What ended the most recent AI winter and started the current boom

A Gradual Shift, Not a Single Turning Point

Unlike the relatively sharp funding collapses that marked AI’s two historic winters, the transition out of the most recent winter and into the current sustained boom happened gradually, as several converging trends accumulated through the 2000s before becoming unmistakably visible in the early 2010s.

Accumulating Gains in Data, Compute, and Technique

Through the 2000s, several enabling factors steadily improved: computing power continued its long-running growth, the amount of digital data available for training AI systems expanded enormously alongside the growth of the internet, and researchers made steady refinements to neural network training techniques that had existed conceptually for decades but had previously been limited by insufficient data and computing power to reach their potential.

The 2012 ImageNet Result as the Clear Turning Point

While these trends had been building gradually, the 2012 deep learning breakthrough on the ImageNet visual recognition competition is widely credited as the moment that made the shift unmistakably visible to the broader research community and to funders, by dramatically demonstrating that these accumulating factors had crossed a threshold where neural network-based approaches could substantially outperform previous methods on a genuinely difficult, real-world task.

How Subsequent Milestones Reinforced and Accelerated the Boom

Following this initial breakthrough, a rapid succession of further milestones sustained and accelerated the boom’s momentum, including DeepMind’s AlphaGo victory over a top Go player in 2016 and the 2017 introduction of the transformer architecture, which became the technical foundation for the large language models that followed.

How Generative AI Products Extended the Boom Further

More recently, the arrival of highly capable generative AI products reaching widespread public and commercial use significantly amplified attention, investment, and adoption well beyond the research community itself, extending and intensifying a boom that had already been building for roughly a decade on the technical foundations established by earlier deep learning breakthroughs.

Bottom Line

The most recent AI winter ended gradually, as accumulating gains in computing power, available data, and neural network technique through the 2000s culminated in the highly visible 2012 deep learning breakthrough — a milestone widely credited with confirming to the field and to funders that a new, sustained period of progress had genuinely begun, a boom subsequently reinforced by further milestones and the rise of widely used generative AI products.

Look Up AI Terms

Search plain-English definitions of AI and machine learning terms in our free AI Glossary.