AI
First Quarter 2026
“At the root of all financial bubbles is a good idea carried to excess.” – Seth Klarman, Baupost Group
As 2025 came to a close, U.S. equity markets completed a remarkable three-year run of double-digit returns. The most recent years’ returns were driven by a resilient economy that weathered significant mid-year tariff shocks and advanced largely on the back of Artificial Intelligence (“AI”) investment. On the fixed income side, bond markets recalibrated to lower short-term rates, while long-term rates such as 30-year treasury yields remained unchanged.
The surging investment in AI has propelled technology stocks higher and fueled broader economic activity over the past several years. The rise in the prices of AI-related stocks has worried many market participants and continues to fuel talks of a speculative AI “bubble,” much like the internet bubble many of us lived through in the late ’90s.
While concern over roaring stock prices is always warranted, we felt some perspective on the nature of market bubbles might help shed a different, more favorable light on current equity markets.
“Are you a Good Bubble, or a Bad Bubble?” (Think Glinda in “The Wizard of Oz.”)
Not all stock market bubbles are inherently detrimental to society over the long term. While all market bubbles do feature a downside, with investors losing money on the back end when the bubble “pops,” the root of the bubble itself really defines whether it could be considered “good” or “bad” for society in hindsight.
Taking a page from Oaktree Capital Management’s Howard Marks’ recent commentary, some bubbles have actually been beneficial for the rapid advance of technology.
- Good – The 1990s tech bubble, 1920s radio/automobile bubbles:
In both instances, new (or newer) technology was seen as a product or service that could be widely adopted, and investor funds rapidly flowed into these spaces. In both periods, infrastructure such as transmission towers or fiber optic cable was laid across the country, and industry was built out to rapidly take advantage of the potential opportunities. Of course, in both cases, capacity eventually outstripped demand, and many investments came crashing down, but there is no doubt that this rapid acceleration of infrastructure laid the groundwork for lasting productivity and innovation in the subsequent decades. - Bad – The real estate bubble of the early 2000s:
The real estate bubble that peaked in 2007 and fueled the Great Recession from 2007-2009 did not have its roots in any particular new form of technology. Rather, this bubble was simply financial engineering and leverage run amok. Mortgage-Backed Securities and Collateralized Debt Obligations (and the further financial creatures these spawned) made large banks heavy profits, which led to excess investment and foolish amounts of leverage. When the party was over, no new technological breakthrough or industrial process had emerged.
There remains little doubt that AI is going to be transformative to society, in one way or another. Clearly, the increase in productivity that has already begun to emerge, allowing companies to do more with less, can be seen as positive. However, as with any new technology, there may be current and future downsides. Fewer job opportunities for young workers and higher utility bills due to data center usage are issues that immediately stand out.
For the time being, capacity is still constrained in many areas of AI. That’s to say that demand for the product remains high and physical constraints like electricity are still an issue, suggesting that recent infrastructure investment, however intense, has not yet resulted in overbuilding.
Productivity Rises
That strong demand for infrastructure is what is driving AI capital investment currently. Three to four years ago, the focus for AI development was on new software and on confirming that this revolutionary tool could actually increase productivity across large swaths of the economy. Some of this increase seems now to have been confirmed. While improved and automated customer service chatbots and software coding AI benefits are largely known, many other real-world examples now exist:
- Walmart’s AI inventory management system led to reductions of out-of-stock incidents by 30%.
- Siemens found that companies using its Senseye Predictive Maintenance reduced maintenance costs by 40% and decreased downtime by 50%. This program employs machine learning and historical data to detect issues from subtle pattern anomalies in equipment use.
- DHL reported that its AI-driven supply chain optimization reduced logistics costs by 15% while improving delivery times.
Overall increases in productivity generally mean higher margins or growth for companies, leading to higher earnings, a key driver in stock prices.
The rise in overall U.S. productivity growth over the past several years may be linked to more rapid and flexible adoption of AI over its European counterparts, who historically have been slow to adopt newer technology and instead attempt to regulate it first.
With productivity increases confirmed, AI companies have turned to the next phase of the buildout over the past 12-24 months: infrastructure. The current data center buildout has been unprecedented. While prior spending on software was only somewhat costly, the infrastructure built out has been far more expensive. This has led to some concern about the interconnected deals and debt financing used to raise the needed capital.
Debt Enters the Chat
The price of the data center buildout has now eclipsed the operating cash flow of many of the biggest players in the AI race. Several of these companies have burned through much of the cash on their balance sheets and resorted to debt financing or other methods in order to keep up with the competition and avoid being left behind:
- Meta has turned to Special Purpose Vehicles, or “SPVs,” to fund large data center buildouts. One example is the 5 GW Hyperion data center in Louisiana, in which a private credit firm, Blue Owl Capital, will own an 80% stake. Meta will operate and effectively run the center, but Blue Owl’s ownership allows Meta to keep the debt off of its balance sheet.
- NVIDIA has invested billions into young “neocloud” providers like CoreWeave and OpenAI, who in turn use much of that capital to buy NVIDIA chips for their datacenters—which are years away from any kind of positive net income. This “circular financing” is reminiscent of similar behaviors during the tech bubble.
Much of the recent flood of capital into the AI space has been funded by private credit funds, which experienced a boom since the financial crisis, as governments had put more requirements on traditional bank lending after 2008. This area, which encompasses non-bank lenders and direct lending funds, has grown rapidly as borrowers seek alternative sources of financing and investors pursue higher yields. The trend has diversified the financial system but also shifted considerable risks outside the purview of traditional regulation.
The Show Goes On….
For the time being, the massive demand for AI by companies in all areas of the economy continues to convince the large AI spenders (Meta, Microsoft, Amazon, Alphabet, and others) to keep the pedal to the floor so as not to lose their place in the AI ecosystem.
At Mitchell Sinkler & Starr we continue to be mindful of the risks inherent in these trends but can also see firsthand how AI has increased productivity, as we are now utilizing it on a more regular basis at our own firm. Our focus continues to be to ensure our clients can participate in this economic growth while still prudently reserving assets for potential future downturns.
– Mitchell Sinkler & Starr’s Portfolio Managers
Bubble Watch Update
We intend to update this table more regularly for our clients, in order to help quantify potential concerns of an equity market bubble.
As it was during the last quarter, valuations for the top companies in the S&P 500 remain stretched, but not at the extreme levels seen in 1999 (with the exception of Tesla.) Concentration remains a potential issue however, as investors who blindly buy the S&P 500 may not receive as much diversification as they may be looking for.
Economic and Capital Markets Data
Decade-by-Decade Market Returns
