Is the AI Boom Heading for a Dotcom-Style Bust?

The Bank for International Settlements (BIS) has drawn historical parallels between the rise of AI and earlier waves of infrastructure and technology investment, each of which tended to trigger a boom that eventually pushed stock prices to unsustainable levels. The central bank of central banks, as the BIS is also known, points to canal mania in the 19th century, the railway boom and the electrification drive of the 1920s as precedents. The dotcom boom of the early internet era in the late 1990s remains the most recent major technology bubble to burst. Across these episodes, a common pattern emerges: a genuine technological breakthrough attracts capital inflows that ultimately outpace commercially viable returns.

The result has consistently been a sharp reversal of investment, with knock-on effects for the wider economy. Stock prices have at times fallen so abruptly that technology shares dragged down entire indices, including firms with little direct exposure to the technology in question. The BIS, headquartered in Basel, is the world's oldest international financial institution, founded in 1930. Its membership is limited to central banks, currently numbering 63, including the US Federal Reserve (Fed), the ECB and the German Bundesbank. Its mandate is to support central banks in maintaining monetary and financial stability. It also houses the secretariat of the Financial Stability Board and the Basel Committee on Banking Supervision, which coordinates the internationally influential Basel III framework.

https://twitter.com/ChristianWPupia/status/2074846871277678886

A Resilient Economy, Propped Up by AI Optimism

In its Annual Economic Report 2026, the BIS notes that the global economy has proven more resilient than expected so far, despite tariffs, geopolitical shocks and energy uncertainty. A key factor behind this resilience has been AI-driven optimism, which is fueling investment in semiconductors, data centers and power infrastructure. This has primarily boosted overall investment in the US and, through supply chains, has also lent economic momentum to Asia and Europe.

The report identifies a critical inflection point under the heading “AI progress and investment boom under pressure”. There, the BIS estimates that the five largest hyperscalers will spend more than $1tn on AI-related investments between 2025 and 2026. Hyperscalers, in this context, refers to major cloud and platform companies such as Amazon, Alphabet, Microsoft, Meta and Oracle. Their spending has now outpaced individual companies' profits and free cash flow, forcing some to take on additional debt.

The Hidden Costs Behind the AI Boom

It is worth noting that the BIS is not dismissing AI as worthless or a dead end. On the contrary, it explicitly acknowledges the technology's productivity potential. At the task level, the report finds efficiency gains of 20%–50% in time savings. Economy-wide productivity estimates, however, are considerably more conservative, typically remaining under 1% over longer periods, a gap the BIS attributes to the slower pace of adoption, transition and integration into production processes.

The problem, then, is not the technology itself but the financing and expectation structure built around it. Companies are committing enormous sums today on the assumption that only a handful of players will ultimately dominate the market. This dynamic risks becoming self-defeating: the more each firm races to build capacity rather than be left behind, the greater the risk of collective overinvestment. Should future AI returns fall short of expectations, a significant share of these investments could come under pressure simultaneously. Indeed, there are already signs that AI companies' costs are running considerably higher than currently disclosed.

You might be interested AI Was Going to Replace Us, But Then Came the Invoice

The Scale and Energy Problem

What distinguishes the current AI bubble from earlier ones is above all the scale and interconnectedness of the buildout. This is not merely a software phenomenon but a physical infrastructure one, requiring data centers of unprecedented size, along with the chips, servers, cooling systems, power connections, grids and generating capacity needed to run them. Meta, for instance, is building an AI data center the size of Manhattan at a cost of $50bn, complete with its own dedicated power plant. According to the International Energy Agency (IEA), data center electricity consumption rose 17% in 2025, roughly 5.7 times faster than global electricity demand overall.

This scale means AI can no longer be financed solely out of tech companies' own cash flows. The BIS points to a broader shift from cash flow to debt: in its bulletin Financing the AI Boom, it notes that investment needs have grown so large that companies are increasingly turning to external financing, with private credit markets playing a rapidly expanding role. Compounding this, hidden obligations are emerging that are not immediately visible on balance sheets, a phenomenon the BIS terms “shadow borrowing”. These include data centers partly financed through special purpose vehicles, leasing arrangements, long-term offtake agreements and private credit structures.

The Debt Hiding in Plain Sight

Economically, these obligations function much like debt, even though they do not always appear as such on corporate balance sheets. The result is a web of mutual dependence: hyperscalers, chip companies, AI labs, data center builders, power providers, private credit funds, insurers and banks are forming an increasingly dense network of symbiotic relationships around AI. The BIS points to “circular financing” as a case in point, whereby chip companies or hyperscalers take equity stakes in AI labs or neocloud providers while simultaneously locking in long-term purchase commitments for chips or computing capacity. The effect is a closed loop of money that leaves the system vulnerable to disruption: should financing or a delivery fail to materialize at any single point, the structure risks unraveling.

Major analysts, including Morgan Stanley and McKinsey, reach broadly similar conclusions. Morgan Stanley frames AI as an infrastructure cycle rather than pure speculation, estimating the global cost of new data centers, including chips, servers and infrastructure, at around $2.9tn by 2028. While the bank stops short of describing this as tech speculation, it does underscore the financing risk: the sheer scale of the buildout is exorbitant enough to pose a danger by size alone. McKinsey's projections go further still, putting global data center spending at around $7tn by 2030. Ultimately, though, the success of this buildout hinges on three factors: capital, energy availability and the practical usability of AI applications.

You might be interested British Foreign Secretary Warns of Hiroshima Moment for AI

The Three Stages of an AI Bubble

The BIS lays out a three-stage dynamic in its report. In the first stage, the AI boom supports growth, stock markets and investment in the short term: the construction of data centers, chip sales, power grid expansion and building projects all generate demand, which helps explain why the global economy has looked more resilient than expected despite other headwinds. In the second stage, high expectations for future AI returns give rise to an investment race, a dynamic already visible today as every major provider seeks enough computing power to avoid being shut out of the next platform market. This behavior is rational at the level of the individual firm, but collectively it risks producing overcapacity.

The third stage is speculation, and history suggests it is the most likely outcome of the two preceding it. Should the trend reverse, and should AI revenues, productivity gains or customers' willingness to pay fail to grow fast enough, the consequences would cascade: investments would be scaled back, stocks re-rated, credit made more expensive and suppliers put under pressure. What began as a growth impulse would then turn into an investment collapse.

The Dotcom Parallel

Much as with the internet, which only began generating meaningful returns after the dotcom bubble had burst, the risk lies not in AI itself but in how expectations around it are being financed, specifically the assumption that AI will quickly deliver very high, monopoly-like returns. The BIS's analysis captures this duality well: the AI boom represents progress and risk simultaneously. The progress is real, visible in productivity gains, forms of automation that did not previously exist and the buildout of new infrastructure. The risk lies in a trillion-dollar bet increasingly financed through debt, leasing, private credit and opaque chains of contracts.

The parallel with the dotcom era therefore holds up. Then, too, the underlying technology was genuine, but capital markets proved too optimistic in pricing in speed, returns and the eventual structure of market winners.