AI Ambition Must Not Mortgage Institutional Resilience

Artificial intelligence often arrives in an institution as a small monthly subscription. The visible price can appear modest: a licence for staff, a chatbot inside a learning platform, automated transcription, a content assistant or an agent that promises to remove administrative work. The surrounding economic system is anything but small.

Models depend on data centres, specialised chips, electricity, water, networks, capital markets and revenue expectations extending years into the future. Institutions do not need to predict whether the AI boom will succeed or fail. They do need to recognise that price, availability, product design and vendor behaviour are shaped by a buildout whose assumptions remain uncertain.

Responsible adoption therefore needs more than a use case. It needs a resilience case: purpose, baseline, evidence, full cost, bounded commitment, retained human capability, continuity, portability and a funded exit. Innovation is not responsible when the institution can begin but cannot stop.

Reuters reported on 24 September 2026 that a new study prepared for a Brookings Institution conference estimates that United States AI infrastructure expansion could absorb about 3.6% of annual gross domestic product through 2032, exceeding $10 trillion. The study estimates 183 gigawatts of new data-centre capacity over seven years and argues that investment is increasingly moving beyond the cash flows of large technology companies into more complex external financing. Reuters report

The researcher, Columbia Business School professor…