He Predicted the AI Bubble in 2023 | Doug Clinton and Gene Munster on Why We're Still in 1996

Key Concepts

• AI as Electricity: The fundamental principle that AI is the conversion of electricity into intelligence.
• The AI Bubble: The theory that AI is currently in a cycle similar to the 1995–1996 dot-com era, with significant room for growth before reaching a peak.
• Inference/Compute Bottlenecks: The physical constraints (power and data center capacity) limiting the deployment of AI.
• Knowledge Worker Disruption: The shift in labor markets where AI acts as a binary tool—either supercharging productivity or rendering roles obsolete.
• Agentic Revolution: The transition from simple chatbots to AI agents capable of writing code and executing complex tasks autonomously.
• Recursive Self-Improvement: The process where AI models are used to train and improve subsequent versions of themselves.
• Hyperscaler Capex: The massive capital expenditure by tech giants (Microsoft, Google, Amazon, Meta) to build AI infrastructure.

---

1. The State of the AI Market

Doug Clinton and Gene Munster of Deepwater Asset Management argue that we are in the early stages of a massive technological bubble, comparable to the mid-90s internet boom.

• Market Timing: While some suggest we are in 1998, the speakers believe we are closer to 1995–1996, implying several years of growth remain.
• The "Utility" Phase: The market has moved past the hype phase into a period of tangible utility, driven by coding agents like Anthropic’s Claude and OpenAI’s Codex.
• Investment Thesis: Investors should evaluate management teams based on their competence in predicting the future. If management is investing heavily in AI, they should be rewarded; if they lack vision, they should be penalized.

2. The "Model War" and Performance

The speakers categorize the current landscape as a race between five major "Western" players: OpenAI (GPT), Anthropic (Claude), Google (Gemini), Meta (Spark/Llama), and xAI (Grok).

• Rankings: As of the discussion, GPT-5.5 is considered the top-performing model, followed closely by Anthropic’s Opus 4.7.
• Differentiation: Models have distinct "personalities." Codex is preferred for production-grade coding, while Claude is often cited as a superior "thought partner" for product ideation.
• The "Infinite Demand" Theory: The market is not necessarily zero-sum. Because demand for intelligence is effectively infinite and providers are capacity-constrained, even the "second or third best" models will find full utilization.

3. Labor Market Impact

The speakers predict a period of "acute knowledge worker unemployment" over the next five years, more severe than the transitions seen during the mobile or internet revolutions.

• The 80/20 Rule: 20% of high-performing employees who master AI tools will become more valuable, while the remaining 80% face significant risk of displacement.
• The "Detective" Role: New job categories will emerge, specifically for "detectives"—individuals who can source unique, non-public data to feed into AI models to create proprietary value.

4. Infrastructure and Energy

A core argument is that the AI revolution is fundamentally an energy revolution.

• Energy Scarcity: The U.S. is "woefully underbuilt" for the power requirements of modern data centers. Nuclear energy (including Small Modular Reactors) is identified as a critical, necessary component of the future energy mix.
• Capex Trends: Hyperscaler capital expenditure is growing at rates far exceeding initial analyst expectations (projected 20–30% growth vs. the expected 10%).

5. Intelligent Alpha: Methodology

Deepwater’s project, Intelligent Alpha, uses AI to manage portfolios by predicting forward earnings.

• The Process: They use a "harness" system where LLMs are fed consistent data packets (earnings transcripts, historical financials, street estimates) to predict revenue and EPS direction.
• Key Finding: AI models are superior to humans in this task because they lack the "endowment effect" (the emotional bias of holding onto a stock because you own it). However, they struggle with "high conviction" calls that deviate from historical base rates.

6. Future Frontiers: Space

The speakers highlight space as a long-term investment theme, specifically regarding:

• Orbital Data Centers: Moving data centers to space to bypass local permitting hurdles and potentially utilize more efficient power sources.
• Space-Based Manufacturing: Partnering with biotech firms (e.g., Varta and United Therapeutics) to develop drugs in zero-gravity environments that cannot be produced on Earth.

---

Synthesis and Conclusion

The overarching takeaway is that AI is a transformative force currently constrained by physical infrastructure—specifically energy and compute capacity. While the short-term outlook involves significant labor market disruption and potential market volatility, the long-term trajectory is one of massive productivity gains. The "winners" will be those who successfully navigate the transition from simple model usage to agentic, recursive workflows, and the companies that secure the energy required to power this new intelligence.