In recent years, stocks associated with artificial intelligence have surged sharply, led by Nvidia. The company first reached a $1 trillion valuation in 2023, and by October 2025 had already crossed a $5 trillion market capitalization.
However, Terry Smith, one of the UK’s best-known investment managers, often referred to as the “British Warren Buffett”, has not joined the buying frenzy. On the contrary, he has reduced his exposure to several leading tech names tied to the sector. He cut his holdings by roughly half in Meta (Facebook and Instagram) and Microsoft, significantly reduced his position in Alphabet (Google), and avoided buying into Nvidia altogether.
The funds he manages hold assets worth tens of billions of pounds, and he is known for a long-term investment approach focused on high-quality, profitable companies with clear competitive advantages. So when he reduces exposure to some of the market’s strongest-performing tech companies, he raises a broader question: are today’s prices justified relative to the profits these firms are expected to generate in the future?
According to Smith, the answer is very likely no. In other words, he believes that AI-related stocks may turn out to be a bubble. Their massive market valuations could shrink if it becomes clear that they are unable to generate the profits and cash flows needed to justify them. “‘There may be a bubble in the stock market that is even more extreme than in 2000,’ Smith argues.”
In 2000, the dot-com bubble burst: technology stocks collapsed and, in 2001, the United States entered a recession. However, even then, the bursting of the bubble did not mean that the underlying technology was not useful or profitable. Rather, it reflected that market valuations had become unanchored from fundamentals, driven in part by easy financing and credit conditions, without companies generating profits that matched their soaring valuations.
Smith does not argue against artificial intelligence itself or its importance. For him, however, that is not the central question for investors. In his view, even world-changing technologies can become poor investments if the price paid for the companies already assumes overly optimistic future profits, high certainty, and clearly identifiable winners.
In letters to investors, interviews, and other public statements, Smith explains why he is concerned about AI-related investments and views them as a potential bubble. According to him, the market is treating AI investments as if their future profits are already guaranteed.
The costs involved are substantial: data centres, chips, servers, electricity, and large-scale infrastructure. At the same time, it remains unclear whether customers will be willing to pay enough for AI services, which companies will ultimately capture most of the value, and whether today’s apparent winners will still be dominant a decade from now.
The argument is that current market pricing is based on the assumption that success is effectively guaranteed, that investments will pay off, that profits will ultimately flow back to shareholders of today’s leading firms, and that rising valuations rest on solid economic fundamentals.
Smith points to alternative mechanisms that may be inflating these valuations. These include technology companies investing heavily in one another, transactions in which capital raised within the sector is quickly recycled back into purchases of cloud services and semiconductors, and index funds that automatically increase exposure to companies that have already risen in value—thereby amplifying the upward momentum.
The costs are enormous, but profits are uncertain
One of the central reasons Smith views AI as a potential bubble is the scale of profits and cash flows required to justify both the current market valuations of companies and the massive infrastructure investments being made.
In addition, he argues that even if large profits do materialise in the future, it is currently difficult to determine which companies will ultimately capture them. Smith attributed his reduction in holdings in Meta, Microsoft, and Alphabet to their heavy spending on artificial intelligence, which he sees as a risk. He explained this at the annual investor conference of Fundsmith, his main investment fund.
He calculated that if four major companies are collectively spending an additional $600 billion per year, this would require roughly $180 billion in additional annual operating profit to justify the investment. This figure is based on Smith’s benchmark assumption of a 30% return on capital.
Smith showed how capital expenditures by Meta, Microsoft, Alphabet, and Amazon grew from under $100 billion in 2020, to around $330 billion in 2025, and an estimated nearly $400 billion in 2026.
These expenditures finance, among other things, data centres, chips, servers, electricity infrastructure, and cooling systems. From Smith’s perspective, this represents a risky structural shift for investors. Until recently, these companies were attractive precisely because they generated substantial profits without requiring heavy investment in physical infrastructure.
Such massive investments demand equally large profits, but for Smith the risks do not end there. He fears an “arms race” dynamic in which companies invest primarily to avoid falling behind competitors, rather than because they see a clear, standalone path to returns on that investment.
He is also concerned that this is not a temporary spike in spending, but a structural requirement that will only intensify over time, forcing companies to commit ever-larger sums just to maintain their competitive position.
