AI Failures, Hype Cycles and Bubbles

Artificial intelligence has failed as a business more often, and more publicly, than almost any other technology. It has done so in a small number of ways, again and again. A striking demonstration persuades people that a general capability exists. Money and expectations run ahead of what the system can reliably do in messy real settings. The gap between demo and deployment is filled with human labour, optimistic accounting or simple waiting. Then the money runs out before the capability arrives. This happened to machine translation in the 1960s, to expert systems and LISP machines in the 1980s, and to IBM's Watson Health in the 2010s. It happened to the self-driving programmes of 2016–2024, from Uber ATG and Argo AI to GM's Cruise, and to Zillow's algorithmic home-buying in 2021. The generative-AI boom added its own cases: Humane's AI Pin, the Rabbit R1, the reset at Jasper and the collapse of Builder.ai in 2025. There were also start-ups whose "AI" turned out to be people, and the first enforcement actions against "AI washing". Each failure also had survivors who learned from it. Waymo outlasted Cruise, Opendoor outlasted Zillow Offers, and the deep-learning researchers who kept working through the second AI winter built the industry of the 2020s. This section explains how AI hype cycles work and walks through the major failures from the first AI winter to 2026. It sets out the recurring mechanics of demo-driven fundraising and "humans behind the curtain", places the 2023–26 boom alongside earlier technology bubbles, and explains how an experienced investor, operator or journalist reads the early signs.