happen; if they don't, enterprise AI procurement may stay stuck at the pilot stage indefinitely.
Content is for reference only, not financial advice.