Cracks in the AI Trade: What the Data Says About Bubble Risk

City analysts and financial economists have increasingly warned that the AI trade — concentrated in seven dominant firms including Nvidia, Microsoft, Alphabet, Amazon, Meta, Apple, and Tesla — is vulnerable to a reversal as companies discover that current AI systems cannot handle the complex, variable tasks they were expected to automate ³. That gap between promise and deployed capability is cited as a structural reason the boom could eventually deflate, even though the timing remains unknown ³.

The most rigorous quantitative case for concern comes from the Bank for International Settlements. Its research models the AI build-out as a contest among firms racing to secure a handful of dominant positions, and finds that this race generates over-investment exceeding the socially efficient level by roughly 50% under conservative assumptions, rising to as much as three times the efficient level where demand is less elastic ². Because AI hardware is highly specialized, the BIS analysis notes that fire-sale dynamics would amplify losses in proportion to leverage if demand disappoints, and a network analysis shows that the failure of a single major firm could cascade through chains of financial exposure to others, particularly given how concentrated the sector has become ².

The Bank of England has separately raised financial-stability concerns rooted in valuation math. AI-related stocks now make up a much larger share of major U.S. indices than they did even a few years ago — JPMorgan's AI stock index, spanning 30 heavily AI-exposed S&P 500 constituents, went from roughly 26% of the index in late 2022 to 44% by October 2025 ⁷. The Bank notes that these valuation multiples have pushed some U.S. equity metrics close to levels last seen at the peak of the dot-com bubble, and flags that the physical infrastructure needed to train and run AI models is expected to require trillions of dollars over the next five years, a significant portion financed by debt ⁷. Importantly, the Bank of England stops short of declaring a bubble outright, acknowledging that if AI delivers transformational productivity gains, current valuations could be justified — but stresses that this outcome is genuinely uncertain ⁷.

Academic voices have focused less on aggregate valuation and more on the structure of deals binding the largest AI players together. Yale School of Management's Jeffrey Sonnenfeld and Stephen Henriques argue that the tangle of cross-investments among tech giants — where one company grants equity to a chip supplier in exchange for data-center financing while simultaneously taking an ownership stake in a competing manufacturer — blurs the line between revenue and equity in ways that look less like disciplined capital allocation and more improvised deal-making ⁶. They explicitly draw the parallel to the run-up to the dot-com bubble, when investors moved with confidence despite deep uncertainty about the right path forward amid a genuine technological shift ⁶. This circularity concern echoes broader commentary suggesting that leading AI firms are engaged in a self-reinforcing flow of investments that some critics argue is artificially inflating stock valuations rather than reflecting organic demand ¹.

Not every institutional voice agrees that these dynamics point toward an imminent bust. Several major financial institutions have explicitly pushed back on bubble framing, arguing that current valuations track real earnings growth rather than speculative excess, with some analysts calling bubble fears "misplaced" or "premature" based on data about median company fundamentals ¹. Fidelity has made a related structural argument: unlike the dot-com era, when technology companies spent more cash than they generated for nearly a decade before the crash, today's large tech firms are largely funding AI capital expenditure out of earnings rather than borrowed money, and earnings restatements — a hallmark of the dot-com-era accounting scandals — have trended downward since 2021 rather than up ⁴. This is a meaningful point of disagreement: the same debt dynamics that the BIS and Bank of England treat as a vulnerability are, in Fidelity's framing, less pronounced than in prior bust cycles because self-funding has replaced heavy borrowing at the aggregate sector level ^{4, 2, 7}.

Practitioner-facing commentary occupies a middle ground, treating a bubble burst as a live risk to prepare for rather than a foregone conclusion. TechTarget's coverage frames the question as one of timing and preparation, advising industry leaders to look past the hype cycle and prioritize AI applications tied to demonstrable business problems, implicitly acknowledging that speculative excess exists even without predicting exactly when a correction would hit ⁸. Taken together, the evidence shows genuine and quantifiable disagreement: central bank modeling points to substantial overinvestment and network fragility, valuation metrics have approached dot-com-era extremes, and deal structures raise governance questions, while defenders of current pricing point to earnings-funded capex and improving accounting discipline as reasons the current boom differs from 2000 ^{2, 7, 6, 4, 1}.