Important thing about AI is that 'AI still requires human input': Monaghan

Key Concepts

• Agentic AI: AI systems capable of autonomous decision-making and executing tasks on behalf of users.
• On-Premise AI: Running AI models within a company’s own physical infrastructure rather than relying solely on public cloud services.
• Compute Constrained: A state where the demand for processing power (GPUs/CPUs) exceeds the available supply and energy infrastructure.
• AI Token Factories: Data centers optimized for the high-speed generation of AI tokens (the output of foundational models).
• Rule of 40: A financial metric for software companies where the sum of the growth rate and profit margin should exceed 40%.
• AI Slop: A colloquial term for low-quality, unverified, or hallucinated AI-generated content produced without human oversight.

---

1. Dell Technologies: The On-Premise AI Play

Michael Monahan identifies Dell as the primary vehicle for corporations looking to implement "agentic AI" while maintaining data security.

• Strategic Value: Dell provides the hardware infrastructure (servers) that allows companies to run foundational models on-premise. This is critical for regulated industries that cannot risk moving sensitive data to public clouds.
• Growth Drivers: The company’s AI server business has seen "explosive" growth. Furthermore, Monahan highlights a new partnership with Nvidia to integrate Nvidia CPUs into desktop environments, positioning Dell to benefit from both server-side and desktop-side AI workloads.
• Risk Assessment: Monahan views the near-term risk profile as low, citing strong tailwinds from the shift toward localized, secure AI processing.

2. Oracle: Infrastructure for the "Boring but Profitable"

Oracle is positioned as a leader in the foundational infrastructure required to support the AI boom.

• Addressing Constraints: Monahan argues that the current AI market is limited by energy availability and data center capacity. Larry Ellison (Oracle’s founder) anticipated this "compute-constrained" environment years ago and invested heavily in building AI-ready data centers.
• Competitive Moat: Oracle serves two specific needs:
  • Legacy Integration: Acting as the "trusted endpoint" and storage layer for existing corporate data.
  • Regulatory Compliance: Providing private cloud solutions for companies (e.g., HIPAA-regulated or government entities) that are legally prohibited from using public cloud infrastructure.

3. Palantir: Human-in-the-Loop AI

Palantir is highlighted for its ability to bridge the gap between raw data and actionable business intelligence through human-AI collaboration.

• Methodology: Monahan emphasizes that AI requires human oversight to avoid "AI slop." Palantir’s platform uses "forward-deployed engineers" to ensure human operators guide and review AI outputs, effectively magnifying human capability.
• Data Integration: Palantir solves the long-standing business intelligence challenge of unifying disconnected data silos (ERP and CRM systems) into a single, coherent view.
• Financial Performance: Monahan points to the "Rule of 40" metric, noting that Palantir’s performance improved from 127 to 145. He suggests that while retail investors have rotated into other sectors (like semiconductors), institutional growth investors are likely to enter the stock in the coming quarters, viewing the current sideways trading as a buying opportunity.

---

Notable Quotes

• On Dell: "Dell is the play for agentic AI moving into the corporates. It allows you to have your own on-premise AI server to run the foundational models to protect your data."
• On Oracle: "Oracle is attacking the boring but profitable side of AI infrastructure... Larry Ellison had the vision to see that the world would be compute constrained several years ago."
• On Palantir: "If you don't have human input, you get referred to what is colloquially referred to as AI slop. However, if you have a highly trained human operator... you magnify that human operator's capability."

---

Synthesis and Conclusion

The overarching theme of the discussion is that the "agentic AI" wave is not a monolithic trend but one that requires different layers of support. Monahan’s investment thesis categorizes these companies by their specific roles: Dell provides the secure, on-premise hardware; Oracle provides the energy-efficient, compliant infrastructure; and Palantir provides the software layer that ensures AI remains accurate and useful through human-in-the-loop integration. The common thread is the focus on "trusted endpoints" and the necessity of managing data within secure, regulated, or enterprise-specific environments.