Free Trades

If AI Is in a Bubble, It WILL Pop

[ My friend Doug Behnfield, a wealth manager and senior vice president at Morgan Stanley in Boulder, has contributed many commentaries to Rick’s Picks over the years. Below is the Q3 report he sent out to clients several weeks ago. Like many observers, he is troubled by the enormous concentration of investment capital in the AI space. Can the eventual payoff ever be big enough to justify the estimated $10 trillion that will flow into AI technology by 2030? Read why Doug thinks there are better places to park your money. With apologies to him, I have dispensed  with his meticulous footnotes and several graphs to simplify typography. The Jetson’s illustration was also my idea, based on his original headline, ‘Thoughts on the Jetsons and Rope-a-Dope’. RA ]

In late 1962, CBS introduced the Hanna-Barbera evening cartoon The Jetsons. It was inspired by their hit series The Flintstones, but set in the future. It lasted for only 24 weekly episodes, but it made an indelible impression on the Baby Boom Generation. Along with flying cars and Rosie the robotic maid, George Jetson worked two days a week, one hour per day (not remotely), and all he did was go in and push a button to start and stop a machine. (the Referential Unisonic Digital Indexer Machine, at Spacely Sprockets).

I was reminded of The Jetsons when reading a Wall Street Journal article describing the rivalry between Mark Zuckerberg and Elon Musk in developing robots. In it, Elon Musk predicts that there will be “at least 10 billion humanoid robots in the world, remaking the idea of work and life” by 2040 (The Jetsons was set in 2062). Zuckerberg’s humanoid robotic aspirations are dependent on gathering data from the microphone and camera in his Artificial Intelligence (AI) Glasses.  With them, he intends to create the Large Language Model (LLM) that will be used to program the machine that can do all your work for you. The glasses will eventually record what you see, gathering “vision data,” even as they make a “wearable” fashion statement.

When reading about the development of AI eyeglasses, I was impressed with the breakthrough represented by having a screen projected inside the lens, providing closed captioning (with foreign language translation) for face-to-face conversations. Next will be AI-generated conversational response recommendations literally right before your eyes! Apparently, AirPods now translate in real time too. These developments represent a leap in communication technology, beyond translators that are common on cell phones.

The irony is that there is quite a debate concerning the impact that screen time can have on mental health in general and childhood development in particular. At least people will stop walking around with their phones in their faces.

There is a wide range of opinion about how AI will impact productivity across the spectrum of human activity and the speed at which AI will be implemented. The list of potential productivity enhancements is long and broad, as are the concerns surrounding the unknown future of how AI will develop. Lacy Hunt put it this way:

“AI decreases labor demand by automating cognitive and repetitive tasks across a wide range of service sector skills. Prior automation mainly affected routine factory roles, but AI goes further. Traditionally, new graduates gained experience through tasks such as data collection and analysis. Now, AI can do these tasks quickly, sharply reducing the need for junior staff. One AI-enabled employee now replaces several people, resulting in fewer hiring needs. AI also enables firms to automate mid-skill roles, thereby pushing down salaries and shifting choices from people to software. This did not happen in prior technology waves. Reports indicate that college graduates already face lower demand because AI can handle advanced tasks. This results in slower hiring, weaker wage growth, and reduced bargaining power for workers without irreplaceable skill.” The expectation is that AI will reduce the need for capital investment and employment in entry as well as mid-level staff, including medical assistants, legal aids and customer service associates as their tasks are taken over by automation.

 Next? Who Knows?

The debate is raging about how the various AI applications will impact most aspects of the human condition. Education is an excellent example of the dilemma AI can present. On one hand, AI can free up time and reduce the number of teachers needed to educate our students by taking over tasks like evaluating essays for content, including plagiarism. On the other hand, AI can create students’ essays that contain “hallucinations” (nonsensical responses) and plagiarism, hampering the learning process. One fear is that students’ relationship with their computers could supplant their human relationships. As a result, only a small percentage of educators are fully supportive of AI applications in education. In a fascinating article in the Wall Street Journal entitled “Giving Career Advice to Kids Has Never Been Harder,” the author advocates, among other things, a return to the liberal arts and away from science, technology and math. She quotes a career counselor who advises: “The most important thing we can advise kids to do is learn how to learn how to think, because the only thing we can do over computers is to be human.” Wow.

The financial press is becoming dominated by articles about AI. Opinions about its impact on the economy and financial markets range across the spectrum. It resembles the Gold Rush. However, regardless of all the frenzy, AI seems to be a much more evolutionary technological development than a revolutionary one. After all, a washing machine is a product of AI, too. (No more tubs, washboards and ringers.)

