A US Economy Powered by AI

The most asked questions we’ve received from clients of late are: “Is AI in a bubble?” and “What’s your strategy for an AI pullback?” We think this largely stems from a series of headlines over the past few months that have been skeptical regarding the run-up over the past few years of a variety of stocks that are considered part of the AI revolution. Many of these articles prey on the public fears of the potentially revolutionary technology and the equal concern that that the promise may not live up to the hype. Inevitably, comparisons to the dot-com bubble,[1] the subprime mortgage bubble,[2] and/or the great depression[3] frequently fail to provide proper context and meaningful facts for the average person. 

For our part, we take a more measured view based on three key observations of what we know now. First, for the past three years, the stock market has done historically better than it has on average. Second, the publicly traded tech companies at the forefront of AI have  generally had strong earnings that are being plowed back into data center construction (however, more recently, they are starting to turn to the bond market to borrow, as their profits are insufficient to make the needed investments for AI). Third, the data and charts that we monitor indicate that the AI infrastructure build-out is responsible for an ever-increasing percentage of the economic growth in the US economy. In light of these observations, we think it’s a prudent time to consider rebalancing back to one’s long-term strategic asset allocation; and, in some cases, even reducing equity exposure may make sense given near-term cash needs (1 – 3 years), the tax consequences of selling, and your overall financial plan.

Equities, both here and abroad, have done historically well in recent years. As you’ll see in the charts below, three measures of stock market performance- namely the S&P 500 (large US companies), the MSCI All-Country World Index (companies worldwide from both developed and emerging markets), and the Russell 2000 (small and medium US companies)- have each done significantly better over the last 3 years compared to their average returns.  Most extraordinarily, the S&P 500 has averaged more than double its annual return at 20.5% versus its annual return going back to 1988 of 9.17%.  Could this outperformance continue?  Yes!  We never know when a pullback, correction or even bear market will occur.  But for many investors, this recent stock growth has meant that the risk level on their portfolio may be bordering or exceeding the long-term level appropriate for their financial plan.

Moreover, unlike during the dot-com era, most of the companies driving returns in the stock market -i.e. the magnificent seven stocks of Alphabet (Google), Amazon, Apple, Meta Platforms, Nvidia, and Tesla- have achieved incredible and growing profits.  As a result, the recent returns are much more in line with the growth of profits versus the 90’s tech bubble.

At the same time, these tech companies are now increasingly spending the vast majority of these profits on capital expenditures for AI infrastructure at levels well above their long-term average, as shown in the chart below. Many are also relying heavily on borrowing. In the case of Oracle below, the company is estimated to spend 132% of its expected 2026 cash flow e.g. earnings, which means Oracle is relying on borrowing estimated to be about $25 billion a year for the next four years.[4]  Oracle and other data center infrastructure companies have been able to borrow vast sums due to future contracts from AI software providers, such as OpenAI and Anthropic, but this borrowing may be reaching its limits.  In fact, insurance on Oracle defaulting on its debts recently hit the highest cost since the global financial crisis.[5]  Bottom line: borrowing, aka leverage, intensifies the potential for gain, but also for loss if the expected revenue to cover the debt fails to materialize.  The debt levels of these companies (and related special purpose vehicles and joint ventures) is one of the critical items to watch to determine whether the current AI investment speculation is sustainable. 

Finally, the US economy has recently been very reliant on AI infrastructure buildout for economic growth at the same time as other growth drivers have been slowing down.  As illustrated by the chart below, nonresidential investment (i.e. spending by companies) has been essentially flat outside of information processing equipment that goes into newly constructed datacenters.  Spending on new homes and improvements to existing properties is similarly lackluster.  By one estimate, the current AI infrastructure build-out as a percentage of US GDP exceeds the dot-com era and continues to rise.

In conclusion, while we don’t know when the datacenter construction boom will end, given the three factors above, we think it’s a good time to look closely at your current portfolio with your advisor and determine whether any rebalancing or adjustments make sense at year-end and the beginning of the year.

[1] https://fortune.com/2025/09/28/ai-dot-com-bubble-parallels-history-explained-companies-revenue-infr…

[2] https://futurism.com/future-society/ai-hype-investment-subprime-bubble

[3] https://www.newyorker.com/news/the-financial-page/the-ai-boom-and-the-spectre-of-1929

[4] Oracle will have to borrow at least $25B a year to fund AI fantasy, says analyst

[5] Oracle Fear Gauge Closes at Highest Since 2009 on AI Worries - Bloomberg