Anyone who absolutely writes off artificial intelligence who knows little of the field, how long it’s been in existence, and the range of different technologies that fall under the umbrella term. There are many public companies that profitably apply AI and offer good investment potential.
But as short-minded it is to dismiss an entire area of development, embracing a hyped addition can be equally foolish. There are signs that some of the generative AI companies face some problems: a lack of practical success at the corporate level, a need to keep scaling up, and massive amounts of debt deals that one day will have to be repaid.
A Quick Background On AI
Too many people toss about the term “AI” as though it were a singular thing. Look at the number of companies currently touting their use of AI in products. This is an old form of technology marketing in which corporations attempt what one might call label-washing. They wrap themselves in the newest buzz term as though it were a competitive advantage and not a coat of paint to be quickly changed over time.
AI started in the 1950s. By 1966, the first chatbot, developed by MIT professor Joseph Weizenbaum, simulated a therapist by responding to human input by asking prompting questions in response. It was a form of natural language processing and a forerunner of ChatGPT, et al. Rule-based software, expert systems, neural networks, computer vision, machine learning, predictive AI, robotics — all these and more come under the rubric.
You’ve probably been using AI for much of your adult life without thinking of it as such. Spellcheck? AI. Troubleshooting wizards? AI. Personalized shopping? AI. GPS directions? AI. Virtual assistants? AI. Predictive maintenance in cars? AI. Smart home devices? AI. Social media timeline generation and the “algorithms”? AI. Chatbots? AI.
Generative AI
Gen AI is a set of technologies that use advanced statistical methods to examine millions or even billions of documents and to find the most common series of components in given contexts. They don’t think — that is something critical to remember. They generate text that might sound like thought.
The stochastic structure is impressive and useful in some settings but also limiting. Statistics always include the potential for unusual situations at the outer parts of distributions. In this case, the larger the number of documents samples and the queries coming in, the more inevitable the appearance of those unusual situations.
An example is so-called hallucination, which the software makes things up. One of the newer public examples is from the high-profile law firm Boies Schiller, where a court filing from them in July contained “erroneous, AI-generated case citations,” according to reporting from The American Lawyer.
The resources necessary for these broad gen AI systems, different from the more targeted implementations that happen, are expensive. Massive amounts of hardware, huge power requirements, rivers of water to cool systems, expensive copyright lawsuits, all while the business models demand constant growth.
The Financial Tremors
Serious business press outlets are noting money problems. The Wall Street Journal asks the question, “Spending on AI Is at Epic Levels. Will It Ever Pay Off?” The costs of building the needed data centers are immense. A “half-built AI factory bigger than 10 Home Depots” has a price topping than $15 billion.
It’s part of a “building spree.” Data centers have become a major focus in commercial real estate investment. They are expensive, take years to complete, and are more complicated than many of the people involved may realize. The design demands are complex, and they often have only a limited lifespan. New equipment, new technologies, and new demands like every growing power needs do not necessarily allow a simple changeover. As Jones Lang LaSalle, a leading global real estate services company, notes that “new data centers risk becoming obsolete.”
Another Journal article noted that tech companies spend billions on data centers and software development. These businesses take on “heavy debt” with relatively “tiny” revenue, and it sounds like a replay of the dot-com bubble in the late 1990s and early 2000s.
Uber was founded in 2009. The company received more than 26 funding rounds for a total of $13.2 billion, according to Tracxn. The company went public in 2019. According to data from S&P Global Intelligence, the first annual publicly reported profit came in 2018, and then not until 2023.
The amount of Uber’s investment was tiny compared to what generative AI companies need. Hedge fund manager David Einhorn, founder of Greenlight Capital, asked whether “spending a trillion dollars a year or $500 billion a year” could deliver good results, Bloomberg reported.
“The numbers that are being thrown around are so extreme that it’s really, really hard to understand them,” Einhorn told Bloomberg. “I’m sure it’s not zero, but there’s a reasonable chance that a tremendous amount of capital destruction is going to come through this cycle.”
This isn’t to say that investment in companies involved in gen AI is bad, wrong, or useless. But it does look awfully risky, which is something any investor should consider.