An “AI winter” is a period in which excitement about artificial intelligence collapses into disappointment, leading to deep cuts in research funding, commercial investment, and public interest. The term, coined by analogy with “nuclear winter,” describes the chilling effect that follows when AI fails to deliver on oversized promises. Each winter typically follows a “summer” of hype, when bold predictions outrun what the technology can actually do.
Historians generally point to two major AI winters. The first came in the 1970s. After early optimism about machine translation, perceptrons, and general problem solving, progress stalled on hard problems, and skeptical assessments such as the UK’s 1973 Lighthill Report, “Artificial Intelligence: A General Survey,” led governments to withdraw broad AI funding. The second came in the late 1980s, when the commercial boom in expert systems and specialized Lisp machines collapsed because the hardware was undercut by ordinary computers and the software proved costly and brittle. Contemporaneous documents from the field, such as the 1988 AAAI Presidential Address by Raj Reddy, capture the profession during that period of strain.
The deeper pattern is a hype cycle: a genuine advance attracts attention and money, expectations balloon beyond what is realistic, results disappoint, and a backlash freezes investment until the next breakthrough revives interest. Both winters were eventually followed by thaws, the 1980s expert-systems boom after the first, and the deep-learning revolution of the 2010s after the second.
Why business readers should care: the AI winters are the strongest historical reason for measured expectations. AI can deliver real value, but enthusiasm has repeatedly run far ahead of capability, and the resulting busts were painful for companies and investors caught in them. Distinguishing demonstrated results from marketing claims is the practical lesson.