Are we in an AI bubble?

In 1998, I graduated from college in the height of the Internet boom, just before the Internet bubble. Crazy returns on stock investments led more investors to flood the market. Ryan Jacob shared with Wired magazine that his mutual fund had grown 150% in 1998, sparking a wave of new investors demanding a piece of the action. Even seasoned experts such as financial analyst Stephen Koffler argued “On the face of it, we certainly agree the stock is in uncharted valuation territory … but as an Internet stock, Cisco is cheap!”

My first real job in 1999 was as a systems engineer at a dot com company, which launched my career in web development. After working at that company for 9 months, the executives suddenly changed their business plan, shifting how they were going to make money. Red flag!  I immediately updated my resume and was unsurprised when I and all the rest of the staff were laid off 2 months later. Eventually, the Internet bubble burst causing the recession of 2000-2001.  The stock market crashed in 2000, 2001, and 2002, with the NASDAQ dropping by 77% between its peak in March 2000 to its lowest in 2002.  

When I say that what’s happening with AI has a similar feel to what happened in the late nineties, I speak from experience. I lived through one bubble. I witnessed the claims of unending growth, the arguments based on feelings, and the suggestions that this time is different. The same things are happening now. 

Are we in an AI Bubble?

Economists generally define a bubble as a situation where market prices for an asset rise significantly above fundamental value, driven by speculative behavior. During such bubbles, the purpose of investment changes – not from an expectation that an asset will directly provide value to the purchaser, but from an expectation that someone else will buy the asset for more money. Investors, in such situations, don’t view assets as objective value but as subjective value. The value of an asset becomes “what others are willing to pay for it”. 

In every market, there are investors that see things in terms of objective and subjective value. What turns the market into a bubble is when a preponderance of investors start using subjective reasoning for purchasing an asset. Stephen Koffler’s statement above highlights this mentality: “As an Internet Stock, Cisco is cheap”. He qualified the recommendation of Cisco as an “Internet stock” not because its fundamental value had improved, but because everyone else was already buying that category of stocks. He expected Cisco’s stock to grow because other people expected it to grow. 

The same mentality occurred during the housing bubble ten years later. People bought houses, not to use or rent, but in the hopes of quickly reselling at a higher value. 

Is that happening with AI today?

A friend recently sent me a couple stories claiming we are in an AI bubble. As evidence of the bubble, they pointed to the astronomical investments major AI companies are pouring into data centers. For example, Sam Altman stated they were committed to developing 30 gigawatts of computing capacity for $1.4 trillion. To put this amount in perspective, there are only 15 countries in the world with a larger GDP than $1.4 trillion. Sam is not alone. According to a McKinsey Consulting report, the IT industry is expected to invest $5 trillion in data centers by 2030 for AI processing alone (and another $1.5 trillion for traditional software processing needs). The entire data center capital expenditure by the big four tech firms Alphabet, Meta, Microsoft, and Amazon TO DATE has been around $700-800 billion. 

Such a historic flow of capital into an unproven technology should give us pause. We should ask serious questions. Are these investments irrational? Can AI capabilities really provide as much value as these companies are planning? Or will our economy implode due to a misallocation of resources? Should I invest in Prepper books and build my own bunker? 

The Singularity is …Now?

Twenty years ago, famed futurist Ray Kurzweil published a book entitled The Singularity is Near. His thesis states that the rate of computing advancements follow an exponential curve. Kurzweil looks back at history, prior to even the computers, and tracks the rate of speed and complexity in computing devices. As he plots these advancements on a graph, they increase at an exponential rate. Eventually the computing power of machines will equal a human. After that, the intelligence of a machine will equal all living humans. If you remember your math, you know that exponential curves always end with a line that sky rockets straight up. That line at which things shoot straight up is called the singularity. If computing follows the same curve (as his data suggests should happen), then eventually computing will reach that singularity. According to Kurzweil, we should hit that singularity by 2045. 

What happens at the singularity? Well no one knows for sure, but it is presumed that computers will achieve super intelligence, intelligence greater than all of humanity put together. With such super intelligence, it can continuously improve its own intelligence while simultaneously solving our hardest problems. The rate of advancement will be so fast that we cannot predict what’s on the other side.  

