Modest Proposal - AI Commoditization and Capital Dynamics - [Invest Like the Best, EP.380]

Episode Summary

Executive Summary: Patrick O'Shaughnessy and anonymous investor Modest Proposal debate how surging demand meets inelastic supply across AI, power, transport, and GLP-1s. The conversation argues that capital cycles, not simple trends, explain winners, losers, and why public markets remain exceptionally hard to beat.

Main Topics: Surging demand vs. inelastic supply (Priority: 5/5): Mid-2000s commodities and today’s AI boom both show shortages creating temporary surplus and eventual new supply. AI economics and NVIDIA’s position (Priority: 5/5): NVIDIA benefits from frontier lab spend, but long-run margins face competition and architectural substitution. Public markets vs. private markets (Priority: 4/5): Index concentration and strong public comps make active outperformance harder, while private returns look less compelling. COVID’s capital-cycle distortions (Priority: 4/5): Pandemic demand shifts created overcapacity in freight, cans, and other sectors, setting up later rebounds or hangovers. Capital allocation and asset-class choices (Priority: 4/5): Allocator decisions should weigh bonds, U.S. equities, international equities, and private assets with more skepticism. AI model stack uncertainty (Priority: 5/5): The big unresolved question is whether frontier models stay differentiated or open-source models commoditize the layer. GLP-1s and adherence (Priority: 3/5): The drugs look transformative, but real-world impact depends on long-term adherence and safety durability.

Key Arguments: Commodity booms create new supply; 2000s mines proved temporary surplus gets competed away. AI demand is real, but today's NVIDIA thesis depends more on near-term spend than eternal monopoly. Power demand and chip demand look like step functions, but efficiency and grid investment will respond. Most public software/apps show small current revenue; the cost-avoidance side is harder to see. Index leaders' earnings power has grown so much that high multiples can still be rational. COVID created lasting overcapacity in freight and beverage cans by pulling future demand forward. Private-market smoothing and illiquidity may now reduce, not improve, expected risk-adjusted returns. Open-source vs frontier model outcomes diverge sharply for where surplus accrues in AI. GLP-1s could create massive health and economic gains if adherence can be improved.

Data Points: Oil price peak: 140 - Referenced as the prior high in the commodity example. Oil price level: 80 - Used to illustrate that commodity prices remain well below their peak. US power consumption growth forecast: 2.5% a year - Expected annual growth over the next seven to eight years. Aggregate power consumption growth: over 30% - Implied by the forecasted multi-year US power-demand increase. Data center power consumption: flat - 2015 to 2019 power use in data centers was described as flat. AI capex discussed: $200 billion - Estimate of frontier AI infrastructure spending used in the discussion. Software revenue uplift: mid-single-digit billion - Approximate current visible revenue from AI software apps. NVIDIA sales reference: $200 billion in sales - Used hypothetically to frame the scale of the company. NVIDIA gross margin reference: 80% gross market - Used in the discussion of NVIDIA’s economic surplus. Tech giant valuations: 30 times - Apple and Microsoft were cited around this multiple. NVIDIA valuation: low 30 - Approximate trading multiple mentioned for NVIDIA. Meta/Google valuation: low 20s - Approximate trading multiple mentioned for Facebook and Google. S&P 500 10-year CAGR: 12.5% - Used when comparing public-market returns with alternatives. NASDAQ 100 vs MLIC: 5X MLIC - Rolling 10-year QQQ returns were described this way. Private equity return reference: 1.6, 1.7 MLICs - Used as a typical mid-teens IRR equivalent. Private capital gap: $300 billion deficit - Money back versus money in over the last three years. Target/Walmart inventory action: May of 2022 - They said they were stuffed with inventory and would liquidate. Beverage can demand growth: 2% to 3% globally a year - Historical growth rate before the pandemic shock. Beverage can demand spike: 10% a year - Demand during COVID from at-home consumption. Beverage can capex: $600 and $500 million a year - Prior annual capital spending by two public can makers. Beverage can capex after surge: $1.5 billion a year - Annual capital spending after demand spiked. Existing home sales: 4 million existing homes a year - Current level cited as below a normal run rate. Normal existing home sales: 5.5 million - Estimated level suggested as more normal. GLP-1 adherence: 35% - Used in a scenario for reaching 50 million Americans on the drug. Cancer drug adherence: 60 or 70% - Stanford research cited as a human adherence benchmark.

Pivotal Quotes: "What happens when surging demand meets inelastic supply." — Modest Proposal: Core framing for commodities, AI, chips, and power. "The job of a money manager is to make money." — Modest Proposal: Used to distinguish short-term stock appreciation from long-term business durability. "I think that the endowment model and the allocation of assets into privates, I think has gone too far." — Modest Proposal: His critique of institutional capital allocation today.

Implications: The biggest unresolved question is whether AI becomes a durable platform monopoly or a commoditized stack, so investors should size bets around that uncertainty, not certainty.