BIS warns AI capex race could trigger dotcom-style bust and recession

The Bank for International Settlements, the so-called central bank for central banks, is raising a serious flag: the AI spending binge could unwind like past speculative manias, with recession-level consequences if the economics do not land.

In its annual report for 2026, the BIS compared today’s hyperscaler capex surge to historical episodes like canal and British railway mania in the 1800s, electrification exuberance in the 1920s, and the dotcom boom of the 1990s. The BIS’s core warning is blunt and specific: each episode shared a technological breakthrough that attracted capital beyond what commercial returns could justify, and that ended with a reversal in investment that induced economy-wide recessions. It also says the “scale and pace” of the current AI investment boom, alongside expectations of large productivity payoffs, “bear resemblance” to these precedents, highlighting near-term downside risk.

So what does that look like in numbers, right now? The BIS analysis points to forecasts already being laid out by major hyperscalers. Amazon forecasts capital expenditures of $200 billion for 2026, Microsoft is projecting $190 billion, Google is projecting some $180 billion, and Meta is up to $140 billion. The report also estimates that the five largest hyperscalers are set to spend more than a trillion dollars on AI-related capex in 2026. BIS frames this as plausible, but the reason it feels dangerous is not just the magnitude. It is the combination of speed, competition, and the reality that inputs like memory can be inflationary.

The BIS argues the capex race is outpacing earnings and free cash flow, pushing some firms to issue debt to raise financing. It also says the investment race may be driven by the perception that only a small number of players with superior technology will ultimately dominate market shares. That incentive structure matters. When competitive pressure drives spending higher, the net economic surplus for the tech industry declines. The BIS even raises the possibility that surplus “could turn negative in adverse scenarios.” In plain terms: if too many companies build too much, too fast, and the expected payoff does not materialize, the winners may not be able to carry the whole system.

That is where the “bubble pop” part comes in. The BIS warns that disappointed returns could trigger a sudden pullback in financing, turning the capex boom into a protracted investment bust. It then points to potential knock-on effects on financial conditions. One scenario it describes is macro-financial: if inflation spikes or AI-led investment collapses, policy rates being tightened to control inflation may precipitate a “sharp pullback in asset prices” after a prolonged period of exuberant risk-taking. The BIS calls out “disruptive macro-financial feedback loops,” which is economist-speak for a loop where losses feed into tighter financing, which feeds into more losses.

The report also highlights vulnerabilities that do not stay neatly inside Big Tech. It cites AI companies’ “rising leverage” and a “growing footprint in credit markets,” warning that a major change in optimistic sentiment toward these businesses could have serious knock-on effects in finance. BIS adds that vulnerabilities extend to the supplier ecosystem, including engineering, procurement, and construction contractors. Those firms, it says, have comparatively weaker balance sheets, leaving them exposed if hyperscalers cut capex. There is also an opacity issue: the BIS notes the “opacity” of AI-sector financing is compounding vulnerabilities as corporations create a web of private arrangements, including circular financing. It also says terms of datacenter facility leases are often not fully disclosed. If you are a board member or CFO, this is the part that should land hardest: in a downturn, it is harder to price risk when the wiring diagram is hidden.

On top of all of that, BIS points to a “supply side roadblock” around electricity availability, chip shortages, and grid connection bottlenecks. AI datacenters are already putting pressure on energy prices and input costs, and BIS points to potential spillovers to inflation. It also notes that shortages can amplify over-investment, because firms try to lock in future capacity through long-dated contracts that further expose them to disappointments in demand.

Finally, BIS adds a reality check on outcomes. It references the backdrop that, while enterprises running pilots report some efficiency gains at an employee level, few report discernible productivity gains from AI projects that went into production environments at scale. That gap between pilot wins and large-scale production results is exactly where capital allocation debates get brutal. If the expected productivity payoffs are slower, smaller, or simply not there, a race built on keeping up with rivals can flip into a supply-and-credit shock.

For executives across hyperscalers, semis-adjacent supply chains, and the finance ecosystem that funds growth, the strategic stake is clear: the BIS is not just warning that spending is high. It is warning about what happens when capex outpaces cash generation, inputs are constrained, financing becomes tightly linked to sentiment, and the system has hidden exposure through supplier credit and opaque private deals. If capital reverses quickly, the knock-on effects could move from corporate balance sheets to the macroeconomy.