PANW CEO: AI Token Costs Must Drop 90%

Palo Alto Networks CEO Nikesh Arora told CNBC on July 9 that AI inference pricing must fall by as much as 90% before enterprise adoption can reach meaningful scale, intensifying industry pressure on frontier model providers.

Arora says AI token costs must shrink 80% within 12 months and 90% within two years to unlock true enterprise-scale deployment.
PANW next-generation security ARR rose 60% year-over-year to $8.13 billion in Q3, fueled by rising demand to secure AI infrastructure.
PANW stock hit an all-time high of $368.17 on July 6, 2026, and has gained more than 72% year-to-date.

Lead

Palo Alto Networks Chairman and CEO Nikesh Arora used a live appearance on CNBC's "Squawk on the Street" on July 9, 2026, to deliver a pointed message to the artificial intelligence industry: current AI token costs are too high for enterprises to deploy the technology at scale, and the gap between pilot projects and full production rollouts will remain wide until inference pricing falls by a factor of ten.

What Happened

Arora's remarks came hours after OpenAI CEO Sam Altman told the same network that the company's latest model had achieved a 54% improvement in token efficiency for agentic coding tasks. Arora offered measured credit for the progress while making clear it is far from enough.

"I think 54% is a good start," Arora told anchor Seema Mody. "I think we probably need another turn at it."

He outlined a two-stage benchmark: token pricing should fall to roughly one-fifth of current levels within 12 months, then continue declining to roughly one-tenth over the following year. That trajectory amounts to a cumulative 90% reduction in Nikesh Arora AI token costs before enterprise AI adoption can cross what he described as the economic threshold chief information officers require.

Arora characterized the current state of demand as unlimited in principle but constrained by cost in practice. "It's important to understand the demand continues to be infinite," he said, "and as long as you have an infinite demand curve that you're facing, I think all these things will rationalize over time." The underlying argument is structural: inference spending has emerged as a line-item problem at the budget level, not merely a technical one.

Market Reaction

PANW stock reached a record closing high of $368.17 on July 6, three days before Arora's interview, and was trading near $338 at the time of publication — still up more than 72% year-to-date and roughly 56% over the prior twelve months. The pullback from the high reflected broader cybersecurity industry sector rotation rather than company-specific concern, with elevated valuations across the peer group drawing scrutiny after Zscaler's guidance disappointed in an adjacent earnings cycle.

With 39 analyst ratings carrying a consensus Buy, the stock trades at a premium multiple that prices in continued outperformance from Palo Alto Networks' shift to platform consolidation and AI-native security.

The Token Cost Barrier

The friction Arora described is not confined to Palo Alto Networks. Enterprise AI adoption across industries has advanced more slowly than early 2025 projections suggested, with CIOs consistently citing inference costs as a primary constraint on moving beyond proof-of-concept deployments. Running large language models at production scale — across thousands of employees or millions of customer interactions — generates token volumes that quickly overwhelm discretionary technology budgets built for traditional software licensing.

Arora is not alone in pressing the case. Palantir Technologies CEO Alex Karp has pointed to open-weight models as a more cost-viable path for enterprise customers, while technology buyers across financial services, healthcare, and manufacturing have publicly flagged the economics of hosted model APIs as a barrier to commitment.

AI and Cybersecurity Angle

Arora's position at the intersection of AI and cybersecurity gives his argument a structural dimension beyond cost management. Every new enterprise AI deployment expands the attack surface — new endpoints, new authentication flows, new data pipelines — that Palo Alto Networks is positioned to secure. The company's Q3 results illustrated the demand directly: next-generation security annual recurring revenue grew 60% year-over-year to $8.13 billion, driven in large part by clients seeking to protect AI-adjacent infrastructure.

A 90% decline in token costs, were it to materialize on Arora's proposed timeline, would accelerate the same enterprise AI buildout that underpins PANW's growth thesis. The company is therefore arguing for a price collapse that would simultaneously challenge AI providers' economics and expand Palo Alto Networks' total addressable market.

Outlook

The call for a 90% reduction in AI token costs from Nikesh Arora represents one of the most specific public benchmarks yet set by a major enterprise technology CEO — and signals that the current pricing structure from frontier model providers is becoming a credibility issue with large corporate customers. OpenAI's 54% efficiency gain marks genuine progress but leaves the gap to Arora's threshold substantial. Whether competitive pressure from open-weight alternatives, hardware improvements, or shifting model architectures closes that gap within the two-year window Arora envisions will determine how quickly enterprise AI adoption scales beyond controlled deployments. For Palo Alto Networks, the answer shapes both the pace of its security revenue growth and its platform consolidation opportunity.

---

Technology }}