Worldwide All Cap

The artificial intelligence boom: are the risks of disappointment growing?

“We can see three years’ worth of data-centre visibility. We just don’t know how they are going to build all that stuff. We can’t get the people to build it, let alone the water and power.”

- (CEO of a US industrial company during a meeting with Stewart Investors. September 2025.)

It has been nearly three years since ChatGPT astonished the world and triggered a wave of investment in the infrastructure needed to support the rollout of artificial intelligence (AI) models. During that time, the share prices of those companies that are building much of that infrastructure, so-called ‘hyperscalers’ such as Microsoft, Meta, Alphabet and Amazon, have risen sharply. Companies who provide the components of the data centres required to train and run the AI models have also enjoyed significant gains. Nvidia has the highest profile of these companies but the past three years have also seen sharp gains for the likes of TSMC, which manufactures advanced semiconductors, and Arista Networks, which makes networking switches.

One consequence of this investment boom is that movements in stock market indices have come to be dominated by the fortunes of fewer and fewer stocks. Such increases in market concentration have happened before, typically when the excitement surrounding a new technology has sparked an infrastructure boom. A similar thing happened with railways in the nineteenth century and then again with the internet 25 years ago. As investors grow enthused by the promise of a new technology, they become willing to project today’s rapid growth persisting into the far-off future and to pay progressively more for those future earnings. While earnings continue to rise, this makes sense. But as market concentration increases around a single theme (AI in this case), and as companies come to be valued at higher and higher multiples of their future earnings, the risks of disappointment grow.

The long-term beneficiaries of today’s AI investment boom will not necessarily be the hyperscalers

In the current boom, one risk arises from the fact that the real world will eventually begin to impose constraints on the number of data centres that can be built. This was highlighted to us by the management teams of some of the companies we met during our recent trip to Texas, one of the states at the epicentre of the data centre boom.

Another risk arises from the fact that it is unclear who will benefit the most from the infrastructure investments being made today. In the railway and internet booms, the financial benefits did not accrue to the companies that first built the infrastructure, but rather to those businesses that were subsequently able to use it in ways that weren’t fully appreciated when it was being built. The profits Netflix and Meta are making today arguably rest on the lossmaking investments that telecoms companies made in laying fibre optic cables during the last technology boom. At this point, no large revenue streams from the end users of the current generation of AI tools have been identified. The hyperscalers are making the bulk of the investments in physical infrastructure – but we are conscious that they built their dominant positions in a world of low interest rates and flourishing advertising revenues.

As market indices narrow, we invest in a diverse range of companies that are benefitting from a broad range of long-term tailwinds

How should long-term, fundamental investors approach the risks, opportunities, and uncertainties being presented by the AI boom? Because our definition of ‘risk’ is the permanent loss of our clients’ capital rather than in diverging from a benchmark index, we are prepared to relinquish some potential gains in exchange for reducing the risk of loss. And although we certainly own some companies that are benefitting from the current wave of investment in AI and data centres, our investment philosophy is based on the premise that the best way to deliver long-term returns is to hold a diversified portfolio of companies. The companies we invest in address a range of unmet needs and we believe they will grow their cashflows and earnings over a decade or more by meeting those needs. We established two new positions during the quarter, both of which we believe meet that definition.

Bajaj Holdings & Investment (India: Financials) is stewarded by the Bajaj family and acts as the holding company for their investments in auto makers and financial businesses. Those underlying companies tend to be leaders in their sectors and are helping to meet India’s needs for finance, housing, mobility, and cleaner transportation. As such, we believe these underlying companies have many years of growth ahead of them.

The second new addition is Spirax Group (United Kingdom: Industrials). We have invested in this company before but sold our position because its end markets appeared to be shrinking at a time when its valuation left little room for error. Over the past 18 months, however, the company’s shares have become more reasonably valued, excess inventory is showing signs of being cleared, and demand in Spirax’s end markets is growing again. Together, we believe this points to healthy long-term returns for its shareholders.

Elsewhere, we topped up our existing position in cybersecurity specialist Fortinet (United States: Information Technology). Its share price fell in August when it signalled that the current upgrade cycle for firewalls was more advanced than realised, implying that future sales of upgraded components could be smaller than expected. The company subsequently increased the size of its share buyback programme, indicating that it thinks the weakness in its share price represents a great buying opportunity.

We funded these purchases by selling holdings in four companies. We have owned shares in HDFC Bank (India: Financials) for several years and it has delivered strong returns in that time. However, because this strategy has the flexibility to invest in companies with lower market capitalisations, it can take advantage of opportunities to invest in younger, smaller companies when they arise. This includes Bajaj Holdings & Investment, where we believe the balance between risk and potential reward is more attractive.

We sold Roche (Switzerland: Health Care) in recognition of the increased risks and uncertainty faced by pharmaceutical companies as the United States reinvents its policies on drug pricing and healthcare. To manage the strategy’s exposure to companies that rely on continued rapid growth in demand for semiconductors, we sold TSMC (Taiwan: Information Technology) and trimmed our position in Samsung Electronics (South Korea: Information Technology). We sold our position in Spectris (United Kingdom: Information Technology) having benefitted from the price appreciation after a private equity firm made an offer to buy it.

We focus on delivering absolute returns in a relative world

Fear of missing out – ‘FOMO’ – is not confined to the young; it is a natural human instinct. We are conscious that our style of investing, which is designed to deliver compelling absolute returns over the long term (a decade or longer) and to minimise capital losses, has not kept pace with returns from global indices over the short term. We don’t believe that means our philosophy of investing in high-quality companies for extended periods of time has suddenly ceased to work. At a time of change in markets, trade and geopolitics, our underlying approach remains consistent: we continue to focus on generating attractive returns over the long term rather than attempting to outperform through every short-term period. We look forward to demonstrating the fruits of that approach over the years and decades to come.