AI(ccident) Waiting to Happen

January 2026

We invest exclusively in companies with qualities that support robust and long-term value creation. A natural consequence of this is that there are many industries we avoid. One of them is AI, which during MIP's lifetime has generated massive media coverage and significant returns. By partially replacing human cognitive tasks with data centers and highly advanced chips, the technology is revolutionizing many central routines and processes in how we live and work. This has led to significant share price increases for the companies at the forefront of developing this technology. Despite this, we have not found investment opportunities in the expansion of AI that possess the characteristics we seek. This statement may seem peculiar given the potential and the growth AI companies have experienced in recent times, and it may prompt the question: is the MIP portfolio missing the investment opportunity of the century?

In hindsight, one could argue that we have made a mistake by not investing in some of these promising and popular companies. The reason is straightforward: we stay within our circle of competence with a focus on long-term returns and avoid speculation despite the appeal that speculation can have when the market is favorably disposed toward a trend or new technology. The potential for AI as a technology is astounding, but in our opinion, the conditions for predictable value creation for shareholders in AI companies are not in place.

In our assessment, the euphoria surrounding AI stems from the fact that the technology has already had a significant impact on our society and will continue to affect how we work and live. However, two centuries of technological progress show that transformative technology does not necessarily guarantee strong returns for investors in those technologies. Historically, the certain winner in such situations has (fortunately) been society, but unfortunately often at the risk of significant losses for those who financed the expansion of the new technology.

The historical precedents for our cautious approach to new technology are extensive. They include the 1840s British railway bubble, the buildup of the internet infrastructure during the late 1990s as well as industry-specific advances such as shale oil and gas extraction.

In 1840s Britain, railways were seen as an infrastructure breakthrough that enabled faster transport and opened new markets. The enthusiasm turned into an investment race with hundreds of railway companies being founded. The result was that many projects were planned simultaneously, often with optimistic assumptions about traffic. Investment levels grew explosively from 1843 to 1846, but a third of approved projects during the bubble were never realized, and investment levels subsequently fell by 70% in 1850 compared to the peak in 1846. Looking at graphs of British railway company indices from that period, they show that share prices from the bubble's peak were not recovered in that century, underscoring the horrible returns that investors received, even though the railways were here to stay. Similarly, the United States witnessed several similar railway crises at the end of the 19th century. The railways changed the world, but in neither Britain nor the United States were they a golden (train) ticket to high long-term returns for the vast majority of investors who invested in the expansion.

In the late 1990s, the internet was perceived as the premier investment opportunity of the era. Massive investments were made in fiber optic cables, network equipment, and server capacity. Many players invested simultaneously because they wanted to be first and largest. In the United States, this led to capacity utilization among internet infrastructure suppliers reaching nearly 90% in 2000. In the following two years, that level fell to 50%, and the number of companies in the sector with payment problems rose from a few percent to 30%. From an investment perspective, the sector was a goldmine in the years leading up to 2000, when almost all internet-related stocks experienced significant price increases. However, if one invested at the peak of the euphoria, based on the Nasdaq 100 index, it took 15 years before returns reached positive territory. During those 15 years, internet usage continued to grow explosively. Technology once again changed the world, but it was yet again not a golden ticket to high long-term returns for the majority of investors who invested in its expansion.

In more recent times, shale oil and gas extraction technology is an industry-specific example of a significant technological advancement that made enormous American oil and gas resources extractable at scale. This led to a focus on volume over profitability in the industry for several years, with many companies investing aggressively in new wells in the years leading up to 2014, expecting that scale and technological advances would make the investments profitable. This development led to 120 bankruptcies in the industry in the United States during 2015 and 2016 alone. The industry ended the period from 2010 to 2019 with cumulative negative cash flows of over $250 billion and write-downs of approximately $450 billion. Investors in the sector achieved good returns in the years leading up to 2014, but since then those returns have been lost, and today an investment in the sector from 2014 is still far from a positive return.

Is there something in the present that sounds familiar from these stories? We think so. The common denominator for such promising technologies with subsequent poor returns for investors is the massive flow of adventurous and hopeful capital that pours into emerging sectors and industries — be it oil, railways, the internet, or AI.

A recurring pattern is that heavy investment is required in the present to harvest uncertain returns in the future. This results in a race to build the most infrastructure to maintain market share despite the uncertain or absent future profitability of this expansion. In that scenario, most executives choose to dance while the music plays rather than lose their jobs.

Furthermore, most of the investments were also undifferentiated. Railways built parallel to one another, oil or internet access are all services or products that suffer from competition with near-identical products. When it comes to AI, one need only look at the broad selection of both Western AI products — ChatGPT, Claude, Gemini, Copilot, Grok, Mistral, and others — and recent Chinese models — DeepSeek, Kimi, GLM, Yi, and Qwen — to observe that competition in that part of the industry is already extremely intense.

AI-ctual Problems

In our experience, heavy capital commitment in unpredictable and competitive industries is a sure recipe for poor long-term returns. Currently, the switching costs of moving from one AI model to another are very low, which means that price competition is expected to be high. In the long run, this will affect the companies that are building out data centers. If data centers become unprofitable and the owners behind them stop investing, it is typically the suppliers who depend on ongoing product sales (including NVIDIA) that are most vulnerable.

The biggest issue we see in the short term is that the expansion of AI is highly dependent on revenue that has not yet materialized. The investment bank, UBS, expects $571 billion to be spent on AI infrastructure in 2026 alone. This involves equipment that will very likely become obsolete within a few years, unlike railways or internet cables. If this level of investment continues, based on a rough and simplified estimate, AI infrastructure would need to generate annual revenue between $1 to 1.5 trillion to achieve a 10% return. By comparison, Microsoft's annual software revenue today is just under $100 billion.

Another issue is that even if these sky-high expectations are met, the current valuation of AI companies is such that even good results do not necessarily lead to extraordinary returns. To illustrate, for investors buying NVIDIA today, its current valuation implies the company must generate $9 trillion in cumulative future revenue simply for them to reach a break-even return. Since NVIDIA chips often constitute approximately 50% of investments in a data center, this would mean that data centers worth over $18 trillion would need to be built for the math to work out. By comparison, approximately $260 billion was invested in renewable energy in North America, Europe, and Australia in 2024. In other words, the aforementioned $18 trillion is equivalent to 70 years of investment in renewable energy in the Western world, which notably would be used solely to build data centers with NVIDIA chips.

During MIP's lifetime, it has been poor for returns not to have exposure to AI. Whether this choice will continue to have a negative impact in 2026, no one knows, but in the long run, returns are determined by earnings growth. We leave it to others to speculate which AI industry players will successfully defend their market positions and generate value-creating returns in the long run. The predictability is close to zero in our view. As in previous periods of technological progress, we also expect that the economic gains from this will largely accrue to society, as capitalism's fierce competition will ensure that most companies will pass any efficiency gains on to customers in the form of lower prices. The companies that will benefit most from AI are likely those with already well-protected and scaled business models. These companies can use the technology to make internal efficiencies and redistribute resources to more value-creating activities without industry competition causing the gains to be largely passed on to customers.