AI bubble

How we got here (the last bubble)

The frequency with which the words "AI" and "bubble" appear together is increasing and many people are comparing the current enthusiasm for companies focused on artificial intelligence to another relatively recent technology stock bubble, sometimes called "the dot com bubble" or just the Internet bubble. To understand whether such comparison is merited and to determine whether there is something worth learning from this past experience to apply to current events we should start with understanding what that bubble really was:

First the simple definition: In five years, between March of 1995 and March of 2000, the Nasdaq index grew from roughly 800 to roughly 5,000 - increasing in value by more than 6 times. It took until 2015 for this index to regain that level (it currently sits at about 14,000). Around the time that ChatGPT was first introduced in 2022 the index was around 11,000 so for calibration, a similar five year growth rate would put the index at 66,000 by March of 2027.

It is worth going beyond the simple definition and examining what actually happened during this period and why the market then collapsed (almost all the way back to where it sat in 1995 - down to 1,139 on October 4, 2002). The story has three primary actors -- technology, companies, and investors.

The Promise of Technology -- way back when, there was a time when we didn't all have instantaneous access to all of human knowledge. In 1995 the total global population of Internet users (estimated by IDC) stood at just 16 million people. And for most of those people this access was very slow and intermittent. To put this in context - that 16 million people represented just .4% of 1995's world population, whereas in 2023 an estimated 70% of a much larger world population can access the Internet (world population has increased from 6 billion to 8 billion over that time). Back in 1995 we also didn't know what to do with this technology. We could look up information (when an information source was available), we could send messages to one another, and maybe play chess or checkers. Nonetheless the innovation engine was fueled.

Companies -- Innovation is tricky -- by definition you don't know what will or won't work. Many of the best companies have started with one idea but then discovered something else completely different while on the journey, and ended up a very different company. Another compounding problem is that there can be dependencies between innovations. While buying things online certainly seemed like something that could work, companies wanting to sell online required new kinds of secure payment systems and a change in consumer confidence around how those goods or services would be delivered. The promise of wealth creation also attracts swindlers who really have no interest or ability to create innovation but can parrot the language of innovation. So in these cycles both existing companies and entirely new companies will be created, each pursuing innovation around a set of plausible ideas with some being more genuine and capable than others. Which leads to some difficult decisions by investors.

Investors -- There are a range of different types of investors characterized perhaps best by their sophistication and their risk tolerance. On the sophisticated side of the equation there are venture capitalists with a very high tolerance for risk, and wealth manager entrusted with safeguarding retirement funds with a low tolerance. On the other end of the spectrum, private individuals have a range of sophistication and risk but generally will be less sophisticated than professional investors and have a lower risk tolerance (and lower ability to invest altogether). Innovation cycles like the 1995-2000 period will be led by venture capital -- risk taking sophisticated investors. They will attempt to discern which ideas are more likely to succeed and which entrepreneurs are more trustworthy and capable. As these companies develop they move to later stage investors. One early mover, Yahoo, went public in April of 1996.

How did the events between 1995 and 2000 play out amongst these three actors? First, the technology continued to improve and grow in reach. From the initial 16 million users in 1995 (0.4%), usage had grown to almost 6% of the world's population in 2000. Higher speed access and many of the foundation problems (secure payments for example) were being solved, enabling an increasing number of use cases. The real problem, the bubble, was on the side of companies and investors - not technology.

Whether you characterize this positively (optimism) or negatively (fraud) or somewhere on this spectrum, the fundamental problem in 2000 was that investors had driven the value of companies in this segment far beyond their business model and technical maturity. Pets.com is perhaps one of the most famous of the dotcom era failures -- building way too big a company for the state of the market. Bubble are created when there is a pipeline of new investors adding to the over valuation of companies. This means attracting investors with a lower risk profile and less sophistication. This is easily done when people seem to be (or actually are) making a lot of money. But as each wave of investors enters a market the focus on making money can replace the focus on properly understanding the value of the underlying asset. Eventually this results in a collapse of value, accelerating as every investor tries to exit the collapsing market.

AI bubble

So do we have an AI bubble today? It is very hard to see that we are anywhere near the irrational exuberance of the dotcom era... yet. But it would be accurate to observe that we have the same conditions being configured -- we have a technology that is growing rapidly both in usage and maturity. This technology has unleashed a wave of innovation by both existing and entirely new companies. And investors -- especially those early risk tolerant venture capitalists -- have been investing in the future, hoping to chose those ideas and entrepreneurs most likely to succeed. It looks a lot like 1995. Can we learn the lessons of past tech-driven growth cycles and calibrate valuations to actual performance and market maturity? Or are we doomed to chase exponential (but short term) rewards through a cycle of over-estimating the business growth scenarios ultimately leading to valuations that become entirely detached from the fundamental value?

As the bumper sticker says "Please God give me just one more bubble. I promise not to screw it up this time." That sort of sums up human nature, don't you think? So who do we blame for this coming AI bubble? The technology? The companies? The investors? "...when after all, it was you and me."