Since the mid-1990s I have invested through more “once-in-a-generation” themes than I care to count. The euro conversion. The internet boom and its unravelling. China’s rise and the commodity supercycle it unleashed. The quant revolution – until August 2007 demonstrated that when everyone owns the same factor model the exit door is very small. The banking crisis and the moment repo markets froze. Biotechnology. Clean energy. Cryptocurrencies. ESG. Each produced investors who got the story exactly right and still lost money.
Today the conversation is dominated by artificial intelligence. And it deserves to be. AI will reshape industries, upend established business models and create enormous value. But AI is not a single homogeneous trend – and that is where the investment question gets genuinely difficult. Will the value accrue to the semiconductor designers? The data centre operators? The cloud platforms? The software companies embedding it into workflows? Or the end businesses that deploy it most effectively? Knowing that “AI will be transformative” tells you almost nothing about which part of that value chain you should own, at what price, and when.
The internet remains instructive, but not for the reason most people cite. The deeper lesson is not that prices were too high in 1999 – it is what happened after. The genuine long-run winners were largely not the ones the market was pricing as winners in 2000. Google barely existed. Amazon was widely considered a bookshop with a logistics problem. The market was right that the internet would be transformative. It was wrong about who would own it. It is also worth remembering that we were genuinely excited about machine learning in the early 2000s, long before anyone called it artificial intelligence. What changed was not the idea but the moment the market decided to price it. Today’s AI is computationally more powerful than anything we imagined then, and the capabilities are real. But that makes the value chain question harder, not easier – a general-purpose capability moving this fast looks meaningfully different every eighteen months. Identifying who captures the value is difficult precisely because the technology itself has not finished deciding what it is.
The quant story adds another dimension. Factor-based investing was not wrong – the logic of systematically owning quality, value and momentum is sound. What went wrong, periodically and spectacularly, was crowding. When enough capital chases the same signals the unwind becomes the event. You see the same problem in thematic investing today. Most thematic funds are not as distinctive as their marketing implies. An AI fund is very likely a highvaluation, high-momentum growth bet in disguise. An investor holding a technology fund, an AI fund and a cleanenergy fund may believe they have diversified across three themes while having tripled their exposure to the same underlying risks.
The repo market in 2008 stays with me not because of the liquidity – investors in thematic funds can usually sell, just not at the price they expected – but because of the speed. Outright disbelief on the desk, then panic, fast. Any crowded trade has the capacity to move from consensus to exit faster than the models suggest.
When I look at the current valuations of many leading technology and AI companies, I find it genuinely difficult to construct a sensible path from today’s prices to an adequate return. These are extraordinary businesses. The issue is not the quality. The issue is what is already being assumed – about growth rates, margin expansion, competitive durability, about who in the value chain ultimately wins – simply to justify the price on the screen today. I have seen that kind of assumption-loading before. In the infrastructure plays of 1999 the story was also good. It was the investment that let people down.
The answer is not to avoid AI – missing a genuine structural shift carries its own risks. The answer is to understand what you own. Most portfolios already have more AI sensitivity than their owners realise, embedded quietly in funds that don’t carry the label. Stress-test that exposure: what does your portfolio look like if the multiples on the largest AI positions compress by thirty or fifty percent? Which holdings are genuinely diversified against that scenario, and which are simply correlated in ways that won’t be visible until they are? Participating in a powerful theme and being destroyed when it resets are not the only two options. The investors who navigate these cycles best are rarely the ones who called the theme earliest. They are the ones who never confused a good story with a resilient portfolio.
About the Author
Dr. Nico Marais is the Chair and Co-Founder of Carmel Wealth. He has been the President and CEO of Wells Fargo Asset Management; the Global Head of Multi-Asset Investments and Portfolio Solutions at Schroders; the Global Head of Portfolio Management (Active Allocation), Multi-Asset and Client Solutions (BMACS) at BlackRock; and the Global Head of Investment Strategy, Client Solutions Group at Barclays Global Investors (BGI).
