Report by The One Hundred Year Study on Artificial Intelligence (AI100) 2026 Study Panel : “The main techniques that underpin most of today’s AI technologies—neural networks—were developed long before the current boom. But it took the emergence of powerful graphics processing units (GPUs) and the era of big data to make neural networks possible at large scale. Neural networks that leveraged these compute and data resources demonstrated their value for tasks like image recognition more than a decade ago. Since then, much of the field of AI has focused on the pursuit of more capable AI systems by increasing computational scale. The use of ever more powerful computing chips and ever more digital data, together with the capital to put them together, allowed for the creation of progressively bigger and more capable neural network models, and ultimately enabled a qualitatively new class of AI systems. Large language models (LLMs), which burst onto the global stage in November 2022 with the release of ChatGPT by OpenAI, are the highest-profile example of what the scaling era enabled.
In what follows, we use the term foundation model1 to mean a very large-scale machine-learning model trained on large datasets and built to perform multiple tasks, and frontier model to mean a foundation model that is at the cutting edge of current capability. Frontier LLMs are the best-known foundation models, although foundation models are increasingly being developed in specialized domains such as medicine.2 Since 2022, we have seen the rapid advance of generative AI: systems that learn from existing data to create new content, including written text, realistic or stylized artistic images, original songs, and computer programs.
The centrality of scale to modern AI has been apparent for some time, but the past five years have seen scaling taken to an unprecedented degree. The companies building frontier models today invest billions of dollars on infrastructure to develop a new state-of-the-art model. They are willing to do this because past experience has shown that increasing the training budget will lead to a predictable rise in quality—and a competitive advantage in the market…(More)”.
