Palo Alto Networks CEO: "AI Found 5 Years of Bugs in 6 Weeks"
Episode Details
In this episode of the All-In Podcast, Palo Alto Networks CEO Nikesh Arora joins hosts David Friedberg, Jason Calacanis, and Chamath Palihapitiya to discuss the massive implications of AI on business. Arora, who previously held leadership roles at Google and SoftBank and sits on the board of Uber, shares how his company utilized the highly capable generative AI model Mythos to discover five to seven years' worth of code vulnerabilities in just six weeks. This breakthrough fundamentally redefines Cybersecurity in the AI era, turning the race between cyber defenders and attackers on its head. However, Arora notes that Hallucination rates (like a 30% false positive rate seen in testing) remain a major hurdle for fully automated digital defense. The discussion heavily focuses on the massive disruption coming to the traditional SaaS industry. Arora points out AI as a technology equalizer and predicts that Analytical SaaS is effectively dead. Instead, AI agents will handle data querying and backend tasks natively through Automation, rendering legacy visual Computing Interfaces obsolete. In contrast, Infrastructure Software such as Data Bricks, Snowflake, and MongoDB are highly undervalued because enterprises will need 10x the data storage to power AI Infrastructure. Consequently, foundational Systems of Record like Salesforce, Oracle, and SAP will have to be completely re-engineered for an AI-native world. The panel explores evolving software business models, observing that base AI Models are becoming a commoditized utility, even as heavyweights like OpenAI and anthropic (led by Dario Amodei and creators of Claude) battle for supremacy. According to Arora, the true Profit Pools lie firmly within Enterprise Applications. This dynamic is accelerating Enterprise AI Adoption while enabling agile startups to undercut legacy SaaS incumbents by utilizing disruptive Consumption-based pricing. On the physical and architectural side of tech, Arora explains the market's ongoing reliance on Hardware and Data Centers. Because financial titans like Goldman Sachs, JP Morgan, and Morgan Stanley demand ultra-low latency, they continue to prioritize owning and operating physical hardware rather than fully migrating workloads to Cloud Computing. This ongoing necessity has fueled a resurgence for hardware providers like Dell. The conversation also spans broader topics including National security concerns over economic havoc on small businesses, the risks of Open source AI (which IBM is spending billions attempting to secure), and Palo Alto's own aggressive M&A strategy, which echoes deep-value philosophies from investors like Bill Ackman. Lastly, Arora offers brief 'Armchair CEO' takes, praising Waymo's functional autonomous vehicles and projecting Alphabet's continued trillion-dollar ascent.
The episode explores the transformative impact of AI on enterprise operations and cybersecurity, highlighting a shift away from traditional analytical SaaS toward agentic, AI-native workflows. While foundational infrastructure software and hardware providers are positioned for massive growth due to increased data and compute demands, legacy SaaS incumbents face significant disruption from consumption-based, AI-driven competitors.
Generated with gemini-flash-lite-latest on 6/27/2026, 5:16:31 AM. For research only. Not financial advice.AI Infrastructure & Data Storage
Enterprises will require 10x more data storage and compute infrastructure to power AI-native systems, making foundational data platforms highly undervalued.
The guest argues that AI agents require massive amounts of enterprise-wide data to learn 'what good looks like,' necessitating a surge in core storage and database infrastructure.
- Enterprises will need 10x the data storage to power AI infrastructure over the next three years.
- Infrastructure software companies like Data Bricks, Snowflake, and MongoDB are identified as critical beneficiaries.
- Increased enterprise adoption of agentic AI workflows.
- Continued growth in data-heavy AI training and inference requirements.
- Potential commoditization of storage layers.
- High capital expenditure requirements for data centers.
- Analyze capital expenditure trends for major cloud providers and database companies.
- Examine unit economics of data storage vs. compute in AI-native enterprise stacks.
The 'Analytical SaaS' Apocalypse
Traditional analytical SaaS companies that rely on manual UI-based data querying are facing obsolescence as AI agents natively interface with backend systems.
The guest suggests that analytical SaaS is 'dead' because LLMs can directly query and analyze data, removing the need for intermediary software modules and visual interfaces.
- A company reduced its software bill by 90% by replacing 17 analytical SaaS seats with direct AI-agent integration.
