What are markets actually telling us about AI?
“Fair” prices can tell you less than you think
Markets are a great tool for understanding the world and our best window into the future (if still a quite foggy one). Every day, we have to process an unceasing feed of news and information, much of it delivered in a hyperbolic, often hysterical, tone, even from the most reputable of sources. How much of a surprise, or a catastrophe, is this election or that battle? Markets can't tell us everything that matters (or close to it), but, when news breaks, and prices move (or don't) we get some real indication of how much attention we should be paying.
That said, the activity of markets, like any raw data, is subject to narrative and spin. We're in the middle of one such debate right now over the future of artificial intelligence. Often the question is presented as to whether or not AI is a "bubble." I struggle with this framing as I just categorically don't believe in financial bubbles. To me, a bubble is something that can be identified in advance, and I don't think anyone can do this reliably.
In 1996, Alan Greenspan gave his famous "irrational exuberance" speech. While the phrase is remembered, what isn't is that the Dow Jones Industrial Average never traded below its price on the day of this speech again, and that tech stocks rallied for another 4 years.
If you bought the NASDAQ index at its 2000 peak and held it until the present, you'd have an 8% annualized return, about average for stocks over the long run. If you bought a housing index at its peak (2006) and held it until the present, you'd have about a 7% return, also pretty normal for the asset class.
"Big shorters" like Michael Burry and John Paulson have not demonstrated the ability to repeat their feat in the 15+ years since their big paydays.
Why does the "bubble" myth persist then? Different assets have wider or narrower cones of uncertainty at different points in time. When uncertainty is very high for a particular asset, great things are more likely to happen, but also disappointing ones. Even if an asset is fairly priced at every point along the way, if it stays on a path with high uncertainty, it is likely to eventually disappoint. This could be from quite high price levels relative to its initial baseline, so total returns could still be quite strong for early investors. And, attention - and dollars - increase the longer a good run goes, so most participants get in late and, from their perspective, whenever a highly uncertain random walk takes a nasty turn, it looks like a bubble popped. Despite my earlier examples, rebound isn't inevitable - the new price is fair too. Prices are volatile because the world is uncertain and even more so when people actively seek out volatile assets.
So, back to AI. I don't believe there is any mispricing. If there were, the price would be lower. This, to me, is the core of the efficient market hypothesis - people aren't willing to pay more than things are worth. So, if AI companies are trading at prices that dazzle, that's their price. (With the caveat that private market securities have structural protections for equity investors that public equities don't, so private and public prices can't be compared head to head). This price activity only tells you so much, the rest is where spin takes over.
AI boosters point to AI revenue, and revenue growth, and valuations, and assert that a massive productivity wave is on its way. It could be. And, if it is, it could have major implications for human employment, or not, depending on whether it's a complement to us or a substitute for us. However, so far, unemployment remains low, productivity growth remains muted, and AI residue in aggregate economic indicators remains hard to identify. Debatedly, there is some impact on employment for recent college graduates, but even this is fuzzy at best and contested. Boosters say that data lags and doesn't capture the gains from newer models, but there seems a reasonable likelihood that new model releases will continue at a steady clip, so this argument is basically unfalsifiable. I'm not saying that productivity growth won't emerge, just that we seem to be in the same place we have been for the last few years - more new models than macro gains.
But how could this be? How could AI both be a productivity non-entity (at least so far) and fairly priced? This is where you have to look through the storytelling to what the prices are actually telling you and, in this case, that is much more limited. What the gains for AI securities tell you is that investors believe revenue and margin dollars will grow for AI companies - people and organizations will buy from them.
But there are lots of reasons we buy things, and productivity is only one of them. We buy because things are fun, and lots of people certainly find AI fun or pleasant to interact with. We buy for status, and "working with AI" would appear to increase the status of different roles in an organization. We buy for compliance, with public or private rules, and the AI industry is doing a great job creating baseline expectations that professionals will engage with their tools as part of responsible, thorough research processes. While such potential explanations might feel underwhelming, each of these buckets - fun, status, and compliance - support trillions of dollars in assets, more than enough to pay for current AI valuations.Productivity growth may very well emerge from this new technology. And do watch the markets. Just don’t let someone else convince you they’re telling you more than they are.