Software engineers are increasingly using AI to speed their programming. According to Sam Altman, CEO of OpenAI, almost all new code written for their organization is assisted with Codex, their AI code generator. This trend has been echoed by CEOs of Microsoft, Facebook, and X. How long will it be before the human in the loop slows the progress of AI development to such an extent that it becomes a hindrance rather than a help? How long will it be before just one organization allows an AI to write code to improve itself?  If it truly is as intelligent as a human, how long will it be before it can reach super intelligence?  How long before we reach the singularity moment? Based on the comments from the AI CEOs, the pace of change in AI looks like it may beat Kurzweil’s 2045 prediction. We may be on the cusp of the singularity now. 

Indeed, that seems to be the major thrust in each of the AI labs, to be the first to reach super intelligence. Why? Because they believe that if they reach super intelligence first, it should grant the owner of the AI the ability to dominate every other business, product, and service, by making better decisions, greater innovations, and streamlined operations faster and more accurately than all competitors. The super intelligence could presumably out predict and out think all competing AIs. It would always make better decisions. The AI labs are all in a race to get there first. And that’s why all of these tech companies expect trillions of dollars in data center builds over the next 5 years. 

Even if the so called super intelligence is never achieved, current generative AI capabilities are hindered by the lack of computing power. Barely 3 years after ChatGPT 3.5 was released, we’re only just beginning to grasp all the ways AI can improve our lives. But the capabilities are limited. When I ask ChatGPT a question, it only “thinks” for a few seconds. It can’t think longer than that because its too expensive to do. The demand from users asking difficult questions soaks up computing resources, so the AI platforms limit how long those processes can take. Just using today’s AI capabilities and providing longer “thinking” compute times would fundamentally improve their outcomes. To do that, today’s tech companies need more data centers. 

How much is too much?

Undoubtedly, the tech firms need to build more AI infrastructure just to keep up with current demand. Undoubtedly, the AI labs will develop models improvements which may require even more computing power to implement. How much capital expenditure on AI infrastructure is “reasonable”? “Reasonable” investments in AI infrastructure are any such sums of money that the tech companies expect to make back on the products they develop. Do they expect to double their revenue? If so, then, doubling their investment in AI infrastructure makes sense. Do they expect to 10x their revenue? Then 10x capital expenditure on AI infrastructure also makes sense. It requires predictions. 

Here, I come back to objective versus subjective value. The tech firms expect tangible value from these investments. They expect to create something of value, of immense value, of astronomical value, that others can consume. They are not, however, paying for data centers in the hopes that someone else will buy the center. They are not investing based on subjective value.

Tech companies spending large sums of money should not be the criteria of a bubble. A bubble emerges when other investors start buying company stocks, not based on fundamentals of the business, but on the hopes of selling the stock to others at a higher price. While I don’t watch the stock market closely, it’s possible that we’re already in the early stages of an AI stock bubble. But consider this… when the Internet bubble popped in 2000-2001 and the entire economy went into a recession, there was one and only one industry that saw positive growth during those years – the Internet industry. While the stocks collapsed, the revenue continued to grow. 

I expect the same thing to happen again. The improvements from AI implementations will shake up multiple industries. Economies will reel. Workforces will shift. Businesses will reinvent themselves or vanish. A massive correction will likely dominate our economy in the next 5 years. Throughout that recession, the AI tech companies will continue to grow, continue to innovate, and continue to increase their revenue, despite any fluctuations in stock prices. 

AI Bubble Take-away

Bubbles are a by-products of a me-too approach to value. They exaggerate the promise returns, focusing on subjective values, leading to misallocated capital, and distorted expectations. Despite these problems, an influx of investments in a booming industry can lead to radical transformations. The investments made during the Internet bubble gave us the infrastructure that powers today’s digital economy.

The AI bubble, if we are indeed in one, may do the same. What matters isn’t whether the bubble pops, but what remains standing afterward. And if history is any guide, the foundations being laid today will form the foundation of our economy for the next 20 years.