92. Is there an AI bubble? Thoughts On What Markets Are Really Pricing In. Mentioning Warren Buffet, Paul Volcker, Larry Page & More

This blog is the best explanation of AI intelligence increase I've seen: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/ ### Defining Market Bubbles - Traditional definition: 20%+ share price decline with economic slowdown/recession - Alternative perspective: hype/story not matching reality over time (dot-com example) - Duncan’s view: share prices ahead of future expectations - Share prices predict future revenue/profit - Decline when reality falls short of predictions ### Historical Bubble Context - Recent cycles analyzed: - COVID (2020) - pandemic-led, quickly reversed with government intervention - GFC (2008) - housing bubble, financial crisis, deeper impact - Tech bubble (1999) - NASDAQ fell 80%, expectations vs reality mismatch - S&L crisis (1992) - mini financial crisis - Volcker era (1980s) - interest rates raised to break inflation ### Current AI Market Dynamics - OpenAI: fastest growing startup ever, $20B revenue run rate in 2 years - Anthropic: grew from $1B to $9B revenue run rate this year - Big tech revenue acceleration through AI-improved ad platform ROI - Key concern: if growth rates plateau, valuations become unsustainable ### Nvidia as Market Bellwether - Central position providing GPUs for data center buildout - Recent earnings beat analyst expectations but share price fell - Market expectations vs analyst expectations are different metrics - 80% of market money judged on 12-month performance vs long-term value creation ### AI Technology Scaling Laws - Intelligence capability doubling every 7 months for 6 years - Progress from 2-second tasks to 90-minute complex programming tasks - Cost per token declining 100x annually on frontier models - Current trajectory: potential for year-long human-equivalent tasks by 2028 ### Investment Scale and Infrastructure - $3 trillion committed to data center construction this year - Power becoming primary bottleneck (not chip supply) - 500-acre solar farms being built around data centers - 7-year backlog on gas turbines, solar+battery fastest deployment option ### Bubble vs Boom Scenarios - Bear case: scaling laws plateau, power constraints limit growth - Short-term revenue slowdown despite long-term potential - Circular investment dependencies create domino effect - Bull case: scaling laws continue, GDP growth accelerates to 5%+ - Current 100% GPU utilization indicates strong demand - Structural productivity gains justify investment levels ### Market Structure Risks - Foundation model layer: 4 roughly equal competitors (OpenAI, Anthropic, Google, XAI) - No clear “winner takes all” dynamic emerging - Private company valuations hard to access for retail investors - Application layer: less concentrated, easier to build sustainable businesses - Chip layer: Nvidia dominance but Google TPUs showing competitive performance