Emerging Markets Equities and our thinking on the impact of AI

Given these phenomenal returns, we have been asking ourselves, can this investment boom in AI data centers continue? And are we seeing an AI led bubble in equity markets?

Aeroplanes without the wings!

To answer these questions, we would like to draw our readers attention to another industry that transformed the world: aviation. As shown in Exhibit 2, the famous first powered flight by Orville Wright of the Wright Flyer in 1903 truly changed the world. In 2026, we take long haul flying for granted, but in 1903 looking at the Wright Flyer and conceptualising not only the implications of how it would change the world, but also the technological advances in airplanes would have been impossible.

Likewise, AI is likely to change the world and develop in ways that we cannot imagine – this we do not dispute. The aviation industry however offers a cautionary tale, as the technological wonder that is powered flight doesn’t mean you have wanted to own airline stocks!

Since its very beginning aviation has been an industry with “blood on the floor” as airlines continually fail to meet their liabilities. We note that since 2000, fifty different airlines have sought bankruptcy protection in the US alone. There seems to be no end to the adventurous spirits willing to lose money on flying airplanes for the benefit of the public. Whilst AI is a different industry with its own quirks, we believe it has three characteristics that are like airlines:

Highly Capital Intensive – according to Bernstein Research, constructing a 1 GW AI data centre facility costs around US$35 billion. Approximately 60% of this cost is for the semiconductor chips, and as the price of these chips increases, the costs of the data centres have also continued to increase. Some estimates put the cost of building an AI capable data centre currently at US$50-60 billion per 1GW. As shown in Exhibit 3 the scale of these facilities is gigantic. For example, Meta’s proposed Hyperion data centre will be four times the size of New York’s Central Park!

Ability to Stimulate Demand – if you want to attract customers to your airline / data centre, dropping the price of the tickets / AI models is a pretty good way to do it. We can see this impact in Exhibit 4, where several AI providers are forced into offering large language models for a lower cost as they try to build scale and market share or were quickly superseded by a newer model.

Large Incentives to Discount – earning just $10, for example, for an hour of processing time in a data centre, even if this doesn’t cover the costs of operating let alone the depreciation cost of the data centre, is better than that server being unutilised (analogous to an airline seat being empty).

It’s not hard to see what will happen in such an industry structure when there is excess supply. Everyone is incentivised to cut prices to stimulate demand even if this means operating at a loss. With very large capital costs often funded by debt, industries with such a structure are also prone to firms periodically going bankrupt.

Without an ability to differentiate their AI offerings through innovation or technology, the AI companies face a dilemma. Either continue to invest in huge and growing processing capacity, or quickly watch their AI offering lag competitors, which puts their existing non-AI cloud businesses at risk. This is a classic ‘prisoner’s dilemma’. In fact, the bigger and faster you can invest in processing capacity, the more likely you are to be the eventual ‘winner’. We are in fact starting to see a new phase in the AI industry where investment in AI processing is growing at such a rate it is cannibalising the existing profitable businesses of some of these companies. For example, Oracle is forecast to see negative free cashflow in 2026 and Meta is forecast to be barely cashflow positive by the consensus of broker analysts (assuming 2026 AI revenues meet expectations).

How will AI be monetised?

“For every action there is an equal and opposite reaction” – Issac Newton, investor who lost his fortune in the South Sea Bubble of 1720.

The Northcape Emerging Markets Equites investment team has been considering what this huge increase in AI processing capacity means. Will all this additional processing power drive rapid improvement in AI models and new use cases that will drive unprecedented profits for the enablers of AI, or will it lead to oversupply and cutthroat competition?

There are some very high value uses for AI, for example Demis Hassabis and John Jumper from Google DeepMind won the 2024 Nobel Prize in Chemistry for using AI to uncover the structure of all 200 million proteins known to science. This discovery is potentially revolutionary for the development of new and novel drugs. Companies such as TSMC are also utilising large quantities of their internally generated data to improve the efficiency of their very complex manufacturing process. We also note AI is already being used in areas such as medical diagnostics.

