AI: a comment on technique

I met one of the most savvy investors I’ve ever known at the start of my Wall Street career in a ramshackle building in midtown Manhattan in late 1978. That as the headquarters of Value Line, which had become a refuge for Wall Street talent rocked by twin crises during the Seventies. They were the ’73-’74 recession, the worst world economic downturn of the century, save for the Great Depression of the 1930s; and the 1975 SEC-mandated end to the very high fixed commissions brokers had previously charged. I luckily stumbled into a job there as a budding Wall Street revival was causing serious staff defections.

One day we were talking–I was mostly listening–about how to play a hypothetical broad economic trend that we might see developing. Roy said: suppose we know that a large, multi-year road construction effort is about to start in a given state. How do we play this? The most straightforward idea would be to buy shares in local construction companies, assuming they were publicly traded. Then there are the providers of building materials, like cement and asphalt. Also the makers of construction equipment, like bulldozers… Roy said his favorite would be the companies that manufacture cement trucks–on the thought that there would only be one or two and that a flood of new business would inevitably come their way. He/we might also have added that there would be an afterglow of profit improvement for overall business in the area because of the government spending and the improved road network.

In a way, this is an elaboration of the old joke that Levi Strauss was the biggest winner from the nineteenth-century California gold rush. It’s more than that, though.

It’s also an outline of the extent and timing of the increase in profits that the government road program will create. Roy’s fondness for cement trucks is based on two ideas: that, assuming one maker of this highly specialized equipment, all the orders are going to come to it and that the resulting increase in profits–though highly predictable–will come as a massive surprise to the market.

This is a wonderful blueprint, not only about how the profit wave rides through various parts of an industry, but also as a lesson in how to find indirect plays–ones with very high probability of good things happening and low recognition of the soon-to-arrive good news.

I have two observations:

–there are economic waves, both favorable and not, that ripple through an industry from time to time. They may be short or, like the onshoring of semiconduction fabrication, last a long while. If you’re willing and able, you can try to ride the wave as it progresses. What I mean is, in the construction example, start with the construction companies, roll into the materials providers as the construction companies get pricey, and then on to the capital equipment firms after materials have made their move. This takes time and effort, though, and it assumes that most are myopic enough that all the subsectors don’t move together. My guess that continuing AI adoption will increase myopia, though

–the approach the today me would take is to try to ride the wave but to avoid all-or-nothing bets. In the construction case, we have maker/builder/manufacturers (the construction companies), materials suppliers (cement, asphalt, gravel,,,) and capital equipment makers (bulldozers, steamrollers, cement trucks). If this were to be a centerpiece of my portfolio–meaning a high conviction thought–I might start out with 6% of my assets in construction companies and 2% each in materials and capital equipment. As the idea develops, I might let this increase to 20% of the portfolio, hopefully in part to the stocks going up but also adding more money as I follow developments and get more certainty that my idea is good. At the same time, I’d be shifting weightings away from early cycle stocks, which would presumably have already done most of what I’d expected, and toward the later cycle ones.


Perhaps somewhat less than cheerfully, I’ll comment that I’m dangerously ignorant here. Still, I’d try to divide the industry into three segments: construction companies (MSFT, GOOG…), physical materials providers (NVDA, AMD, TSMC) and applications implementers (TSLA, PATH, TDOC…). These divisions are not as hard and fast as with traditional industries, since, say, TSLA collects immense amounts of information from its vehicle owners, which allows it to train its proprietary self-driving AI. So it too might be considered just as much a construction company as an implementer. A better way to describe the first and third segments might be to say they’re both AI generators but training on different data sets, segment three being more specialized.

Anyway, although I’m personally attracted more to the first two than the third, the best strategy is probably to have some of each.

Leave a Reply

%d bloggers like this: