Hyperscalers Designing Custom In-House AI Accelerators
Major cloud providers are increasingly designing their own custom AI accelerator chips rather than buying exclusively from merchant silicon vendors, using their own scale to justify the enormous non-recurring engineering cost of custom silicon development. This trend threatens to cut merchant vendors out of a growing share of the largest, highest-volume accelerator contracts anywhere, even as those same hyperscalers continue buying merchant chips for workloads that do not justify custom design investment. Merchant vendors are responding by offering more customizable reference designs and co-development partnerships to retain a foothold in accounts pursuing custom silicon.
Market Impact: Adds 12 billion inference chip units








