An AI Bubble?

As 2025 came to a close, it was obvious what would be the focus of the 1st quarter newsletter. It's the question on the mind of every stock market observer. But to answer this question (as best I can), requires a discussion of numerous factors. Whether or not AI stocks are in a bubble matters not just for tech investors, but for anyone with money in diversified stock portfolios. As such, I have dedicated this entire newsletter to this matter, with a few related tangents.

The Promise of AI

There is a lot of hype around AI, but also plenty of criticism. Many have pointed out the shortcomings of AI, noting spotty or erroneous pictures and video. What those detractors miss is the incredible rate of improvement of AI. The fact that the technology went from nothing to creating very realistic images in a short 5 years is astounding. More importantly, AI capabilities are improving at an exponential rate. The technology is able to do things today that were impossible a mere 18 months ago. So, while it may produce some comically bad content today, the most impressive creations will come in the very near future. This is not to mention the fact that the creation of images or video, while impressive, is only the tip of the iceberg of AI capability.

The most important indicator of AI's potential is its actual adoption. While some studies have pointed out that AI has been tested, but dropped, by many organizations, this is most certainly a result of either a haphazard attempt at adoption or the cost not yielding immediate benefits. But the largest and most successful organizations, and many individuals, are adopting AI tools at a breakneck pace, and with great success.

This pace of adoption is evidenced by the revenue numbers at the two largest "pure play" AI companies, OpenAI and Anthropic, both of which are private. OpenAI’s ChatGPT and Anthropic’s Claude are generative AI assistants—applications built on large language models (LLMs) that can understand and produce human-like text. OpenAI reportedly hit $5.5 billion in annualized revenue in December 2024. Less than 6 months later, that number had more than doubled to $12 billion. Anthropic reportedly hit $3 billion annualized revenue by late May 2025, and was approaching $7 billion by October. 80% of Anthropic's revenue came from enterprise customers, meaning the usage among businesses has been increasing rapidly. And we are only in the earliest stages of AI adoption.

It’s also important to recognize that many of the economic benefits of AI do not show up as a separate line item labeled “AI revenue.” In many cases, AI is improving existing businesses rather than creating entirely new ones. Better ad targeting can increase pricing and efficiency for advertising platforms. Productivity gains can reduce costs or allow employees to do more in the same amount of time. Improved recommendations can increase engagement and sales. These benefits are real, but they tend to be embedded across income statements rather than easily isolated, which makes AI’s financial impact harder to measure in the early stages.

Make no mistake, AI will be transformational.

The Data Center Boom

To achieve this exponential growth in AI capability will require the training of the LLMs, which takes a huge amount of "compute" (industry-speak for computing power). This, in turn, will require a truly massive buildout of data centers. We are seeing this in our own backyard. Aligned Data Centers is nearing completion of an AI data center on the former New Departure site in Sandusky. There are reports that even larger data centers in this area are on the horizon.

The data center numbers are staggering. According to private research, data center and AI-related investment accounted for about 80% of U.S. private GDP growth in the first half of 2025. Investment in 2024 was around $60 billion. In 2025 it tripled to $180 billion. By 2026, it's estimated to reach $370 billion; a six-fold increase in two years.

One reason the data center boom has attracted so much investor attention is that data center real estate is increasingly being treated as non-cyclical infrastructure, rather than traditional commercial real estate. Unlike office buildings or retail space, data centers serve mission-critical functions for their tenants. Computing capacity is becoming as essential to large organizations as electricity or logistics, which means capacity planning often takes precedence over short-term economic conditions.

This has led investors to re-rate data center properties—often through publicly traded data-center-focused REITs—at higher valuations than traditional real estate. The debate is whether this re-rating is fully justified. While long-term demand is clearly strong, data centers are still capital-intensive assets, and periods of overbuilding have historically followed every major infrastructure boom. If demand grows more slowly than expected, returns could disappoint even without a broader economic downturn.

