An ASIC, or application-specific integrated circuit, is a chip built for one particular task rather than for general use. Giving up flexibility lets it do that task faster and with less power. Google’s TPU is the best-known AI example: Google’s 2017 paper says the first TPU had been in its data centres since 2015.
ASICs take years and large teams to design, so they mostly make sense for the biggest buyers of computing power. You meet the term in stories about cloud companies and AI labs building or ordering their own chips, such as the Broadcom “XPU” accelerators behind a $63 billion financing deal announced in October 2026.