From Savanna to Silicon: The Red Queen Meets AI

For thousands of years, cheetahs have hunted gazelles on the African savanna. Over the millennia, cheetahs evolved traits for raw speed while gazelles evolved traits for lateral agility. In the end, their adaptations offset and neither side gained a durable competitive advantage.

In 1973, biologist Leigh Van Valen proposed the Red Queen hypothesis to explain these evolutionary stalemates. The phrase comes from a scene in Lewis Carroll’s “Through the Looking Glass,” in which the Red Queen explained to Alice that in her country “it takes all the running you can do, to keep in the same place.”

Importantly, these evolutionary advances also came with some drawbacks. As cheetahs got faster, their turning radiuses got wider. Springy tendons made gazelles more nimble but less suited for foraging.

While this concept is more likely to be found in a biology textbook than an economics course, the Red Queen effect has real-world applications in business, and we are witnessing this play out in the current artificial intelligence (AI) infrastructure build-out. A group of mega-cap tech companies (along with a few large venture-backed startups) have committed hundreds of billions of dollars that are expected to cascade through the AI ecosystem. The question is: what will the returns on those investments be?

The firms leading the AI build-out are carefully calculating their own returns on invested capital, but they seem to be overlooking how much capital their peers are deploying. Each individual company may be acting rationally; however, the group’s collective spending decisions have an impact on the potential returns and the odds of success for each participant. One possibility is that these investments merely help them retain market share rather than grow it. Like the cheetah and the gazelle, these companies are attempting to adapt, but the competitive responses they are triggering in each other may neutralize any relative benefits.

The trade-offs these companies are making to participate in this arms race are real. For the last decade, these businesses have been cash-generating machines with fortress balance sheets, high margins and capital-light business models. Shifting their focus to AI will require massive capital investment, entail uncertain margins and involve direct competition with more than a handful of deep-pocketed rivals. Why would a rational operator pursue this strategy? In short, no one wants to risk getting left behind. No CEO wants to be remembered as the leader who missed a major platform shift. So instead, they’ve stepped onto the Red Queen’s investment treadmill, and it just might take all the running they can do to keep in the same place.

Sometimes standing on your tiptoes doesn’t help

There are many historical examples of this concept. Forty years ago, Berkshire Hathaway’s textile business was going through a period of substantial technological change. In his 1985 letter to shareholders, Warren Buffett explained the dilemma of being forced to choose between falling behind or irrationally deploying capital into an industry where returns were being driven below the cost of capital, writing:

“Viewed individually, each company’s capital investment decision appeared cost-effective and rational; viewed collectively, the decisions neutralized each other and were irrational (just as happens when each person watching a parade decides he can see a little better if he stands on tiptoes). After each round of investment, all the players had more money in the game and returns remained anemic.”¹

Scale now, profit later

If that example feels like ancient history, consider the more recent streaming wars that kicked off roughly six years ago. Within 18 months, Disney, Peacock, Paramount and Apple all launched new streaming services. As each new product launched, investors cheered as low introductory prices attracted subscriber numbers that exceeded initial expectations. Emboldened by their initial success, streamers expanded their marketing budgets and poured billions of dollars into creating original content. Bidding wars erupted for top talent, with showrunners like Taylor Sheridan, Ryan Murphy and Shonda Rhimes signing deals that were each reportedly worth more than $200 million.

The prevailing logic at the time was scale now, profit later. Streaming was thought to be a once-in-a-generation platform shift, where whoever amassed subscribers the fastest would lock in an unassailable competitive position. The accepted wisdom was that this was a winner-take-most market, which further justified significant upfront spending and big, bold bets. If any of this sounds like the justifications you are hearing about the AI infrastructure build-out today, you’re not alone.

