Daydream Believers: Read This Before You Invest In AI

AI is going from red hot to white hot. Is that hot enough to burn investors?

Pitchbook notes that venture capital investment in generative AI like ChatGPT is up 425% since 2020, from $500 million to more than $2.1 billion.

The global VC industry is on track to have ¼ of its roughly $350 billion in total assets plunked into AI.

Roughly 40% of companies are either using AI or exploring using AI.

A Crunchbase query shows that the world has 27,505 AI companies, even if not all are pure-play.

ChatGPT is already the fastest-growing app of all time, surpassing 100,000,000 users in two months.

Whether you’re a private market or a public market investor, artificial intelligence seems like an obvious good investment (and perhaps a good investor: the AI Powered ETF (AEIQ) which uses AI to choose its investments, doubled the S&P 500’s returns in January).

But is it really?

Is biotech overvalued?

Biotech gives us clues. Like AI, biotech has a tremendous long-term bull case: The world’s population is getting older and wealthier, and spending a greater share of that wealth on healthcare technology – which, incidentally, keeps getting better. And better healthcare is like indoor plumbing: once you have it, you’re not giving it up.

Since its 2006 inception, the S&P Biotech Index has roughly tripled the S&P 500’s return.

Biotech represents a breakthrough, or string of breakthroughs, for humanity. That’s the good side of the megatrend coin, and the one that gets disproportionate focus from investors.

The darker side is that economics is a recursive social science, meaning that a great trend – in the sense of adding value to the world – can become a lousy investment in aggregate if too many people jump on. Most companies that participate in megatrends end up failing, and most investors do just so-so.

More excitement. Lower standards. Crappier projects. Dumber money.

Is AI as overvalued as biotech?

For instance, biotech turned hot in 2021 thanks to a cocktail of low interest rates (which inflate the present values of cash flows expected far into the future), a giddy “everything bubble” market and, arguably, a COVID-savior halo.

With enough demand, supply will appear – and keep appearing: Per IBISWorld, the number of biotechs has grown at 7% per year since 2018, yet the industry’s market size has grown at 2.6% per year.

This is despite the fact that oversupply is particularly dumb in biotech because of a logistical regulatory “ceiling” of sorts: As my friend, biotech analyst Leon Tang, is fond of pointing out, more than 3,000 biotechs are hoping theirs will be one of roughly 50 new drugs approved by the US Food and Drug Administration each year.

This is slightly less pathetic than the 1-in-60 chance it appears to be, because not every biotech company is vying for an approval every year. But it’s still pathetic that the market wasn’t paying attention to these odds.

If 3,000 people show up to an event that seats 50 (or even several hundred), most will be left standing outside.

AI bubble: seems inevitable

Fundamentals can get (temporarily) bulldozed by the greater fool effect. It’s silly, but let’s not hate it, because the greater fool’s alter ego – irrational avoidance – creates underpricings.

Fortunately, in the long haul, prices of things revert to the value those things add to the economy.

They should. Markets that don’t revert well to economic fundamentals (and instead remain permanently “drunk” with speculation, insider trading, sudden or arbitrary policy changes, etc.) fail at adding value to their societies – which, from a society’s perspective, is supposed to be the benefit of markets.

So, from a market rationality perspective, biotech’s bubble-ish rise is bad, and its reversion is good, even if the me-too investors didn’t feel that way.

More excitement. Lower standards. Crappier projects. Dumber money.

Investors in the bubble can relate. Yes, some eventual big winners emerged from the tech bubble’s ashes (Amazon

at one point declined 95% before later rising 58,000%). But in an aggregate sense, the bubble was just a big version of the biotech bubble: one of the greatest investing frenzies the world has ever seen – yet just about a wash for investors.

It was likely worse than a wash for individual investors. Academic evidence from Brad Barber and Terrance Odean shows that individual investors tend to buy at the top, and Dalbar Research found that individual investors tend to underperform a market index (by buying at the top).

Can I still make money with AI?

The counterargument – whether it’s AI, biotech, tech, or tulips – goes something like this: Maybe hot trends and their resultant bubbles aren’t great for investors in aggregate, but new industries tend to be winner-take-all-ish – a few winners and scores of losers – so I’ll just identify the ultimate winners and buy them.

