If you’re not living under a rock, you should have heard of the so-called “AI-bubble” that will surely explode and shatter something into a thousand pieces.
AI critics are already feasting on the imminent AI bursting.
But for starters, we have to ask, beyond the intuitive idea (or desire!) that AI will explode, is any of this real? What is the evidence?
If you have read my previous articles, you are aware that I am affiliated with the philosophical current of the “skeptics,” and you may have come across my “The Skeptic AI Enthusiast” Substack publication. So I’m not taking the “AI bubble” prophecy for granted.
Let’s first clarify what exactly “AI is a bubble” means and then take a look at what would come after it, as the title of this post promises.
To be clear, many people say that AI will burst mainly because they hate it. And I get them: we are (at least I am) fed up with hearing about AI all the time, and even worse, reading content on the Internet that is obviously generated by a machine.
In my opinion, AI-generated content rots the Internet and sucks the value out of it.
But that’s not what “a bubble” is about.
What are bubbles?
A bubble is about inflated expectations; it’s about companies with impossibly high valuations. It’s about AI’s exaggerated claims and AI companies that appear to be a house of cards.
A bubble is the highest point in a cycle that economists (the little I know about the economy comes from my economist brother) are familiar with. That cycle goes as follows:
First, a discovery / a technology / an ingenious scam scheme is discovered.
Then, money flows in the direction of the novelty. Investments attract other investments in a seemingly continuous positive feedback cycle.
Companies’ valuations reach ridiculously high points. Speculation becomes a frenzy.
An unexpected event cuts the speculation frenzy. Sometimes, something that has been simmering for a long time reaches a critical point.
Valuations fall in a thunderous crash, leaving many with devastating losses.
This cycle has been seen many times already. Perhaps the most recent one was the cryptocurrency collapse, which followed the pattern to the letter:
Blockchain technology was a real discovery. It led to cryptocurrencies.
Many people sought to make easy money, so they invested in cryptocurrency as quickly as possible. The mounting valuation was seen as a magnet instead of a red flag. One of my cousins invited me to make a joint investment in crypto (I declined).
Fraud by the leaders of crypto companies has been discovered. Some platforms (like FTX) misused customer funds recklessly.
FTX’s valuation plummeted to zero, along with that of other crypto companies. Clients lost their money.
Other instances of this nefarious cycle come to mind, like the 2008 real estate crisis. However, bubbles date back to the 19th century.
Yes, during the 1840s, a bubble took place in England related to railway public investment. We tend to think that investments made by the average Joe are something recent, but they’re not. Between 1844 and 1847, hundreds of railway companies were floated on the London Stock Exchange.
However, shares were often sold on expectations of future profits, rather than actual earnings. Greed led to reckless financial behavior. Then, what was called “overconstruction” translated to a fall in prices of railway infrastructure. Then, the collapse came. Some elders lost their lifelong savings.
Some have pointed out that companies’ valuations are today clearly inflated. This can be verified with hard data and is not just subjective opinions. In particular, the ratio of market value to earnings, known as the price-to-earnings (P/E) ratio, is typically around 20.
Now, consider the dot-com bubble at the turn of the century. Valuations of internet companies were unreasonably high; the NASDAQ index reached a high of 5,048 in March 2000. That’s above 250 times the “normal.” Of course, investors thought they were living exceptional times because the Internet multiplied everything… That is, until it multiplied the losses of investors of countless dot-com companies.
In this example, you see again how greed takes the better of us: instead of seeing an unreasonably high price-to-earnings ratio as a red flag, many saw it as an opportunity.
The proponents of the “AI bubble” term obviously think we are in the third phase of this process. But are we?
The AI bubble
The 2022 crypto bubble is still fresh, and its consequences are evident: at this point, it’s unclear whether, in the future, cryptocurrencies will regain popularity. Some say crypto was a failed technology.
Will AI, if the bubble explodes, be forgotten, too?
Not so fast.
I’m going to present to you the evidence I gathered pointing to AI being a bubble, and then I’ll show what could come after it bursts.
In my opinion, the most crucial evidence of an AI bubble is inflated valuations.
As I mentioned above, inflated valuations are signaled by an excessively high stock’s price-to-earnings (P/E) ratio, which is typically well above a ratio of 20. We can find countless examples of AI companies with P/E ratios ridiculously above 20.
A notable example with an overly high P/E ratio is Tesla, with a ratio of 216 for 2024 – this is ten times the reasonable value of 20. However, in principle, Tesla is not an AI company, although this is disputable given its increasing reliance on autonomous driving capabilities, which are, of course, AI-powered.
Tesla’s stock valuation could crash at any moment. The fact that its valuation is inflated can’t be disputed.
Although OpenAI is typically considered an inflated company, its P/E ratio is approximately $300 billion divided by $10 billion, which is 30x, a value not too far above the reasonable range of 20.
What about Google (Alphabet)? Its market cap is around $2,513 billion, while its income for 2024 was around $100 billion. That gives a P/E ratio of about 25, which is entirely reasonable. Alphabet doesn’t have an inflated valuation.
But smaller AI companies are indeed prone to inflated valuations. For instance, Character.ai has a P/E ratio of around 80x — a notable inflation. Another example is Cohere, with a $6.3 billion valuation and $100 million in income, resulting in a 63x ratio.
An extreme case is Perplexity, with a valuation of $8 billion and a revenue of merely $8 million. That gives a ratio of 1,000x!
Discussing extreme cases, there are companies with no income, such as “Safe Superintelligence” (founded by Ilya Sutskever), which obviously have an infinite P/E ratio, as the denominator is zero.
Another frequently cited piece of evidence for the AI bubble thesis is a study published by MIT, “The GenAI Divide,” which shows that 95% of AI pilot projects in enterprises failed to achieve their goals, despite receiving an estimated $40 billion in budget. That would make a loss of $38 billion. Ouch!
I think this study struck a nerve because it supports the narrative that investment in AI is unsustainable, and in extreme cases, that AI is a fad doomed from its inception.
I found an article: “That Viral MIT Study Claiming 95% of AI Pilots Fail?’ Don’t Believe the Hype.” It questions the methodology used by the MIT study, as well as its results. According to this article, the definition of success was too narrow, it ignored key metrics, and some data sources were unreliable.
My point of view is that shortcomings in pilot technological programs within enterprises are a regular occurrence, not a specific issue related to AI. Also, putting a clear-cut “failure” flag on pilot projects is, in general, too reductionist.
Many people discussing the MIT report think that it’s a tell-tale sign that AI as a whole is going to crash, sooner than later. But they are wrong.
Look, I’ve been dabbling in the entrepreneurial world for several years (I had a startup, “Avalinguo,” related to language learning), and I’m aware that most startups ultimately shut down (as my own startup did), and one thing I learned is that the chances of startups developing as a consolidated company are slim.
In the natural world, this is called “Natural selection,” and was what Darwin discovered (by the way, many people wrongly believe that Darwin proposed the evolution of species, when in reality, evolution was well known before him).
This leads us to the last part of this article: what comes after the AI bubble?
After the AI bubble
Keep reading with a 7-day free trial
Subscribe to The Skeptic AI Enthusiast to keep reading this post and get 7 days of free access to the full post archives.