With the benefit of hindsight, it’s clear that the streaming industry went through a period of irrational exuberance that led to higher costs, an oversupplied market, lower returns and more competition. And what did it have to show for all that? After the dust settled, the relative positioning of the participants remained mostly unchanged, and the optimistic estimates of the addressable market had to be revised sharply lower.

Why are we skeptical of the AI infrastructure boom?

Just like Netflix emerged as a winner in the streaming wars, we expect winners to emerge from the AI infrastructure race. However, the expectations embedded in today’s prices appear to reflect a very rosy future and imply that it is nearly certain this boom will deliver both significant growth and monopoly-like returns to its winners. That scenario is possible, but unlikely in our view.

Another potential error we see is that investors are capitalizing the earnings of suppliers involved in the initial build-out phase as if these profits represent through-the-cycle normalized earnings. In other words, they are putting peak multiples on peak earnings at peak margins. In doing so, we believe they are mistaking a cyclical boom for a secular shift.

It’s hard to grasp the scale of capex currently underway. Harvard economist Jason Furman calculated that information processing equipment and software accounted for 4% of GDP and roughly 92% of the incremental growth in the first half of 2025. We’ve also seen estimates that AI investment is comparable to the levels we saw during the height of the internet bubble. The exact figures can be debated, but there’s no doubt the magnitude is staggering.

To be clear, this spending is real, it is happening now and it is already affecting the real economy. As long as there is abundant capital that is willing to invest in physical infrastructure, it is nearly certain that we are going to see profound impacts on the economy. What is far less certain is exactly how long capital will stay cheap and plentiful, and whether the returns on that capital will justify the investment. The uptick in creative and leveraged financing structures to support the rapid expansion may be the first sign of reticence from investors. No one has summarized the recent deal delirium better than JP Morgan’s Michael Cembalest, who observed that:

“Oracle’s stock jumped by 25% after being promised $60 billion a year from OpenAI, an amount of money OpenAI doesn’t earn yet, to provide cloud computing facilities that Oracle hasn’t built yet, and which will require 4.5 gigawatts of power (the equivalent of 2.25 Hoover Dams or four nuclear plants).”²

While every cycle differs, past investment booms ranging from railroads to energy to internet infrastructure provide insights into how the current cycle may unfold. First, overinvestment tends to create excess capacity. Second, competition typically intensifies as entrants crowd into similar opportunities.

Taken together, these factors can weigh heavily on near-term returns on capital and on the ultimate return investors hope to realize from the surge in capital spending. That doesn’t mean that all the investment will be wasted. Rather, it underscores the need for prudence in the prices we are willing to pay for these businesses. We are bullish on AI’s long-term potential, but the path to progress typically involves one or possibly even two major downturns along the way. If that pattern holds, it would meaningfully dilute the rosy return scenarios priced in at the top of the market.

Perhaps we’ve buried the lead, but as you may have gathered by now, we have limited exposure to AI infrastructure beneficiaries in our Large Cap and Large Cap Concentrated portfolios. That is an intentional decision we’ve made to ensure we don’t expose our clients’ capital to the inherent risks that accompany this very difficult-to-predict spending cycle. We’ve owned many of these businesses in the past and would gladly own them again when growth is reasonably priced, and the shares trade below our estimate of intrinsic value.

In the meantime, we have 1,000 companies in our universe whose fundamentals we are studying carefully. We are finding some companies, like Colgate, that we believe are capable of delivering double-digit total returns³ without a change in their earnings multiple. If we can populate the portfolio with companies delivering these sorts of results, it will be well positioned to succeed over the long run. We are committed to staying true to our investment philosophy; that said, we are open-minded enough to acknowledge that if we are wrong about the fundamentals of this AI capex cycle, we are willing to change our minds when the facts change. In the meantime, while others are joining the crowd and standing on their tiptoes, we’re choosing to seek out a better vantage point with a clearer view. Our job as owners is to fund growth only when incremental return on invested capital comfortably exceeds the cost of capital. Revenue growth without that spread is activity, not value creation.