This can work. And if it does work, it will make you extremely rich. There’s a “but” and there’s a “but” to the “but.”

You can guess the first “but:” the odds are against you.

  • Few things changed the world – and the US in particular – as much as the automobile. But more than 3,000 US car makers have come and gone. Sure, some losers were bought by winners, but on balance, if you bought any particular participant, you likely lost money.
  • Many people know the story of the American Gold Rush of the mid-1800s: the average California miner made $10-15 per day (probably a generous figure if we consider everyone who tried). The real money was in selling supplies to the people attempting to strike it rich*.
  • Although the extent of the Dutch Tulip Mania of the 1600s is doubted by some academics, whoever paid the price of a mansion for a one bulb probably regretted it.

*One could argue that this is the cause of bubbles in the first place, or at least an exacerbator: Instead of picks and shovels, modern enablers are selling the aspiring (if indiscriminate) investors inferior, overpriced versions of the “winners” they hope to buy.

Don’t let me be a wet blanket. If you are truly skilled at identifying the future winners, by all means you should do this. It’s an enormous relative advantage (for instance, there’s evidence that only about 10% of individual investors are actually good stockpickers).

What will AI’s dud-to-winner ratio be?

If you’re not sure but want to play anyway – and this is the “but” to this “but” – we need to talk about histograms.

If you’ve bought a pack of Pokemon cards (or baseball or football cards – or run a VC fund), you understand that most cards in the pack are going to be duds. In fact, many packs are all duds or near-duds. Therefore, anyone aspiring to invest by buying Pokemon packs – probably a dumb thing to do, but this is theoretical – would need to consider two histogram-y questions:

  1. How many packs will I have to buy before getting one or more winners?
  1. Will the aggregate winning-ness of my winners offset the aggregate dud-ness of my duds?… i.e., will my winners pay for my losers?

It’s a bit unfair to equate Pokemon card pack investing to VC investing, because randomly buying Pokemon packs requires no skill*, whereas VCs at least try to apply skill to separate wheat from chaff.

*Index investing, by contrast, is a bit closer to Pokemon pack investing, although most index investing isn’t an attempt to net runaway winners per se.

I’m generally skeptical of VC returns data – shop around and you’ll see “averages” ranging from stock market-ish returns to 57% per year – but VC is like the Pokemon pack effect on steroids: Correlation Ventures found that 65% of VC investments lose money, and that just 4% return 10x or more. (And 0.4% return 50x or more.) In fact, Institutional Investor magazine pointed out that a VC fund would need to hold more than 500 investments to capture one of these mega-winners.

The upside of buying all those Pokemon packs? Look at the white rotated square in the middle: It’s higher (although note that these are pre-fee numbers, and VC funds may charge 2-and-20 fees that get negotiated up or down depending on the hotness of both VC and the particular VC fund at the moment). Plus, the dispersion of returns is a lot tighter than I’d have expected, too.

How to win the AI war

We’ve discussed

  • A gold rush where the real money was made on people showing up for the gold rush.
  • A tech bubble where values were propped up by people showing up for the tech bubble.
  • A biotech bubble that happened because people put more money into biotech stocks, and not because biotech added more value to the world.
  • A planet that got motorized – while most of the companies trying to help (and their investors) failed.

Still, my intention isn’t as negative as may seem. I’m really trying to establish that:

  1. Investors often assume that a hot new thing that moves humanity forward will automatically be a great investment, but so many investors assume this that the hot new thing may end up a disappointing investment in aggregate.
  2. Mega-winners from megatrends indeed create fortunes for a few people and a few investors, but mega-winners are usually surrounded by hundreds, or thousands, of also-rans.
  3. It’s possible to diversify by investing in a fund, but there’s a high chance the fund’s returns are also watered down by the also-rans.
  4. As the market realizes that it owns a lot of also-rans instead of actual mega-winners, prices come down.

Everything I’m saying is a sweeping generalization. Almost recklessly vague. Funds can do very well by investing in hot things. So can individual investors. It happens frequently. And if you invest in AI, my wish is that it happens for you.

My goal is not to dissuade you from investing in AI. AI will change the world. Precedence Research predicts its market size will grow from $120 billion now to $1.6 billion by 2030. AI will birth AI millionaires and AI billionaires.

But if history is a guide, it will leave a lot of dead companies and burned investors in its path.