- AI agents can now perform backend tasks natively, rendering legacy UI-heavy software redundant.
- Widespread adoption of natural language interfaces for enterprise data.
- Increased pressure on IT budgets to consolidate redundant software.
- High switching costs for deeply embedded systems of record.
- Potential for legacy incumbents to successfully pivot to AI-native architectures.
- Evaluate the percentage of revenue derived from 'analytical' vs 'system of record' modules in major SaaS companies.
- Monitor churn rates in traditional analytical SaaS platforms.
Hardware Resurgence for Low-Latency Computing
The demand for ultra-low latency in high-stakes industries like finance ensures that physical, on-premise hardware remains a critical, high-growth sector.
Financial institutions are reluctant to migrate to the cloud because increased latency directly reduces profitability, forcing them to maintain physical hardware.
- Large financial services firms like Goldman Sachs and JP Morgan prioritize owning physical hardware to maintain low latency.
- Hardware providers like Dell are seeing a resurgence in market value.
- Continued growth in high-frequency trading and data-intensive financial services.
- Supply chain constraints for high-performance hardware components.
- Technological breakthroughs in cloud latency reduction.
- Increased regulatory pressure on on-premise data security.
- Track capex spending in the financial services sector for hardware vs. cloud services.
- Monitor global manufacturing capacity for high-end server hardware.
Watchlist
- Palo Alto Networks
- Dell
- Snowflake
- Data Bricks
- Waymo
- Alphabet
Open Questions
- How quickly can legacy 'systems of record' like Salesforce or SAP re-engineer their platforms to be AI-native before losing market share?
- What is the actual false-positive rate of current frontier models in production enterprise environments?
- Will the profit pools in AI ultimately reside in the model layer or the application layer?
- How will the '3-month' timeline for open-source AI to reach 'Mythos-level' code vulnerability detection impact global cybersecurity?
Key Topics & People
The delivery of computing services over the internet, critical for hosting AI models.
Large facilities housing servers and GPUs for AI compute, which are increasingly hard to power and zone.
The physical and technological backbone required to train and run AI models.
The intersection of AI capabilities and cybersecurity, raising concerns about automated vulnerabilities.
CEO of Anthropic, noted for navigating regulatory hurdles surrounding AI model releases.
AI models with weights and architectures freely available for download and modification.
.jpg?width=300){kind=avatar}
Investor and podcast host analyzing AI infrastructure, politics, and markets.
.jpg?width=300){kind=avatar}
Investor and podcast host moderating discussions on startups and tech markets.
Host of the All-In Podcast conducting the interview with Ryan Cohen.
The podcast hosting the interview with GameStop CEO Ryan Cohen.
Cybersecurity company whose CEO tested Anthropic's models.
The defense and strategic safety of the US, which Fetterman argues is synonymous with energy security.
Software designed to collect and analyze data, which is becoming obsolete due to LLMs.
The segments of the market where the majority of profit is generated, migrating to AI applications.
Foundational software like databases and data storage that will see massive growth due to AI data demands.
A business model where customers are billed based on their specific usage, disrupting traditional per-seat SaaS models.
The use of agents and AI to complete background business processes without human intervention.
The software layer sitting atop foundational AI models, predicted to capture the majority of AI-driven profit.
The integration of AI technologies by businesses into their workflows, operations, and software systems.
False positive results generated by AI models, a significant challenge for enterprise defense mechanisms.
The concept that AI is standardizing and democratizing intelligence across enterprise outputs.
_(cropped).jpg?width=300){kind=avatar}
An investor mentioned regarding identifying overbeaten companies ripe for acquisition.
A multinational investment bank avoiding full cloud transitions due to high-frequency latency concerns.
A financial services institution noted for utilizing physical hardware over the cloud.
.jpg?width=300){kind=avatar}
A data analytics and infrastructure platform listed as essential infrastructure software.
A major cloud-based software company recognized as a core system of record.
Traditional user interfaces (UIs) that are predicted to disappear as AI agents take over backend data interactions.
Core enterprise applications that hold business data, poised for reinvention via AI automation.
The Software as a Service sector, currently undergoing massive disruption due to AI agents.
.jpg?width=300){kind=avatar}
CEO of Palo Alto Networks and former executive at Google and SoftBank.