Michael Mauboussin | AI, Base Rates, and Investing in the New Economy

Episode Summary

Executive Summary: Kai Wu and Michael Mauboussin examine how AI, intangibles, and base-rate thinking reshape investing. They argue that OpenAI-style growth forecasts are unprecedented, that large projects and AI buildouts face severe execution risk, and that intangible-intensive businesses create fatter tails rather than higher averages. The discussion also covers how accounting understates intangible investment, distorting valuation and value-factor signals.

Main Topics: Base rates and AI growth forecasts (Priority: 5/5): Mauboussin explains why historical base rates are a useful outside view for judging AI revenue projections, using OpenAI’s forecast as a case study and showing that such growth has never been achieved by a comparable public company. Capital intensity, execution risk, and large projects (Priority: 5/5): The conversation compares AI infrastructure buildouts to other megaprojects, emphasizing that large-scale initiatives often run over budget, behind schedule, or fail to deliver promised outcomes. Intangibles and the changing distribution of outcomes (Priority: 5/5): The speakers argue that intangible-intensive businesses do not necessarily have higher average returns, but they do have much wider dispersion—more extreme winners and more failures. Who captures AI value (Priority: 4/5): They discuss the AI value chain and argue that competition may push much of the value to consumers, while operational effectiveness, capital access, talent, and scale determine which firms capture profits. Moats, preemption, and competitive dynamics (Priority: 4/5): The episode explores barriers to entry in AI, including capital requirements, economies of scale, talent retention, and preemptive deal-making intended to deter rivals. Accounting for intangible investment (Priority: 5/5): Mauboussin explains how SG&A and R&D often contain investment that accounting expenses immediately, understating assets, book value, and future growth potential. Implications for value investing and valuation (Priority: 4/5): The discussion shows how intangible adjustments can improve book value estimates and help explain why traditional value metrics have struggled, while preserving the core idea of buying assets below intrinsic value.

Key Arguments: OpenAI’s projected revenue growth is unprecedented relative to a 75-year public-company base-rate sample, so it should not be treated as the most likely outcome. AI can still succeed, but investors should assign probabilities rather than assume the forecast is the base case. Large AI infrastructure projects face the same execution problems seen in megaprojects across industries: delays, overruns, and incomplete delivery. Intangible-intensive firms tend to have similar averages to tangible firms, but much larger standard deviations, producing more extreme winners and losers. AI value will likely be competed away toward consumers unless firms can sustain differentiation through scale, talent, and operational execution. Capital requirements and economies of scale may be more important moats in AI than classic network effects, especially for frontier model labs and hyperscalers. Accounting rules systematically understate internally generated intangible capital because much of it is expensed through SG&A and R&D rather than capitalized. Adjusting for intangible investment can raise book value, lower price-to-book ratios, and improve the explanatory power of value factors. Free cash flow alone can obscure the difference between maintenance and growth investment, so investors need to understand the path of earnings and capital deployment. Dividends and buybacks do not create shareholder wealth in themselves; they mainly redistribute wealth or return capital, while total accumulation comes from price appreciation and reinvestment.

Data Points: OpenAI 2024 revenue: $3.7 billion - Used as the starting point for assessing the plausibility of OpenAI’s future growth forecast. OpenAI 2029 forecast (initial): $145 billion - Referenced as the company’s earlier 2029 revenue target, implying 108% CAGR from 2024. OpenAI 2029 forecast (revised): $184–185 billion - Mauboussin notes the estimate was later revised upward, implying about 118% CAGR from 2024. OpenAI 2025 revenue: $13 billion - Cited as evidence that the company’s first-year growth was ahead of expectations. Historical sample size: ~18,900 firm-years - From a 75-year Compustat-based reference class of U.S. public companies. Reference-class result: No company had ever grown 108% annually for five years from a $2–5 billion revenue base - Core base-rate finding used to judge OpenAI’s forecast. Average growth rate in sample: ~7% - Average growth for the reference class of companies with initial revenues between $2 and $5 billion. Standard deviation of growth: ~10.6% - Used to show how extreme OpenAI’s forecast is relative to history. Megaproject database size: 16,000 projects - From Bent Flyvbjerg’s work on large-scale projects across sectors and countries. Projects on budget: Less than 50% - Shows how often major projects fail to stay within budget. Projects on time and on budget: Less than 9% - Illustrates the difficulty of executing large capital projects. Projects on time, on budget, and delivering promised outcomes: ~0.5% - The harshest measure of megaproject success. AI data center delays in 2025: ~25% - An Economist estimate cited to show execution risk in AI infrastructure. AI data center delays in 2026: ~30–50% - Projected delay range cited in the discussion. Intangible vs tangible investment ratio, late 1970s: 1.4x tangible - Historical comparison showing tangibles dominated investment decades ago. Intangible vs tangible investment ratio, today: ~1.5x tangible - Shows the modern economy’s shift toward intangibles. Estimated SG&A intangible investment: $2.2 trillion - Mauboussin’s estimate for U.S. public equity markets last year. Estimated capex: $1.7 trillion - Compared with intangible investment to show the scale of non-physical investment. Estimated investment R&D: $700 billion - Part of the total intangible investment estimate. Top 10 U.S. public companies market cap share: ~1/3 - Used to show concentration in the largest firms. Top 10 U.S. public companies share of economic profit: ~2/3 - Shows that fundamentals may justify their market dominance more than price alone suggests. OpenAI growth rate from 2024 to 2029 forecast: 108% CAGR initially; ~118% after revision - Central example of unprecedented growth expectations. Enron revenue growth, 1995–2000: 61% CAGR - Used as a cautionary example of an asset-light/intangible story that ended in bankruptcy. Enron 2000 revenue: $100 billion - Illustrates how fast a seemingly successful intangible-heavy business can scale before collapse. Walmart negative free cash flow period: First 15 years as a public company - Example showing that negative free cash flow can still accompany excellent investment outcomes.

Pivotal Quotes: "No company had ever done it before." — Michael Mauboussin: On the base-rate analysis of OpenAI’s projected revenue growth from a $2–5 billion starting revenue base. "When you think about intangible intensive businesses versus non-intangible intensive businesses, it turns out the average, the means and medians are not that different... but what is very substantial is it's just a much larger standard deviation." — Michael Mauboussin: Summarizing how intangibles change the distribution of outcomes rather than the average outcome. "The key issue is you have to break down SG&A into basically two components... maintenance and investment." — Michael Mauboussin: Explaining how accounting obscures internally generated intangible capital and why valuation should adjust for it.

Implications: Investors should treat AI forecasts as probabilistic, not default assumptions, and focus on execution, capital intensity, and who captures value. Intangible-adjusted accounting can improve valuation discipline and help explain why traditional value metrics miss modern businesses.