Nikesh Arora: Mythos is Real, Analytical SaaS is Dead, and Google can be a $10T company

What It Covers

Palo Alto Networks CEO Nikesh Arora analyzes how AI reshapes cybersecurity, enterprise software, and business operations. He covers Anthropic's Mythos model finding code vulnerabilities in weeks instead of years, the death of analytical SaaS, infrastructure software as undervalued, and Google's path to a $10 trillion market cap.

Key Questions Answered

• •AI Vulnerability Detection Speed: Anthropic's Mythos model identified code vulnerabilities in Palo Alto's own codebase within six weeks — work that would have taken five to seven years manually. The cost was low millions of dollars. However, the model carried a 30% false positive rate, making it currently more useful for offense than defense.
• •Analytical SaaS Obsolescence: Any SaaS product whose core value proposition is collecting and analyzing data is effectively dead. Enterprises can now run LLMs directly against raw data, eliminating the need for third-party analytical modules. Businesses are already cutting SaaS seats by 90%, connecting remaining data sources to Claude or similar models via Slack integrations.
• •Infrastructure Software Undervalued: Enterprises will need ten times their current stored data volume within three years to train AI systems on normal versus anomalous behavior. Database and data infrastructure companies — Snowflake, Databricks, MongoDB, Oracle — are undervalued relative to this demand curve and represent a durable growth category regardless of which AI models win.
• •False Positive Rates as the Real AI Bottleneck: The critical unspoken metric in enterprise AI deployment is false positive rate. Mythos ran at 30% false positives. Deploying models at 10–20% false positive rates in business processes like insurance claims or security patching causes direct financial losses. The real competitive moat is reducing false positives to near zero without increasing false negatives.
• •Profit Pools Sit in Applications, Not Models: AI model providers are moving toward the application layer because that is where enterprise revenue concentrates. However, most enterprises will not build their own applications — they will buy vertical AI-native replacements for existing SaaS. The highest-velocity revenue opportunities are replacement TAMs, where existing budgets already exist and switching from an inferior product is straightforward.