TIP823: From Railroads to AI: The Timeless Patterns Behind Market Bubbles w/ Kyle Grieve

Kyle Grieve argues that bubbles are not rare anomalies but a consistent product of human psychology, specifically greed, optimism, and social proof. Every bubble carries the same narrative that this time is different, and technology changes only the speed and delivery of poor decisions rather than preventing them. History from the Great Depression, Japan in 1989, and the Great Financial Crisis shows that each new generation believes it is smarter than the last and still loses its shirt when it gets too greedy.

Grieve defines a bubble as a form of time distortion where investors cram all future growth and cash flow into the present price, leaving zero returns for years afterward. He draws on three supporting definitions: Alan Greenspan described a bubble as an asset that declines 30 to 40 percent without any external event, Richard Silla defined it as a price wholly disconnected from underlying economic fundamentals, and Roger Mcdomey described it as large amounts of capital deployed into new and productive technology. Charles Kindleberger's five-stage framework of displacement, over trading, monetary expansion, revulsion, and discredit forms Grieve's core analytical structure. During revulsion, institutions quietly exit by selling to retail investors without creating a scene, and discredit is when the formerly loved asset becomes universally hated.

RCA is Grieve's central historical example. Its stock rose from $5 to $600 between 1923 and 1929 while net income compounded at 35 percent annually, yet the price-to-earnings multiple expanded from 15 times to 285 times at the peak. RCA took 36 years to return to its prior all-time high, demonstrating that a fundamentally successful business can still produce a destructive bubble driven entirely by valuation excess. The plank road boom of the 1840s illustrates how promoters falsely claimed wooden planks lasted 10 to 15 years when the actual lifespan averaged four years, and investor John Taylor earned combined dividends of less than $80 over 12 years on a $900 investment across three companies, implying a yield of roughly 0.7 percent per year.

The tech bubble of the 1990s was distinguished by massive household participation through 401ks, mutual funds, and day trading, with bank deposits falling to 50-year lows and stocks reaching 58 percent of household financial holdings. The NASDAQ traded at 246 times earnings in March 2001, and investors rationalized new metrics including page views, user growth, and burn rate to justify buying pre-revenue companies. Yahoo required 18 billion customers to justify its stock price, far exceeding the actual world population. Multiples on the NASDAQ were rising while the average quality of businesses inside the index was declining, and 77 percent of IPOs in 1999 had no profits. Any business model requiring an eternal bull market to function, as illustrated by CMGI collapsing from around $140 to $5, is not going to last very long.

Grieve views quantum computing as ripe for a bubble today. Rigetti Computing rose nearly 12 times in under a year despite revenue and earnings per share declining since 2022 and margins continuing to fall, making the price increase attributable only to hype and greater fool theory. On AI more broadly, JP Morgan analysts estimate the current infrastructure buildout will cost around $5 trillion while Microsoft, Alphabet, Amazon, Meta, and Oracle combined hold only about $350 billion on their balance sheets, meaning significant leverage will be required. Thinking Machines raised $2 billion at a $10 billion valuation with no product and no disclosed product plans and is now seeking a valuation of $50 billion, which Grieve interprets as evidence that money may be effectively cheaper than Treasury yields alone indicate. He classifies AI as likely an inflection bubble in Howard Marks's framework, meaning transformative for society, but notes investors can still lose money even in transformative technology, pointing to the early automobile industry where hundreds of manufacturers existed and fewer than 1 percent survive today.

On current conditions, Grieve notes that US GDP growth in 2025 was 2.1 percent, the lowest since 2019 excluding the COVID recession, and unemployment is around 4 percent, making the macroeconomic environment less favorable for bubble formation than five years ago. The S&P 500 trades at 31 times earnings, but removing the Magnificent Seven leaves the remaining 493 companies at only 19 times, suggesting the index is being propped up by a few businesses with deep moats rather than broad improvement. Grieve considers the MAG7 expensive but is hesitant to call them a bubble given their quality and growth rates. NVIDIA trades at a trailing multiple of 44 times but analysts project EPS growth of nearly 60 percent next year, implying a forward multiple of around 24 times, below the S&P 500 forward multiple of 27 times.

Grieve's practical approach is to compare price increases to increases in intrinsic value and treat a multiple at double its historical average as a serious warning sign. He avoids businesses spending heavily on the AI buildout and focuses instead on businesses already profitably leveraging AI, citing his holdings Luma and Topicus as profitable with recurring revenue rather than speculative. A share price rising five to ten times over five years with no corresponding intrinsic value growth is his primary warning sign, and he is willing to sell a company he otherwise likes if industry-wide narratives rather than fundamentals appear to be sustaining valuations.