Bank for International Settlements publishes working paper finding AI investment may exceed efficient levels by around 50% and heighten bust risks

The Bank for International Settlements has published a working paper examining whether the current AI build-out could follow earlier technology boom-bust cycles. The paper models AI investment as a winner-take-most race and finds that contest dynamics can push firms to over-commit capital relative to the social return, with over-investment estimated at about 1.5 times the efficient level under a conservative baseline. It argues that debt financing and circular equity ties between hyperscalers and AI labs can turn that excess into financial fragility, making a downturn more likely and more disruptive if AI productivity falls short. The paper says the current AI build-out ranks among the largest technology-driven investment booms in US history, with major hyperscaler capital expenditure set to exceed USD 700 billion in 2026 and aggregate AI infrastructure spending widely expected to run into the trillions. In the model, larger booms require stronger productivity outcomes to remain sustainable, while specialized AI hardware amplifies fire-sale losses when borrowing is high. Over-investment can rise to around three times the efficient level when demand is less elastic. Calibrated results indicate that once the build-out reaches roughly USD 3 trillion, expected net economic surplus turns negative, while network analysis suggests that the failure of one firm could propagate through chains of financial exposures, especially where funding is concentrated through shared backers and circular financing links.