Intel Corporation

Contribution to AI

Moore's law was an observation before it became a schedule. A 1965 trade-journal article drew a line through five data points and predicted that the number of components per integrated circuit would keep doubling; once its author was running a company organised to meet that curve, the doubling turned into an industry-wide planning assumption, and researchers could design for machines that did not yet exist on the confidence that they soon would. The 4004, shipped in November 1971 out of a calculator contract, put a whole processor on one die and made computation a part you could buy rather than a room you had to build. For four decades afterwards the practical substrate of the field was an x86 chip: expert systems, statistical natural language processing, kernel methods and the first decade of web-scale data all ran on it, while the specialised Lisp hardware of the 1980s lost to commodity parts. The failures matter as much. An analog trainable neural-network chip late in the 1980s found no buyers; the deep learning wave went to GPUs instead; the Nervana processors acquired in 2016 were cancelled before volume shipment, and the Gaudi accelerators that came with Habana Labs in 2019 have not displaced Nvidia. Loihi, the spiking research chip announced in 2017 and lent to academic groups rather than sold, is the quieter continuing contribution — and the whole sequence is the standard case study in how an incumbent's dominance fails to transfer across a change in workload.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).