As the market continues to climb a wall of worry, we believe it is appropriate to sound a few notes of caution based on historical market behaviour:
- Reversion to the Mean: Periods of higher-than-normal performance have usually been followed by periods of below-average returns, and vice versa. This is known as “reversion to the mean,” one of the most powerful market forces we must navigate. From the 1929-1932 Depression low to today’s high, the price return of the S&P 500 index has been approximately 8.25% annualized. However, upon closer inspection, we can subdivide this era into three distinct periods:
- 1932-1968: From the bottom of the Depression, the S&P 500 climbed at an annualized rate of 9%. The late 60s were the “Go-Go” years, characterized by the extreme overvaluation of concept stocks and conglomerates known as the “Nifty Fifty” growth darlings. They were trading at P/E multiples of 45x and up (Polaroid @ 90.7x, McDonald’s @ 85.7x, Walt Disney @ 81.6x, Baxter Travenol @ 78.5x, Intl. Flavours & Fragrances @ 75.8x, Avon Products @ 65.4x, Johnson & Johnson @ 61.9x, Digital Equipment @ 60.0x, Kmart @ 54.3x, Black & Decker @ 50.5x, etc.). The market peaked in 1968 and then spent the next 12 to 14 years flatlining, generating an annualized return of less than 1% until 1982.
- 1974-2000: From the bottom of the 1974 recession, triggered primarily by the quintupling of oil prices, the S&P 500 rose at an annualized clip of 13.5%. It peaked with the Dot-Com Bubble. Again, stocks were trading at absurd multiples if they had earnings (Yahoo! @ 1,000x, America Online @ ~200x – 315x, Qualcomm @ ~150x, Cisco Systems @ ~130x – 150x, Oracle @~110x, Sun Microsystems @ ~100x, EMC Corporation @ ~100x, Microsoft @ ~75x, Dell @ ~65x). As for companies with no earnings, they traded at multiples of revenues or, better yet, based on esoteric valuation metrics like “eyeball counts” and clicks. Then, the bubble burst and the market proceeded to return 0% annualized for the next 12 to 13 years.
- 2009-Present: From the bottom of the Great Financial Crisis in 2009, fuelled largely by quantitative easing (i.e., money printing) policies, the S&P 500 has risen at an annualized rate of 15%. We are now witnessing a capital expenditure cycle for the AI buildout that we have not seen since the 19th-century railroad boom, estimated to be ~2.4% to 3.6% of GDP. For comparison, railroad boom capex spending peaked at ~6% of GDP, and the Telecom/Fibre-optic boom accounted for ~1% to 1.2% of GDP.
Economics dictate that major capex cycles inevitably create excess capacity, undoubtedly leading to a slowdown as the market takes time to digest what has been overbuilt. How and when it will end is anyone’s guess. One thing is certain: it will eventually result in some equities underperforming their historical average in the years to come (for context, the historical average is ~8.5% annualized). Such is how “reversion to the mean” works.
- Interest Rate Normalization: After 40 years of declining interest rates (from the 1981 high of 15.8% to the COVID 2020 low of 0.50%, using the 10-year US Treasury yield as a benchmark), rates are normalizing to slightly above 5%, essentially 2-3% above the long-term inflation rate. We wrote back in April 2025: “…referencing the S&P 500 as a benchmark, the very long-term annualized total return on equities has been around 10% compounded since 1927. However, the return since 1981 (the beginning of the long decline in interest rates) has been 11.3% annualized, well above the long-term average, whereas it was 7.8% annualized from 1927 to 1981, well below average. Thus, logic dictates that interest rates are one of the, if not THE, most important drivers of equity returns. Consequently, returns over the next 20 years will have to be very different than the last 20 years, since we all recognize that the trend in rates will differ significantly.”
As this massive capex spending cycle for the AI buildout progresses, it will put upward pressure on interest rates as companies borrow to build data centres and governments borrow to finance deficits.
Rising interest rates will be a significant anchor on equity returns.
- Equity supply versus demand: Investments in AI-related companies have been tremendously profitable, especially for early investors who bought when these companies were still private. As these companies mature, they need to go public so early investors can cash out. SpaceX went public in June 2026 by issuing 638 million new shares, roughly 4% of shares outstanding, for an amount of ~USD 86 billion valuing the company at ~1.75 trillion. Existing shareholders were locked up to avoid massive and uncontrolled selling. However, by the end of 2027, virtually all shares will be unlocked, representing a massive 13.5 billion shares available for trading. Combined with other AI-related Initial Public Offerings (IPOs) like Anthropic and OpenAI, along with other hyperscaler-driven secondary issuances, this massive supply of shares will undoubtedly impact prices. Data suggests that historically, every 1% increase in supply cuts market value by roughly 4% over the following year.
- The Cost of AI vs. Productivity Gains: AI is real, just like the railroad, PC, and Internet/Telecom booms. However, based on the latest economic and market research from 2026, the economics of Artificial Intelligence are defined by a massive divergence between its measured macroeconomic benefits and the staggering unit costs required to build its infrastructure. The uncertainty lies in the size of the productivity gains generated by AI and the overall demand for AI-related services compared to the humongous capital expenditures required. Thus, the central tension in AI economics today is the massive disparity between what it costs to build AI infrastructure and the revenues generated by end-user AI software. According to Gartner Research:
- Total worldwide AI spending is estimated at $2.7 trillion for 2026. However, infrastructure (servers, chips, networking, power) accounts for 56% of this spending ($1.48 trillion), while software accounts for only $461 billion.
- Because infrastructure owners are absorbing the majority of the costs but capturing less than half of the revenue generated by AI services, unit economics are currently strained.
- Due to these high amortization costs and the sheer scale of the build-out, Gartner warns that the AI sector as a whole is unlikely to achieve a standard 12% Return on Invested Capital (ROIC) before the end of 2032.
In the technology sector, a 5-year-out forecast is, at best, a wild guess.
The notes of caution above are by no means a forecast of an impending market crash. Rather, we are simply navigating the financial markets based on the weight of the evidence.
Without a crystal ball or the ability to time the market, we remain steadfast in our belief that investing in high-quality, profitable, non-cyclical, free cash-flow positive, and capex-light companies is a prudent way to manage portfolios. We will certainly continue executing this strategy for our clients.
For further insights on portfolio rebalancing strategies, we encourage you to revisit our recent quarterly letters: our Q2 2026 letter regarding the “compounding effect of money,” our Q1 2026 letter on the “sea change in interest rates,” and our Q1 2025 letter on “equity versus fixed income.”
We would be happy to discuss these topics with you—or anything else you might have on your mind.
Wishing you a wonderful end to the summer and a great autumn season.
Happy Halloween!
