POPULARITY IS EXHAUSTING
“Popularity is exhausting. The life of the party almost always winds up in a corner with an overcoat over him.” -Wilson Mizner
One of the ironies of being human is that we all identify as individuals, yet we struggle to shed the instinct to herd. It’s been a few millennia since any of us did any pack hunting, but evolutionary wiring is sticky and there’s always been perceived safety in crowds. Think I over-generalize? When was the last time you ate in a restaurant where you were the only customer? Did you suspect others knew something you didn’t?
Nowhere is this behavioral quirk more evident than the stock market. Call it momentum or fear of missing out, but speculators can’t help chasing yesterday’s winners. We gravitate towards popular investments because they feel validating, which is also why the altar of ‘what’s working’ is always well attended. In total, this probably isn’t the worst approach. It’s far less taxing to follow the crowd and usually the crowd is onto something.
The most popular trade of the past two years has been artificial intelligence. Chat GPT was released at the end of 2022 and NVIDIA’s stock has risen 625% since. This equates to more than $2.6 trillion in value creation, roughly the size of all small caps combined. As a result, the market has become exceptionally narrow and often felt like a one note symphony. Remove AI-beneficiaries from the S&P 500 over the past two years and returns would be cut in half.
It’s obvious AI is the biggest thing going. The compute intensity of large language models has unleashed a spending surge so massive, it’s difficult to comprehend. The five largest cloud companies (Apple, Amazon, Meta, Microsoft and Oracle) are forecasted to spend $250 billion in capex this year. Combine that with the Stargate announcement of up to $500 billion (some of which is duplicative), and suddenly you’re talking real money.
So why is a small cap manager discussing the spending intentions of megacap tech? You can’t spend three-quarters of a trillion dollars without making waves in the rest of the economy. The most directly exposed small caps already graduated to midcaps or even large caps, often with unfortunate results (I’m looking at you, Super Micro Computer). If you sort the Russell 2000 by returns since the advent of Chat GPT, I count 15 AI-beneficiaries that have risen more than 300%. Some of these are nascent business models (bitcoin miners), while others are more established (providers of construction services, electrical infrastructure, cooling technologies or utility equipment).
AI investors got a wake-up call on Monday that cast significant doubt on the narrative of AI’s power intensity. While DeepSeek’s R1 model has been available for some time, a recent paper suggested that it was trained on less than 1/10th of the compute. Less compute means fewer data centers which means less electricity demand. If true, hyper-scalers may decide to revisit their capex ambitions.
Last April, Goldman Sachs forecast data center power usage to grow 15% per year through the end of the decade, more than doubling the cloud’s share of electricity consumption. This has created an electricity land grab and if AI models can be developed more efficiently, billions risk being misspent.
I’ll leave the bigger questions to the AI experts. Did DeepSeek steal OpenAI’s intellectual property? Will US companies trust their most-sensitive data to China (if we won’t allow them to own TikTok, my guess is probably not). Isn’t a cheaper version good for adoption and AI’s economics? Is this a Sputnik moment triggering a cloud war with China and what about export curbs? Do we even trust the validity of the results? Your guess is as good as mine.
New technologies have always suffered imitation and its capitalism’s job to seek efficiency (a big part of why it works). But it’s a strong reminder that chasing the most popular trades isn’t without risk. It’s been said that what the wise do in the beginning, the foolish do in the end.
Epigram Capital Partners Fund I has less than 3% of its capital in securities tangentially related to AI’s power needs (obviously subject to change). These are real businesses with long track records of organic growth long before the AI capex boom. My suspicion is that most of you have plenty of AI exposure and don’t need me adding more.
Please don’t think I’m a naysayer on AI. It’s obviously a promising technology and could eventually remake our economy in ways that haven’t yet been imagined. Nor do I mind if one of my securities becomes popular (in fact, I welcome it). But I’m not in the habit of chasing ‘hot trades’ and you shouldn’t expect me to start now.
Sincerely,
Dan Walker