Custom Silicon (ASIC)

Overview

Custom Silicon (ASIC) refers to Application-Specific Integrated Circuits designed to the specific requirements of a customer — typically a hyperscaler or cloud service provider — rather than using general-purpose merchant silicon. In the AI era, the primary use case is custom AI training and inference accelerators (TPUs, NPUs, and XPUs) optimized for specific model architectures, memory configurations, or workload profiles that general-purpose GPUs do not address as efficiently.

The custom AI accelerator market (excluding merchant GPU silicon) was approximately $6.1 billion in 2023 and is projected to exceed $40 billion by 2028 at a CAGR above 46%. Some third-party analyst estimates place the 2028 or 2030 total at $80–100 billion, depending on the pace of hyperscaler insourcing.