In addition, we are seeing AI being used in more prosaic applications such as in automation of call centres, computer programming, search functions, and document creation. What we can see is that the value created by AI can vary greatly, with what appears to be some very high value applications for AI, and more what we would term “mass market” or “economy class” applications. The AI providers however are offering a commoditised product and therefore are unable to differentially charge customers by the value they can generate or differentiate their AI services from other providers.

Given the competitive dynamic of the industry is to ‘fill the seats’ regardless of price, this means the price of AI services is likely to be set by the marginal buyer. To understand this dynamic, ask yourself these questions, “How are we currently using AI in our business, and how much would we be willing to pay for this?”. We have posed these questions to companies we have met, and the majority are seeing AI implemented slowly, and for lower value functions.

It seems that customer willingness to pay higher prices for AI services will therefore be limited, particularly given they can shop around amongst providers. As shown in Exhibit 6, the responses for corporate management we met aligns with signs that AI adoption rates are slowing. Researchers at the US Census Bureau who surveyed firms’ use of artificial intelligence “in producing goods and services” found that AI usage has stalled at around 11% of the US workforce, with adoption falling sharply at the largest businesses (>250 employees).

We therefore strongly suspect that even if new applications for AI are developed and existing models improve, the huge increases in capacity within the AI industry will lead to escalating competition for customers and intensifying price competition. There is strong evidence that this is already occurring, with AI data centre rental rates falling (see Exhibit 7), whilst at the same time the capital cost of AI data centres is skyrocketing due to the severe shortage of semiconductor chips. As increased capacity comes online over the next two to three years this trend of increasing competition could in fact accelerate.

Over the Capex Cliff?

Given the level of proposed capital investment, Bain Capital recently estimated total industry revenues of US$2 trillion by 2030 will be required to justify the expenditure. Current estimates of the revenues being generated by AI vary widely – anywhere between US$40 billion to US$250 billion. Despite this high level of uncertainty in the current state of the industry, it is clear usage of AI services will need to start growing very rapidly to justify the capital investment being poured into the construction of data centers.

However, what we appear to be seeing is AI adoption rates stagnate. The need to “fill the seats” in the data centres (i.e. maximise utilisation of the very expensive AI servers) if demand doesn’t grow materially will potentially lead to cutthroat competition slashing revenue growth and profitability for the AI operators. We suspect that the proposed build out of AI data centers over the next five years is at high risk of not fully materialising.

We have therefore reduced our exposure to stocks linked to the construction of data centers as well as stocks within the semiconductor supply chain.

AI investment’s economic impact

Several leading economists have concluded that US GDP growth in the first half of 2025 was almost entirely driven by investment in data centers and information processing technology. This includes Harvard economist Jason Furman who calculated that excluding these technology-related categories, US GDP growth would have been just 0.1% on an annualised basis in H1 2025; a near standstill that underlines the increasingly pivotal role of high-tech infrastructure in shaping macroeconomic outcomes. The risks from a reversal in data center investment on the broader US economy is therefore substantial. This is particularly the case given weak consumer sentiment, which in the past few months has hit lows last seen during COVID, limiting the scope for a consumer led recovery (see Exhibit 8).

A slowdown in data centre investment, if it were to occur, could be a major factor in slowing US GDP and slowing inflation (rising energy costs have been linked to surging demand for power from data centers). In turn, this would lead to a decline in Federal Reserve policy rates and downward pressure on the US dollar.

China’s data centre risks

The other major economy seeing substantial investment in AI data centers is China. It is hard to obtain accurate data on expected data center investment in this market, but we suspect it is substantial. Again, any slowdown in Chinese data center investment could have serious economic impact for the Chinese economy given other major drivers of economic activity are either depressed or showing signs of deceleration. This includes continued weak consumer sentiment, the Chinese housing market activity remaining at 20-year lows and a major deceleration in fixed asset investment in China which has recorded negative YoY growth in the last two months.

Learn more about Warakirri Asset Management

For more information on Warakirri Asset Management and their funds and strategies visit their website.