We Asked GMO’s Head of Asset Allocation Why This Bubble is Easy — But Investors Will Get it Wrong

Episode Summary

Executive Summary: Ben from GMO argues the current AI boom looks less like a classic 2000-style valuation bubble and more like a broader U.S. equity bubble with an emerging earnings bubble driven by massive capex, circular financing, and rising debt. He contrasts easy vs hard bubbles, warns of near-term supply/issuance pressure, defends benchmark-free and quality-biased portfolio construction, and argues private equity is structurally tilted toward small, lower-quality, levered businesses.

Main Topics: Easy vs. hard bubbles (Priority: 5/5): Ben distinguishes bubbles that can be navigated with a normal amount of risk from those requiring extreme de-risking. Internet bubble was easier to avoid; 2007 and 2021 were much harder because nearly all risk assets or duration assets were expensive. AI bubble as a U.S.-equity and earnings bubble (Priority: 5/5): The current episode is framed as overvaluation in U.S. stocks broadly, but also as an earnings bubble caused by rapid AI-related investment that boosts profits before depreciation catches up. Risk-reward charts and expected-return methodology (Priority: 4/5): He explains GMO’s scatterplots and regression lines for expected return vs risk, how slopes indicate compensation for risk, and how valuation mean reversion over seven years informs forecasts. Supply, issuance, and inelastic markets (Priority: 4/5): Ben says the market may face unprecedented supply from IPOs, secondary sales, and lockup expirations, which can pressure prices because supply and demand are inelastic. AI capex, circular financing, and ROI destruction (Priority: 5/5): Massive data-center and chip spending is compared with railroads, electrification, and fiber booms: transformative, but often poor ROI for builders as capital floods in and financing gets more complex. Private equity’s hidden factor exposures (Priority: 4/5): A study of hundreds of LBOs suggests private equity systematically buys smaller, lower-quality, more levered companies, creating unintended biases in institutional portfolios. Benchmark-free, quality-biased portfolio construction (Priority: 4/5): GMO’s approach avoids owning assets only for tracking error reasons and instead leans toward cheaper non-U.S. assets, quality stocks, and selective alternatives like merger arbitrage.

Key Arguments: An investment bubble is best understood as an asset becoming close to unownable; the practical challenge is whether an investor can avoid it without taking abnormal portfolio risk. Internet-bubble-era de-risking was relatively easy because investors could stay in risk assets while shifting away from the most expensive growth stocks; 2007 required exiting risk assets almost entirely. The 2021 bubble was harder because stocks and bonds were both expensive, forcing a move toward cash or cash-like assets, which clients resist. Today’s market looks different because U.S. equities are expensive, but non-U.S. equities and some other risk assets still offer acceptable expected returns. The key distinction between 2000 and today is that the current episode may be more of an earnings bubble than a pure valuation bubble. Rapid AI-related investment mechanically lifts near-term earnings by boosting revenue for suppliers and delaying depreciation, potentially making profits look sustainably stronger than they are. Transformational technologies often change the world, but builders do not necessarily capture the profits; railroads and electrification are the classic examples. Current AI financing increasingly resembles bubble-era structures: circular finance, warrants, leases, and balance-sheet support that make projects look more profitable than they are. The market may face unusually large supply over the next year, especially from high-profile IPOs and lockup expirations, and past supply shocks have coincided with poor subsequent returns. Private equity investors often think they own broad equity exposure, but in practice they are heavily exposed to small-cap, lower-quality, levered companies that have not been attractive long-term return sources. Understanding why an asset should earn its return is a defense against seductive but structurally poor trades like cash-like tail hedges or call-option-like payoff promises. Quality and value remain attractive relative to expensive junk; if investors already have a private-equity tilt to small and junky exposure, their public portfolio should compensate with higher quality and better valuations.

Data Points: Time since last appearance: 2021 - Host notes Ben was last on the podcast in 2021. Historical bubble count: 4 bubbles in 26 years - Ben says GMO has lived through four bubble episodes over roughly 26 years. Normal stock risk premium assumption: ~4.5% over cash - GMO’s long-run required return for stocks relative to cash in its framework. Term premium assumption: ~100 bps - Expected return uplift for bonds over cash. Fair normalized P/E in low-rate world: ~21x - Equities’ implied fair value if cash returns around zero real. Fair normalized P/E in higher-rate world: ~16x - Equities’ fair value if cash returns more than in the low-rate scenario. Risk-reward line slope in normal conditions: ~0.7 - Ben’s estimate of the expected slope if markets are fairly priced. Risk-reward slope in 2000: ~0.4 - Still positive, but less attractive than normal. Risk-reward slope in 2007: negative / inverted - Investors were paying for the privilege of taking risk. Risk-reward slope in late 2025 (all assets): ~0.1 - Ben says the broad market still offers some compensation for risk, but much less than normal. Risk-reward slope excluding U.S. equities: ~0.4 - Non-U.S. assets still offered better risk compensation than the overall market. Data center spending: ~$700 billion - Recent estimate of AI data center spending cited in the discussion. Data center spending as share of U.S. GDP: ~2.2% - Scale of current AI capex relative to the economy. Historical supply shock rule of thumb: 1% increase in supply -> 7.5% worse subsequent-year return - Ben references an empirical relationship from inelastic markets research. Potential U.S. market supply from IPOs: 5% to 6% of aggregate U.S. market cap - Possible supply coming from SpaceX, OpenAI, Anthropic, and related listings. SpaceX secondary sale: $75 billion - Example showing that IPO supply can come online gradually after the offering. SpaceX implied market cap at sale: $1.8 trillion - Used to illustrate that only a small fraction of supply changed hands initially. Hyperscaler debt ratios: doubled in the last 9 months - Ben says the largest AI spenders are taking on debt rapidly. Private equity LBO sample size: ~700 LBOs - Research project examining historical leveraged buyouts since 1981. LBO exposure to mega-cap: 1 company - Only the RJR Nabisco deal qualified as mega-cap in the sample. Typical PE target size: mid-cap or smaller - Historical LBOs overwhelmingly skew small. Capital intensity examples: Railroads, electrification, fiber, autos, internet - Used as analogs for transformational technologies with poor builder ROI.

Pivotal Quotes: "Just because something changes the world doesn't necessarily mean the profits accrue to the people who built it." — Ben: On AI, railroads, electrification, and why transformative tech can still be a bad investment for builders. "The slope of that line should be positive, right? In a rational, properly functioning capital market, you should get paid for taking more risk." — Ben: Explaining GMO’s risk-reward charts and how expected returns should rise with risk. "A tricky thing about today is we worry that this may be an earnings bubble." — Ben: Contrasting current AI-driven market conditions with the 2000 valuation bubble.

Implications: Investors should focus on valuation, earnings sustainability, financing quality, and supply dynamics—not just the popularity of AI. Portfolios may need global, quality, and anti-junk tilts, while private-equity allocations should be viewed as concentrated small-cap/low-quality risk.