Not Once in 18,900 Tries | Michael Mauboussin on What History Says About AI Growth

| Stock Investing | April 01, 2026 | 19.5 Thousand views | 1:01:13

TL;DR

Michael Mauboussin applies historical "base rate" analysis to AI growth projections, revealing that OpenAI's target of 108% annual compound growth has zero precedent in 18,900 firm-years of data, while warning that intangible-intensive business models create fatter tails of both spectacular success and catastrophic failure.

📉 Historical Base Rates vs. AI Growth Targets 3 insights

Unprecedented revenue growth requirements

OpenAI's forecast to grow from $3.7 billion (2024) to $145 billion (2029) implies a 108% compound annual growth rate—a feat no company in the $2-5 billion revenue range has achieved in 75 years of historical data covering 18,900 firm-years.

Statistical implausibility

This growth trajectory represents a 9.5 standard deviation event from the historical mean of 7% growth, and remains an outlier even after OpenAI revised 2029 targets upward to $185 billion (118% CAGR).

The three-legged execution stool

Achieving these targets requires simultaneously maintaining a competitive product, attracting elite talent, and burning through $218 billion in cash before reaching free cash flow neutrality—a combination of challenges no previous company has solved at this scale.

⚖️ Intangibles, Risk, and Market Concentration 3 insights

Fatter tails in intangible-heavy businesses

Companies intensive in intangible assets show similar mean returns to physical-asset firms but exhibit substantially higher standard deviations, creating more extreme outcomes—both spectacular winners and total failures.

The Enron cautionary tale

Enron grew 61% annually to reach $100 billion revenue by 2000 using an 'asset-light' intangible strategy, only to file for bankruptcy nine months later, illustrating that intangible-intensive models can collapse as rapidly as they ascend.

Magnificent 7 economic dominance

Today's largest tech companies comprise one-third of total market capitalization but generate roughly two-thirds of all economic profits, growing faster than historical large companies due to proprietary software creating unique moats that resist competitive diffusion.

🏗️ Infrastructure Reality and Project Risk 2 insights

Dismal big project base rates

Analysis of 16,000 large-scale projects reveals fewer than 50% finish on budget, under 9% hit both time and budget targets, and only 0.5% deliver on time, on budget, and with promised benefits.

AI data center delays mounting

Despite modularity advantages, 25% of AI data centers faced delays in 2025, with industry estimates suggesting 30-50% of 2026 projects will be delayed due to permitting, energy, and cooling constraints.

Bottom Line

Investors should anchor AI growth forecasts using historical base rates as a starting point, recognizing that while intangible assets enable unprecedented scale, they concentrate risk in fatter left-tail outcomes that demand probability-weighted analysis rather than linear extrapolation of current hype.

More from Excess Returns

View all
Labor Market Cracks, Wild Fed Credibility Data and Semis Running Out of Pie | Last Call
1:04:09
Excess Returns Excess Returns

Labor Market Cracks, Wild Fed Credibility Data and Semis Running Out of Pie | Last Call

The video analyzes SpaceX's volatile IPO through the lens of complex index flow dynamics, while macro strategist Andy Constant presents his 'Fab Five Freddy' thesis arguing that semiconductor equipment manufacturers are the next critical bottleneck in the AI supply chain as the industry faces oversubscribed earnings expectations against finite GDP growth.

2 days ago · 9 points