The Questions You Should Be Asking About the AI Bubble
Rather than trying to time when the AI infrastructure bubble bursts, executives should assess genuine economic exposure — history shows bubbles often finance transformative infrastructure even as investors lose money.
Summary
BCG's chief economist Philipp Carlsson-Szlezak and senior economist Paul Swartz argue executives are asking the wrong question about AI investment: not "when will it burst" but "does it threaten the real economy." They cite capex estimates ranging from hundreds of billions to several trillion dollars in AI data-center spending, and note that historical bubbles (rail, telecom, dot-com) often left behind valuable infrastructure and innovation despite investor losses. The piece counsels a more nuanced, less alarmist reading of bubble risk and redirects attention toward capturing value from the current investment wave rather than predicting its collapse.
Editor's opinion
A rigorous, well-quantified reframing of AI-bubble anxiety -- the GDP-sizing breakdown and the bank-exposure analysis explaining why this boom differs structurally from 2008 are both checkable and specific rather than asserted, and the “engage to thrive and survive” argument (being right about a bubble doesn’t protect you if competitors end up owning the resulting productive capacity) is a genuine strategic point, not just economic color. The main limitation is durability: the analysis is pinned to specific 2026 figures -- the $630B/$315B capex estimates, the household equity-wealth share, the state of hyperscaler balance sheets -- that will need updating within a year or two as the buildout evolves, so its value is more as a snapshot of how to think about the risk than as a lasting reference.
Key concepts
AI infrastructure bubble · historical bubble/innovation parallel