‘Is it an AI bubble?’ is the wrong question

IF EVERYONE is asking whether artificial intelligence (AI) is a bubble, it is likely the wrong question. The focus on whether AI is overhyped distracts from the larger, more critical issue – the misallocation of capital and the physical constraints that limit growth.

Interest rates are the mechanism by which societies allocate capital and resources. Every financial decision must hurdle the risk-free rate. When capital is not allocated based on its economic utility, inefficiencies arise, leading to economic and financial market distortions – eventually manifesting as bubbles.

The artificial suppression of capital costs, which began in the early 2010s, accelerated the diversion of capital from the real, physical economy to a “paper economy” (for instance, stock buybacks and financial engineering).

While outsourced manufacturing and falling fixed investment generated immense wealth for owners of capital, they had minimal actual economic utility. The resulting cost is clear today: a lack of sufficient US housing stock, as well as a deficit of goods and skilled labour necessary to construct the foundational infrastructure that the AI revolution demands.

A tale of two economies

The AI buildout is progressing at a staggering pace, yet the broader macroeconomic environment feels recessionary. The Federal Reserve is cutting rates in response to a weak labour market. Credit weakness is evident in credit cards, car loans and private debt. Oil prices – a common proxy for growth – are down double-digit percentages year-to-date, signalling excess economic capacity.

However, the AI sector tells a different tale, one of acute capacity constraints. Makers of mission-critical components – such as high-bandwidth memory chips, advanced gas turbines, very large electrical transformers and grid interconnection equipment – are sold out, with lead times measured in years, creating construction delays.

Shifting funding dynamics and the path to profitability

Until recently, AI investment was primarily funded by two sources: massive cash flows from hyperscalers (such as Alphabet, Amazon and Microsoft) and private capital, which has backed model providers and neo-clouds.

However, the funding landscape has clearly shifted, moving towards public debt markets, asset-backed vehicles and vendor financing. The latter, where suppliers essentially act as both seller and lender to fund customer purchases, can sometimes mask financial health and true default risk.

These changes highlight two critical points about the AI paradigm:

SEE ALSO

Nvidia aims to begin H200 chip shipments to China by mid-February

1. The scale of AI investment is larger than anticipated. Even the largest hyperscalers are seeing their free cash flow decline as they fund the immense capital expenditure requirements of AI development. The deliberate shift to debt financing may be a strategic move to slow this decline and alleviate pressure on their stock valuations.

2. AI adoption is rising, but prices are falling. As observed with other general purpose technologies of the past, AI is following a competitive pattern that mirrors historical trends seen in industries such as air travel, cars and personal computing.

AI revenue and profits will come as the technology offers increasing, demonstrable value to users.

However, due to infrastructure bottlenecks, the timeline may be longer than what is implied in valuations. The AI bubble may merely be one of expectations and time, which would be typical.

Additionally, and perhaps more importantly, a key challenge that does not seem to be recognised by many market participants is that AI models operate with negative economies of scale.

Each complex client query triggers expensive compute processes, where initial operational costs can often exceed the revenue generated by that single interaction.

This is in stark contrast to the Internet 2.0 giants, which leveraged powerful network effects to achieve monopolistic positions and historic profit margins. AI services, being highly commoditised at the utility layer, may lack these inherent network effects and possess less convex profit profiles than the previous generation of technology giants, requiring more capital, capex and investor trust.

Opportunities amid constraints

Despite these challenges, we believe that the AI boom presents significant opportunities for investors. The bottlenecks in the physical economy may create underappreciated profit potential for suppliers who sell into these constrained areas.

We are excited about the potential opportunity for best-in-class operators in businesses such as electrical equipment, machinery and tools, speciality chemicals, semiconductor capital and network equipment, power management solutions, and so on.

Considering the rapid advancement and capital-intensive nature of AI, it is essential to examine how physical and financial constraints will present both risks and opportunities for investors.

We believe that the focus should be on companies that enable the physical infrastructure required for AI and also possess durable competitive advantages. And of course, on avoiding owning businesses where product differentiation is low and the risk of rapid obsolescence is high.

 The writer is portfolio manager and global investment strategist at MFS Investment Management