When spending numbers get this large, the next logical question is whether enthusiasm has outrun fundamentals. That brings us back to the original question: if there is an AI bubble, where exactly is it?

Where is the "Bubble"?

If there is an AI bubble, it does not appear to be broad-based.

Much of the public attention has focused on NVIDIA (NVDA), yet its valuation looks far more reasonable than headlines suggest. NVIDIA’s forward price-to-earnings ratio is in the mid-20s, and when adjusted for growth, the stock trades at a PEG ratio below 1. In other words, investors are paying a relatively modest premium for extraordinary earnings growth.

Bubble concerns tend to concentrate elsewhere:

• A small group of “AI narrative” software stocks, such as Palantir (PLTR), trade at very high price-to-sales and price-to-earnings multiples. These stocks are priced for near-perfect execution, leaving little margin for error if growth slows.

• Certain data-center and AI-infrastructure suppliers have seen their valuations rise sharply as the buildout accelerates, raising questions about how durable current demand will be once the construction cycle matures.

• The loudest bubble accusations focus on private AI companies, such as the previously-mentioned OpenAI and Anthropic, whose valuations have climbed into the hundreds of billions of dollars without the same level of public financial disclosure.

Notably, this enthusiasm is concentrated in a relatively small number of companies, rather than the stock market as a whole.

The Blast Radius - Where This Leaves Investors

The AI debate isn't about whether the technology matters - it clearly does. The real question is whether today's prices reflect realistic timeframes for monetization. History suggests that transformative technologies often go through valuation overshoots without negating their long-term impact. The internet didn't fail because the dot-com bubble burst.

Importantly, today’s AI enthusiasm is far more concentrated than past market manias. Many of the companies leading the AI buildout are highly profitable, well-capitalized, and funding investment internally rather than relying on fragile credit markets. That limits the potential “blast radius” if expectations cool.

At the same time, the financial system itself is very different from 2000 or 2008. After nearly two decades of aggressive monetary intervention, central banks have demonstrated a strong willingness to support market liquidity during periods of stress. While corrections are always possible—and healthy—large, systemic market collapses tend to require a breakdown in credit or liquidity. At present, those conditions appear far less likely.

For investors, this argues for perspective rather than panic: understanding where risks are concentrated, staying diversified, and remembering that technological revolutions rarely move in straight lines—but they do tend to move forward.

How We’re Thinking About Portfolio Risk

Periods of rapid technological change often come with higher volatility. That doesn’t mean investors should avoid innovation, but it does mean portfolios should be constructed thoughtfully. As we assess the broader risks of a potential AI bubble, we'll be sticking to some tried and true investment strategies like value investing and diversification. Companies with strong balance sheets, durable margins, and real cash generation tend to weather periods of valuation adjustment better than purely speculative names.

Additionally, we've been leaning on so-called "defined-outcome" strategies, where appropriate. Specifically, for many clients, we use buffered ETFs from Allianz that are designed to provide participation in the market upside (up to a stated cap), while offering a level of downside protection (usually 10%) over a defined period. These strategies don’t eliminate risk, but they can help reduce the emotional and financial impact of market drawdowns, particularly during periods of uncertainty. They are one of several tools we use to tailor portfolios to each client’s comfort level and goals.

Epilogue: How FWM is Using AI

AI tools helped us research and synthesize the data used in this newsletter more efficiently, allowing us to evaluate multiple perspectives, verify figures, and focus our time on interpretation rather than information gathering. Last year, we also used ChatGPT to build a custom tax-planning model using Python, a powerful programming language. This enables us to evaluate Roth conversion strategies for each client across a wide range of personalized assumptions and scenarios. Building this level of analysis using spreadsheets would have been extremely time-consuming, and hiring developers to create a custom solution would have been cost-prohibitive. Instead, we were able to run the calculations using the Python model, which allowed us to focus our time on thoughtful planning and meaningful client conversations.