Palo Alto Networks CEO Says AI Security Costs Hinder Endpoint Deployment

Palo Alto Networks CEO Nikesh Arora cautioned that artificial intelligence expands the cybersecurity threat surface. He added that the current operational costs of running frontier models at the endpoint level remain too expensive for mass adoption. Arora highlighted the economic gap between high-performance AI requirements and standard enterprise security budgets during a presentation at the Goldman Sachs Communacopia + Technology Conference.

"Mithos has been more useful for me as a marketing tool than anything I did for eight years," said Nikesh Arora, Chief Executive Officer, Palo Alto Networks, in a statement regarding the company's internal vulnerability discovery tool.

Economic Barriers to Endpoint AI

Arora noted that deploying large language models directly onto individual devices is currently unviable. AI-driven security promises faster threat detection, but the compute costs required to run these models would exceed the typical annual security budget of $30 to $40 per device. This creates a bottleneck for organizations looking to integrate advanced AI agents into their local security stacks.

Palo Alto Networks relies on centralized data processing rather than local deployment. The company currently processes 19 petabytes of data daily via Google Cloud infrastructure, supported by a network of 180 million deployed sensors worldwide. This scale allows the firm to leverage centralized AI capabilities without requiring individual endpoints to bear the processing burden.

Internal Vulnerability Management

To manage its own security posture, the company has deployed its proprietary AI-driven vulnerability tool, Mithos. According to a transcript of the conference session, the tool has successfully identified 1,200 internal vulnerabilities since its launch. By automating the discovery process, the company aims to reduce the time security teams spend manually auditing code and infrastructure.

What Is at Stake for Security Platforms

The tension between AI-driven threat surfaces and the cost of defensive AI represents a shift in how cybersecurity firms prioritize their product roadmaps. Attackers utilize AI to scale their operations, forcing platform providers to balance the need for high-compute security tools with enterprise budget constraints. By focusing on cloud-based processing rather than endpoint-heavy models, Palo Alto Networks is attempting to provide centralized AI-driven protection that avoids the cost limitations described by Arora.