AI currently involves a significant initial investment in conventional brick-and-mortar infrastructure for data centers and the electricity to power them. The estimates of dollars and Gigawatts required seem to be growing at an extraordinary pace. It remains to be seen whether the current infrastructure being implemented to facilitate AI will survive the rapid technological development underway. Overbuilding in response to new technology is perennial. It is even conceivable that a Large Language Model will contribute to making the current business model obsolete. It is not clear whether a substantial portion of AI will be delivered by “open source” software, fee-based programming from dedicated AI modelers, or in-house machine learning hyperscale data centers that the major search and software companies are building. From an investment perspective, the most important question is: “Who will make money on AI and when?”

During the Dot-Com era, fiber optic companies were essential infrastructure providers that generally went bust. According to the “AI Overview” on a Google Search of the topic: Severe overestimation of internet traffic and demand, fueled by venture capital and market hype, left fiber optic providers with a catastrophic glut of ‘dark fiber’ after the dotcom bubble burst. This overcapacity was exacerbated by technological advances that increased network efficiency, leading to a collapse in prices and the bankruptcy of many telecom and fiber optic companies.

According to the same source, $7 trillion is projected to be spent by 2030 globally on AI Data Centers. In the Wall Street Journal this morning, the number grew to $10 trillion! The amount of revenue that must be earned to support that level of investment is almost unfathomable. Just the announcement of a circular agreement between hardware, software and hyperscale companies to collaborate can result in dramatic jumps in stock prices.

Another Bubble?

Looking back at the Dot-Com Bubble, it is easy to draw comparisons to the current frenzy. The arguments for why “this time is different” are similarly compelling, but there is no certainty about what the economic impact

of AI will ultimately be. Examples of revolutionary human progress include the internal combustion engine, rural electrification, urban sanitation, telecommunication, pharmaceuticals and chemicals. While dramatically improving productivity and the “standard of living,” they all resulted in robust employment and capital spending.

In so many ways, the expectations for AI seem uniquely speculative. But calling it a bubble is a reference to public opinion and ultimately to stock prices. It is not necessarily a characterization of the legitimacy of AI as an important technological event. (The term “bubble” is used because bubbles pop.) It is therefore important to keep in mind that trying to guess when the pin will show up is equally speculative.

Stock prices in the Dot-Com Bubble peaked in mid-2000 and crashed. Personal computers and the Internet nevertheless turned out to be revolutionary technologies by most measures. Their impact on the economy proceeded uninterrupted, with little regard for the losses in the market post-bubble. However, the leading companies in those industries, having risen in price by five-  to ten-fold in the four years leading up

to the bubble peak, saw their stock prices crash by 65%-100% the following year (2001) , They all went down together, and the NASDAQ Index declined 80% peak-to-trough. The lesson is that stock prices are driven by fear and greed, and they can “go further than you think” in response to optimism or disappointment. The NASDAQ took 15 years to get back to its 2000 high.

Concentration Risk 

Today, the price of the “Mag 7” stocks (the top AI companies), have traced out a trajectory that is practically identical to the leading companies in the dot-com era. In doing so, they have dominated the performance of the entire S&P 500 Index. The ten biggest stocks in the S&P 500 now represent over 40% of the total index value, and the Mag 7 are all in the top ten.

S&P 500 Index funds are the most widely held vehicle for equity exposure among the investing public (particularly in 401Ks) because they represent Blue Chip large-capitalization growth stocks. In addition, the S&P 500 Index is “capitalization weighted” and that is why the top ten stocks can be such a substantial percentage of the total index value. The remaining 490 companies in the Index represent only about 60% of the total (of five hundred companies). The S&P 500 Index funds have outperformed all other major stock indexes for decades and, as a result, they are also the most popular stock market holding in individual portfolios. If the AI Bubble pops, it could take the S&P 500 Index Funds with it.

The Wealth Effect

The term “wealth effect” refers to the impact that a change in household wealth has on consumer confidence. Very simply, people are more willing to make discretionary consumer purchases (like restaurant meals, cars or Florida vacations) when their net worth is appreciating than when it is declining in value. The wealth effect is a two-edged sword, and as a result, it can be positive or negative in its impact on spending. Household wealth is made up primarily of savings and home equity, so society is typically impacted together as interest rates and the value of stocks, bonds, and residential real estate ebb and flow together on a national basis. Therefore, the state of the economy in general can be driven to a large extent by the wealth effect.

