Nikesh Arora: Mythos is Real, Analytical SaaS is Dead, and Google can be a $10T company
Palo Alto Networks CEO Nikesh Arora discusses how AI is fundamentally reshaping cybersecurity, enterprise software, and business operations. He argues that analytical SaaS is effectively dead, that AI models like Mythos have demonstrated unprecedented vulnerability detection capabilities, and that the race between cyber defenders and attackers is intensifying. He also shares his views on Google's potential to become a $10 trillion company and the future of enterprise software architecture.
Summary
Nikesh Arora, CEO of Palo Alto Networks, opens by framing AI as 'democratizing intelligence' in the same way Google search democratized information. He argues AI will standardize output quality across large workforces, enabling 5,000 customer-facing employees to perform consistently rather than relying on individual expertise.
On cybersecurity, Arora shares a striking data point: using Anthropic's Mythos model in 'ultra mode' (persistent thinking), Palo Alto found vulnerabilities in their own codebase in six weeks that would have taken five to seven years to find traditionally — at a cost of only low millions of dollars. He notes Mythos was 'not oversold' and warns that similar capabilities will be available in open-source or Chinese models within three months. He frames this as a race between defenders patching vulnerabilities and attackers exploiting them, one he believes defenders are currently losing. He also highlights a critical limitation: Mythos had a 30% false positive rate, making it great for offense but problematic for defense without additional fine-tuning and harnesses.
Arora delivers a blunt verdict on the SaaS industry: analytical SaaS is 'dead' because LLMs can now perform data analysis that previously required specialized software modules. He categorizes enterprise software into three buckets: (1) analytical SaaS — obsolete; (2) infrastructure software like databases and storage — in high demand as enterprises will need 10x more data storage; and (3) systems of work/record — due for reinvention as agents replace UI-driven workflows. He predicts UI itself will largely disappear as agents handle data entry and workflow automation, enabling dramatic workforce efficiency gains.
Arora is skeptical that AI model companies (OpenAI, Anthropic) will successfully capture the application layer, arguing that profit pools lie in applications built on top of models, not in model usage itself. He compares the current landscape to waiting for a new generation of agentic application companies to emerge, similar to how enterprise software companies were built on top of prior infrastructure generations.
On national security, Arora expresses less concern about attacks on hardened critical infrastructure and more concern about economic chaos from attacks on small businesses and healthcare systems — citing the Change Healthcare ransomware attack as a cautionary example. He notes 89% of breaches occur due to stolen credentials, not sophisticated exploits.
Regarding Google, Arora predicts it will be the first $10 trillion company, citing its combination of model capabilities and enterprise sales force. On hardware, he argues it remains essential due to latency requirements, particularly in financial services, and that production bottlenecks — not design — are the limiting factor in the current AI infrastructure buildout.
Finally, on M&A strategy, Arora describes Palo Alto's evolution from acquiring product companies to feed their go-to-market engine, to now pursuing larger strategic acquisitions like their recent $25 billion identity security company. He suggests the next phase may involve using AI to dramatically improve operating margins, potentially enabling acquisitions of underperforming companies and transforming their economics. Contrary to the narrative of AI reducing headcount, he claims Palo Alto is actually growing its technical workforce because AI is driving demand for transformation across every function.
Key Insights
- Arora claims Palo Alto found vulnerabilities in their own codebase in 6 weeks using Mythos that would have taken 5-7 years through traditional methods, at a cost of only low millions of dollars.
- Arora argues that Mythos had a 30% false positive rate, making AI highly effective for cyber offense but dangerously unreliable for defense without significant post-model tuning and harnesses.
- Arora declares analytical SaaS categorically dead, arguing that enterprises no longer need third-party software to analyze their data when LLMs can run directly against raw data instead.
- Arora predicts that enterprise UI will largely disappear as AI agents replace human-driven data entry and workflow management, enabling what he calls 'five people becoming one' efficiency gains.
- Arora argues that profit pools in AI will accrue to the application layer, not to model providers, and that we are still waiting for the new generation of agentic enterprise application companies to fully emerge.
- Arora states that mythos-level AI capabilities for finding code vulnerabilities will be available in open-source or Chinese models within three months, framing this as an urgent national and enterprise risk.
- Arora asserts that 89% of breaches occur due to stolen credentials rather than sophisticated exploits, and expresses greater concern about economic chaos from attacks on small businesses than about breaches of hardened critical infrastructure.
- Arora predicts Google will become the first $10 trillion company, citing its unique combination of frontier model capabilities and the largest enterprise sales forces among the three major hyperscalers.
Topics
Full transcript available for MurmurCast members
Sign Up to Access