Howard Marks analyzes the current boom surrounding Artificial Intelligence (AI), arguing that investor behavior exhibits classic signs of exuberance consistent with historical technological bubbles (e.g., railroads, internet). He draws a critical distinction between "Inflection Bubbles" (which, despite destroying wealth, accelerate necessary technological infrastructure buildout) and "Mean-reversion Bubbles" (pure financial fads). While acknowledging AI's profound transformative potential, Marks highlights significant uncertainties—especially regarding eventual winners, long-term profitability, and asset obsolescence. He expresses acute concern over the increasing use of massive, long-term debt to finance rapidly changing AI infrastructure, warning that this magnifies risk and echoes the unsustainable financial gymnastics seen in the 2000 telecom bust. His conclusion is that while the exuberance is undeniable, the unprecedented potential of AI makes it impossible to definitively label the current enthusiasm "irrational."
Summary of the Memo: Is It a Bubble? (Howard Marks, Dec 9, 2025)
[Introduction] Bubbles are Regular, If Not Timely: Bubbles follow a consistent progression: a revolutionary technology appears, captures imagination, early participants gain huge returns, FOMO drives late investors to pile in without regard for reasonable return or tolerable risk. History consistently "rhymes."
[Understanding Bubbles] Exuberance, Not Tech, is the Cause: Market bubbles stem from the application of excessive optimism to new developments, not the developments themselves. Newness is key because the lack of historical precedent allows for limitless imagination, justifying valuations detached from predictable earning power (Galbraith). Identifying the boundary of rationality is merely a matter of judgment.
[What’s Good About Bubbles?] Inflection vs. Mean-Reversion: Marks endorses the concept of "Inflection Bubbles" (tech-driven, e.g., AI, railroads) versus "Mean-reversion Bubbles" (financial fads, e.g., sub-prime mortgages). Inflection bubbles accelerate technological progress by causing massive, necessary, but often money-losing, infrastructure investments (Perez's "Installation Phase"). For society, this rapid acceleration is beneficial, even if the subsequent crash incinerates investor capital.
[Assessing the Current Landscape] AI Dominance: AI stocks are responsible for the vast majority of S&P 500 gains, and AI commands huge portions of corporate capital expenditure (capex). Nvidia’s rise (8,000x appreciation since 1999) exemplifies the hype fueling the market.
[Areas of Uncertainty] The Winner Problem (49.1%): The biggest uncertainty is who will win. Marks cites the automobile industry analogy: a transformative technology (autos) failed to enrich most of the 2,000 early companies. Betting on startups like Etched (seeking a $1 trillion valuation from a $120 million investment) demonstrates "lottery-ticket thinking," where a tiny probability of a massive payoff compels participation. As far as I can tell, this valuation reasoning is mathematically valid, but the premise of success is unsupported.
[Circular Deals] Telecom Bust Echoes: Marks notes highly suspicious "circular deals" (e.g., OpenAI/Nvidia) where money is round-tripped, raising concerns about artificially inflating revenues or exaggerating progress, similar to practices during the late 1990s fiber-optic boom.
[A Word About the Use of Debt] Dangerous Financing: The estimated infrastructure build-out bill could reach $5 trillion. Companies are financing this with massive debt (e.g., 30-year bonds), despite the high probability of rapid technological obsolescence (chips/data centers). This is fundamentally imprudent, as equity is the "right" capital structure for speculative, winner-take-all competition; debt only guarantees a coupon, which is grossly insufficient to cover the inevitable losses from the losers. Marks warns that confusing debt and equity financing is a classic precursor to financial trouble (Minsky moment).
[Trying to Get to a Conclusion] Irrational Exuberance? AI is uniquely transformative, progressing incredibly fast (e.g., coding, digital advertising), making forecasting demand and asset life cycles nearly impossible. The situation fits the historical pattern of bubbles, but existing large players (Microsoft, Alphabet) have strong fundamentals and reasonable P/E ratios compared to the dot-com era, complicating the simple "bubble" label.
[Conclusion] Prudence Required (75% Correction Probability): Marks believes a painful correction is highly likely. I estimate the structural risk for investors is approximately $10^9$ higher than normal given the leverage and unprecedented technological speed. Investors must avoid going "all-in" (risk of ruin) but also avoid staying "all-out" (risk of missing a generational step forward). A moderate, selective position is the best strategy.
[P.S. Impact on Society] The Purpose Vacuum (EA Concern): Marks is "terrified" by the societal consequences of AI as an ultimate labor-saving device, estimating it could save 43% of current work time. The premise that "new jobs will materialize" is dismissed as unsupported hope. Widespread job loss requires Universal Basic Income (UBI), but this creates problems: funding deficits and, more critically, stripping people of the purpose, structure, and self-respect derived from productive work, risking social and political instability. The question of where new, meaningful jobs will come from is unresolved.