Back at the turn of the century, the dot-com stock bubble had a very minor direct positive wealth effect on the consumer. Similarly, there was only a minor negative wealth effect when the bubble burst. Participation in the stock market by households was more limited than today due to demographics (Baby Boomers were in their 40s) and the fact that 401K retirement plans had begun to replace traditional Defined Benefit Pension Plans offered by employers. The recession in 2001 was very short and shallow. Whereas stock prices crashed, home prices appreciated throughout the late 1990s and early 2000s as Baby Boomers entered their peak earning years. Home prices had a much greater impact on consumer confidence, as Baby Boomers could leverage their home equity for a wide range of consumption.

It has become evident that consumer spending is skewed heavily toward the top 20% of consumers, exemplifying both income and wealth inequality. Older households in particular are benefiting from the wealth effect due to greater accumulation and access to their savings. 401K balances have a muted impact on spending when the participants are younger, and retirement is far in the future, but today, Baby Boomers have reached retirement age and remain the dominant cohort demographically. For the most part, retirement means substituting investment income for employment income, so their 401Ks and other forms of retirement savings have evolved from accumulation vehicles to income vehicles that fund their lifestyle.

Asset allocation is up to the participant in 401Ks and IRAs, and these Baby Boomer retirees have 65% of their allocation in stocks, primarily in S&P 500 Index Funds. Because of the combination of an aging population and the changes introduced 35 years ago with the 401 (k), the impact of financial asset values on the wealth effect is much greater than at any time in history. Just as the economy has avoided recession in large part due to the positive wealth effect of the current bull market, the economy is more vulnerable to a bear market in stocks than ever before.

Rope-a-Dope

On October 30, 1974, Muhammad Ali entered the ring in Kinshasa, Zaire, to fight George Foreman in a bid to regain the World Heavyweight Boxing Championship Title. It was billed as “The Rumble in the Jungle.” 60,000 people attended the fight in person and close to one billion people watched it live on pay-per-view television and closed-circuit theater TV. Ali was getting older and had lost the title to Joe Frazier in 1971. Many thought he was over the hill, and he was a 4-1 underdog. George Foreman was the defending heavyweight champion and considered indestructible, having won all 40 of his previous professional bouts, 37 by knockout.

Miraculously, Muhammad Ali pulled an upset win in the eighth round, and he did it by using a tactic that he named “rope-a-dope”. It is considered one of the most important sporting events of the 20th Century and, by far, the most important boxing match in history. By holding back defensively against the ropes and deflecting most of Foreman’s constant and powerful blows, Ali succeeded in exhausting Foreman while preserving his own strength. George Foreman was so tired by the eighth round that he could not defend himself when Ali came off the ropes. Ali brilliantly executed a series of combination blows and won the match with a knockout.

In the current market environment of unrelenting new highs in the stock market, fueled by the apparent indestructible promise of AI, it is extraordinarily challenging not to jump into the ring and participate, even though the lessons of history argue against it. It may be a better strategy to stay defensive and let this mania punch itself out.

Here is a great quote from Stan Druckenmiller, the iconic stock market investor who, in the 1970s and 1980s, was Jack Dreyfus’ top advisor and portfolio manager for the Dreyfus Funds. He gained fame by blowing out of stocks right before the 1987 Crash. Druckenmiller was running George Soros’ equity portfolio in the Quantum Fund at the peak of the Dot-Com Bubble. This is how he described what transpired:

“So like around March [2000] I could feel it coming. I just—I had to play. I couldn’t help myself. And three times the same week I pick up a — don’t do it. Don’t do it. Anyway, I pick up the phone finally. I think I missed the top by an hour. I bought $6 billion worth of tech stocks, and in six weeks I had left Soros and I had lost $3 billion in that one play. You asked me what I learned. I didn’t learn anything. I already knew that I wasn’t supposed to do that. I was just an emotional basket case and couldn’t help myself. So, maybe I learned not to do it again, but I already knew that.”

It is not just AI stocks that reek of a speculative bubble. While bonds are signaling an economic slowdown or worse and the Fed is back to cutting rates because of labor market weakness, crypto, sports gambling, meme stocks and day-trading on phone apps are all the rage. Even the price of gold and silver has gone parabolic. Focusing on capital preservation seems more appropriate than ever. In their most recent report entitled A Fond Farewell to 10-Year Treasury Yields Above 4.00%, the interest rate strategists at Morgan Stanley are estimating that the Fed will cut short-term interest rates by an additional 1.5% to 2% over the next 12 months in response to a slowing economy and muted inflationary pressures.  Historically, the yield on long-term Treasury bonds captures about 70% of Fed rate cuts, suggesting strong total return performance in the investment-grade bond market over the near term. This should provide a far better opportunity to reduce duration risk in bond portfolios.