“‘It is unlikely that we know today who the winners will be’”
Another concern raised by Smith is that it is not possible to know which companies will ultimately emerge as the biggest winners from artificial intelligence in the coming years. Even if AI does prove to be a technological revolution, it is unclear where most of the profits in its value chain will accrue: whether to chip manufacturers such as Nvidia, to cloud and infrastructure providers like Microsoft, Amazon, and Google, or to model developers such as OpenAI, Google, and Anthropic.
Smith draws a parallel here with the dot-com bubble. At the time, it was already clear that the internet was a transformative technology, but it was not obvious in real time which companies would ultimately capture most of the profits from it. Some of the biggest winners of the internet era were not the obvious candidates at the height of the bubble, and some were not even major public companies yet. Amazon, for example, was then a relatively small, unprofitable company that later became one of the most dominant firms in the world.
For Smith, this serves as a warning against the assumption that today’s companies most closely associated with the AI revolution will necessarily be the long-term winners. “It is unlikely that we know today who the winners will be,” he said.
In addition, competition can emerge at every layer of the AI industry, making it even harder to identify clear winners and potentially eroding profits. As more companies develop AI models, it becomes increasingly difficult to determine which model, if any, will achieve lasting dominance in the future. Similarly, if more firms succeed in producing high-quality chips, Nvidia’s market power could weaken over time. For that reason, even strong and widespread demand for AI does not resolve the central investment question: which companies will actually succeed in converting that demand into sustained, high levels of profit.
The income is circular, and the index funds inflate the value of the shares
Smith is also concerned about the rapid rise in share prices following “every announcement” by companies, and the fact that the rally has extended to “any company perceived to benefit indirectly” from the technology. As a result, the relationship between companies’ valuations and their actual sales has surged, reflecting, in his view, a widening gap between market value and current business activity.
One mechanism that may be inflating share prices without being directly linked to long-term profitability is what he describes as circular revenue flows. This refers to demand that may be driven by an internal loop of investments and transactions between AI companies, cloud providers, and chip manufacturers, rather than by durable end-user demand.
Another factor he highlights is the role of index funds, which mechanically increase exposure to companies whose valuations are already rising, thereby reinforcing inflows into the same stocks and potentially amplifying the upward price momentum.
A further and central mechanism is passive investment funds that track stock indices such as the S&P 500. These funds offer index-tracking investment products and therefore purchase the shares of the companies included in those indices. As a result, when a company’s valuation rises and its weight in the index increases, passive funds are required to buy more of its stock. This additional demand can push the share price even higher, reinforcing the cycle.
Smith argues that the term “passive” is misleading, and that these flows are effectively momentum-based investing: the more a stock rises, the more it is bought. He sees index funds as a major source of inflated valuations and therefore believes this represents a “bubble more extreme than in 2000.” According to him, in 2023, assets in US equity funds held in index-tracking vehicles exceeded 50% for the first time. In contrast, less than 10% was held in such funds during the late-1990s dot-com bubble.
The problem is compounded by the fact that not only passive funds replicate the index, but also active fund managers, who, according to Smith, are increasingly afraid of underperforming benchmark indices. This is where Smith’s personal context also comes into play. After years in which his flagship fund delivered exceptionally strong annual returns, it has in recent years lagged behind major indices such as the S&P 500, whose gains have been driven largely by large technology companies.
One reason for this underperformance is Smith’s decision to avoid joining the AI-driven rally, meaning his funds have benefited less from the surge in tech stocks. His public statements on the issue also serve, in part, to explain the rationale behind this positioning to his investors. Nevertheless, he maintains his view: over the long term, some AI-related stocks may turn out to be poor investments.
Against Smith’s argument stands the view that the market may in fact be correctly pricing the economic potential of artificial intelligence. According to this perspective, companies such as Nvidia, Microsoft, Google, Amazon, and Meta could benefit for years from growing demand for chips, data centres, cloud services, and AI models.
From this angle, the enormous costs themselves may act as a barrier to entry, protecting the dominant players that have the capital, infrastructure, and customer base needed to compete in the race. However, even if this scenario proves true, it does not necessarily negate Smith’s warning. His argument is not that AI will fail, but that the market is pricing it as if success is certain—assuming that winners are already known, that profits will be large and predictable, and that the massive investments will generate returns quickly and reliably.