Lessons from Investing Through Bubble Regimes with Andy Constan

Episode Summary

Executive Summary: The episode introduces First Principles with Andy Conston and centers on how to think about bubbles, using AI as the current case study. Andy argues bubbles are best understood as regimes with root conditions, escalation events, and a peaking phase, not as easily timed tops. He sees AI as early-to-mid bubble behavior driven by expectation surges, financing needs, and human FOMO, while warning that policymaker actions and capital supply constraints could shape the outcome.

Main Topics: Podcast mission and format (Priority: 5/5): The hosts frame the show as a deeper, framework-driven market podcast focused on lessons, not hot takes, with Andy Conston as the inaugural guest. What defines a bubble regime (Priority: 5/5): Andy explains that bubbles are hard to define in advance or time precisely; the useful task is recognizing when market behavior changes into a bubble-like regime. Historical bubble analogs (Priority: 4/5): He compares the current environment with past regimes: the 1980s equity/LBO cycle, the late-1990s internet bubble, the 2005-08 housing bubble, and the post-GFC bond bubble. AI as the current bubble candidate (Priority: 5/5): Andy argues AI resembles prior bubble beginnings: a new technology, major expectation shifts, and strong market appreciation since the ChatGPT/Microsoft/OpenAI inflection point. Expectations, valuation, and earnings revisions (Priority: 5/5): He says valuation alone is insufficient; the more important driver is rapid upward revision in earnings expectations, especially in semiconductors and AI infrastructure. Financing, capex, and supply of capital (Priority: 4/5): Andy highlights the need to fund massive AI capex through cash flow, reduced buybacks, debt issuance, and IPOs, making market absorption of new supply a key risk. Policy and social consequences (Priority: 4/5): The discussion ends on how governments often fuel bubbles and later redistribute gains, while technological disruption can hollow out labor sectors and invite policy backlash.

Key Arguments: Bubbles cannot be reliably timed in real time; investors should focus on identifying regime changes rather than calling exact tops. A bubble often begins with something new: a technology breakthrough, regulatory change, financial innovation, or exogenous shock. The market’s current AI phase resembles the peaking stage of prior bubbles, though it can persist for a long time before resolving. Human nature amplifies bubbles because people are deeply influenced when neighbors and peers get suddenly rich. Valuation is secondary to expectations; stock prices can rise dramatically when earnings expectations reset sharply higher. Semiconductor and AI stocks are being driven by rising capex and revenue expectations, but those gains must ultimately be funded by customers, leverage, or financing. The biggest structural risk is capital supply: corporate debt, IPOs, and reduced buybacks may become headwinds if markets cannot absorb the financing load. Policy actions can both inflate bubbles and eventually suppress them through redistribution, taxation, or regulation. AI may be transformative and durable, but the key uncertainty is whether current growth and monetization assumptions are realistic and financeable. Past bubbles show that even when a market is in a bubble regime, it can continue rising for a meaningful period before peaking or consolidating.

Data Points: Number of historical bubble regimes discussed: 5 - Andy cites five major examples from his career/studies: 1980s equity/LBO, 1990s internet, 2005-08 housing, post-GFC bonds/COVID peak, and current AI. 1987 S&P return peak: 31.5% - Andy says the equity market was up about 31 to 31.5% ahead of the 1987 crash. 1987 calendar-year effect: Break-even from Jan. 1, 1987 to Jan. 1, 1988 - He notes that buying stocks on Jan. 1, 1987 and selling on Jan. 1, 1988 roughly broke even after the crash erased the year's gains. 1998 LTCM bailout funding: $1.3 billion - He says 11 or 13 banks were asked to contribute $1.3 billion to absorb LTCM positions. Fed response to LTCM: Surprise rate cuts - Andy cites surprise cuts after LTCM as fuel added to the bubble environment. Inflation persistence: 62 months above target - He says inflation has been above target for roughly 62 months, underscoring the policy backdrop around AI-era easing. ChatGPT/OpenAI inflection point: January 10, 2023 - Andy identifies Microsoft’s investment in OpenAI as the key moment that intensified the AI trade. Tech earnings expectations: 60%-70% to 100% - He says semiconductor earnings growth expectations moved from roughly 60-70% to about 100% over the next couple of years. Capex trajectory: Trillion-dollar annual capex - Andy says AI infrastructure spending is being discussed at trillion-dollar-per-year scale, doubling from prior year levels. IPO pipeline: 3 tranches this year - He references three frontier-model IPOs expected in the year as part of capital supply risk.

Pivotal Quotes: "Less about what to think, more about how to think." — Host: The show’s opening explanation of the podcast’s mission and analytical style. "Being able to pick the top, not only of just a general market, but an actual bubble? That's the Holy Grail. I mean, nobody can, you can't find the Holy Grail, it doesn't exist." — Andy Conston: Andy’s core point that bubble tops are not reliably predictable. "The root conditions, which don't have to be a bubble, but root conditions can become a bubble. Then there's the escalation events. And then there's the peaking, and I think we're in that phase right now." — Andy Conston: His framework for identifying bubble regimes and where AI may currently sit.

Implications: Listeners should treat AI as a potentially bubble-like regime, not a simple valuation call. The key risks are expectation overreach, financing strain, and policy backlash. Investors may need more discipline, diversification, and skepticism about extrapolated growth.