Nvidia CEO Jensen Huang (Photo by: Artur Widak/NurPhoto via Getty Images)
NurPhoto (from Getty Images)
“What about an AI bubble?” This is the most common question asked by economists and investment professionals today. Skepticism is appropriate regarding two issues: bubble claims and valuations of companies in new sectors. Although bubbles are difficult to identify in real time, it may be easy to see underlying demand.
The underlying fundamentals are very positive for artificial intelligence companies, and two major economic activities are almost certain to be strong. Companies that create large-scale language models (LLMs) will become a major driver of productivity across a vast range of businesses, nonprofits, and governments. Additionally, many companies are developing specialized applications that use LLM to improve productivity for specific tasks such as billing or product design. In addition to these two main approaches, smaller language models for specialized tasks have been developed, but it is less certain that this approach can compete with specialized apps connected to LLM.
However, strong fundamentals do not always justify very high company valuations. The key is to understand how strong the fundamentals are relative to current valuations. It's incredibly difficult, except in retrospect.
The AI industry is figuring out the best business models for applying technology to real-world problems. Add to this the human tendency for speculative exuberance, and evaluation becomes difficult. Business model challenges have been under-appreciated in recent discussions.
In 1908, there were more than 250 automobile manufacturing companies in the United States. Two major trends followed. Car sales skyrocketed to previously unbelievable levels and most car companies went out of business. That's an amazing result.
Car companies needed to figure out how to efficiently manufacture the products that Ford excelled at. They needed to figure out what customers wanted most from their cars, and what General Motors was good at. In industries with large economies of scale, underperforming firms lose market share and their costs exceed those of other firms. So they failed.
The dot-com boom of the 1990s had similar results. Many companies were founded in e-commerce and most of them went out of business. However, online shopping has reached levels that few people expected 20 years ago. The business concept was great, but most of the concrete attempts ended in failure.
At the root of this dichotomy is a concept I call the “economy of trial and error.” Companies need to experiment, even if many experiments fail. Or success can mean that the company learns what doesn't work. Major innovations rarely work perfectly on the first iteration.
In the field of AI, many errors will be observed. Some companies will fail. Some people stumble badly and still survive. And a small number of companies will find the right model to help their customers significantly improve their productivity and ultimately deserve the high recognition they deserve.
Market structure is part of the puzzle that determines a company's ultimate success. In some cases, one company may dominate a particular field. Who is the #2 company for small business accounting software? That's because every accountant can read computer files created with QuickBooks. Small business owners using alternative accounting programs have even more problems.
Some industries have a few large companies, while others have many small companies. Think of a restaurant. I wrote this a while ago. “The large-scale language model will resemble an Airbus-Boeing oligopoly, but the applications sector will look like sushi, hamburgers, pizza, etc.” That means there will be many companies using very specific knowledge in a domain to provide very specific applications. When electrical contractors want to reduce time to bid, they turn to AI products developed with deep knowledge in the specific field of estimating the cost of electrical installations. That program won't help car dealers trying to increase sales in their service department, but other apps will. However, these application providers may use the same LLM as their engine behind the scenes.
When it comes to stock valuation, consider that Amazon's stock price reached its all-time high in 1999. Two years later, the stock price was down 95%. But it turns out that buying at 1999 highs isn't so foolish. The stock is now worth 52 times its 1999 peak. An equivalent investment in the Standard & Poor's 500 Index would be worth approximately eight times more than the amount invested, with dividends reinvested.
Amazon's story is worth it not only for its final monetary value, but also for the wild journey it takes along the way. This stock has experienced many setbacks along the way. And investors should remember that many dot-com investments have become worthless.
Is the AI industry a bubble? To be honest, I don't know. Individual stocks in this sector are quite risky at current prices, but some may prove to be worth more than today's prices. Many investments result in a total loss. However, there is strong evidence that AI will be a major force in the business world for years to come.
