July 2026 – Quarterly Letter

Since the debut of ChatGPT in the fall of 2022, according to a Morgan Stanley analysis, AI-related stocks were responsible for as much as 75 percent of the returns of the S&P 500. The dependency isn’t just limited to the stock market. During the first half of 2025, data centers accounted for 92 percent of America’s GDP growth, according to a calculation by Harvard Kennedy School economist Jason Furman.

As a percentage of the nation’s GDP, the AI infrastructure build-out is on track to exceed the construction of the American railroad system during the second half of the 19th century, the building of the Interstate highway system 100 years later, and the Apollo space program.

If you’ve spent any time in Austin, you are familiar with the frustration of our expressway traffic. There is nothing worse than sitting in a lane at a standstill while the other lanes flow freely. Occasionally, you might be tempted to make a quick lane change, hoping to capitalize on the seemingly impossible phenomenon that one lane will get you to your destination faster than the lane that is inexplicably going nowhere. Of course, as soon as you decide to move to the faster lane, that new lane comes to a standstill, and the lane you previously left becomes the fast lane.

Recently, the new era and old era segments of the overall S&P 500 Index have acted as though they are completely different species. They have become totally disconnected. Since April 17th, most of the time when new era stocks rise, old era stocks decline, and vice versa. Over the past four months, it’s as though the AI and non-AI segments of the stock market need a whole new classification — investor choices are now cash, bonds, AI stocks, or non-AI stocks.

As highlighted on the chart, beginning on April 17th, the sectors related to the AI buildout (Technology and Communication Services) surged by almost 27 percent, peaking on June 2nd. During this same period, the rest of the S&P 500 declined by about 2 percent. Then, from the June 2nd high until last Wednesday’s close after the FOMC meeting, S&P 500 new era stocks collapsed by 15.1 percent while old era stocks rose by 4.3 percent. It is so tempting to chase what has been working, but the reality is that when the tide turns, the speed of the sell-off is much faster than the pace of the rally.

For the first half of the year, AI and the build-out of the necessary data centers was the only theme that mattered in the market. Companies like ASML, Micron Technology, and Quanta Services became the new market favorites. Magnificent Seven stocks like Tesla, Meta, and Microsoft fell. Year-to-date, the Mag 7 stocks are down slightly while the S&P 500 is up almost ten percent.

I feel very good about the portfolio’s diversification. We are very modestly underweight AI relative to the S&P 500, but we will participate in any big rallies in the technology and communications sector. We are slightly overweight energy and industrials, which is basically an old-economy way to get exposure to the AI infrastructure buildout.

Some investors are concerned that the AI boom is starting to resemble the doomed internet bubble of the late nineties. On the surface, the comparison is reasonable. However, this time around, profits and positive operating cash flow are funding the build-out. Corporate earnings are much more supportive of the companies leading the market rally, and earnings are growing faster than expected. The consensus earnings estimate for the S&P 500 in 2027 is now $387, up from $358 at the beginning of the year. So, despite the market rallying almost ten percent so far this year, the S&P 500’s valuation is comparable to what it was in January.

The vibes today are completely different from the arrogance of late-nineties investors. Back then, it didn’t matter how sketchy the outlook was for an IPO. If it came public, and it had “dotcom” in its name, it was going to be a hot IPO — no questions asked.

Six weeks ago, SpaceX completed one of the most anticipated IPOs in history. The price of the initial public offering was set at $135 per share. After briefly rallying to a high of $225, the price has dropped more than 50 percent to a low of $104.83. Unlike 1999, investors are not yet willing to disregard fundamental valuation principles. SpaceX may have further to fall. Even at a price significantly lower than the IPO price, the current price-to-sales ratio is 35. Compare this to Google, which trades at nine times sales, and it seems hard to justify the current price of SpaceX.

We can’t know for certain how much longer the AI buildout will continue at this pace. All we can go by is what companies estimate their AI capital investment budgets will be for the next couple of years. The companies doing most of the AI buildout are what investors refer to as “hyperscalers.” These are the large technology companies that provide cloud computing. The big three are Amazon, Microsoft, and Google.

The second tier consists of Meta, Oracle, and Alibaba. In 2024, the combined capital investment budget for these companies was roughly $250 billion. By 2027, these companies are projecting to invest roughly $1 trillion into their cloud computing infrastructure.

At some point, the question will turn to whether all this capital investment makes economic sense. It’s way too soon to know what the return on AI investments will be. Will the AI build-out work better for Microsoft, Amazon, and Google, who are seemingly all in, or will the more patient approach of Apple pay off? Time will tell, but until then, I’m confident that, as always, diversification is the most prudent approach to portfolio management.