Palo Alto CEO Arora says AI pricing needs to fall 90% as token costs skyrocket

Palo Alto Networks CEO Nikesh Arora warned AI token costs must fall sharply—about 90% to enable large-scale adoption—with a target drop to ~20% over the next 12 months and ~90% by the following year. He said current token pricing is straining enterprise AI budgets and reducing implementation willingness, even as AI spending accelerates to new highs. The backdrop includes OpenAI’s claim of 54% token-efficiency for agentic coding, but Arora signaled “another turn” is still needed for pricing to rationalize.

Analysis

The market is still treating token efficiency as a software feature, but it is really a pricing reset for the entire AI stack. Near term, that is bearish for the revenue-per-query assumptions embedded in premium model vendors and for any software names whose AI narrative depends on metered usage rather than workflow lock-in. The first-order risk is not lower demand; it is lower monetization per unit of demand, which can compress multiples before volume catch-up shows up in reported numbers.

Second-order, cheaper tokens should accelerate open-weight and self-hosted deployments, shifting spend from external APIs toward cloud infrastructure, orchestration, and enterprise controls. That is constructive for hyperscalers with distribution and balance-sheet capacity, but only if volume elasticity overwhelms the lower take-rate; otherwise AI inference becomes a commoditized utility with weaker margin capture. PLTR is vulnerable to this framing because the stock still trades partly on AI optionality, while PANW is comparatively insulated unless AI budget scrutiny spills into broader enterprise software spend.

The contrarian miss is that “cheaper AI” is not automatically bearish for AMZN; it may be the cleanest beneficiary if lower model costs expand the addressable market and push workloads into AWS-hosted environments. The falsifier is simple: if enterprise AI consumption inflects faster than token prices fall, the whole debate becomes a volume story, not a margin story, and the recent caution will look premature within 1-3 quarters.

Key Decisions for Investors

• Pair trade: long AMZN / short PLTR for 1-3 months. Thesis: token compression favors infrastructure and distribution over AI-premium multiples. Use a 6-8% stop on the spread if PLTR re-accelerates or AWS AI commentary weakens.
• On AMZN, use weakness to add rather than chase; the setup improves if the market overreacts to AI capex headlines. Risk/reward is favorable if AWS can show AI consumption growth outpacing price-per-token declines over the next two earnings cycles.
• Avoid adding to PANW on this tape; it is not a direct beneficiary of lower token pricing, and AI budget scrutiny could delay some adjacent enterprise spend. Revisit only if management data shows security attach rates rising alongside AI deployment.
• Treat PLTR as a relative underperformer until it proves that cheaper tokens expand, rather than commoditize, its enterprise AI wallet share. Falsify the short if commercial growth re-accelerates materially or if management quantifies AI-driven net retention above current trend.