Custom accelerator
An AI chip designed by a cloud or internet company for its own data centres, such as Trainium, Maia or MTIA.
Also: ASIC, in-house AI chip
A custom accelerator is an application-specific chip, or ASIC, built by a company for its own use rather than for sale. Google's TPU was the first at scale; Amazon followed with Inferentia in 2018 and Trainium in 2020, Microsoft with Maia and Meta with MTIA. Chip design firms such as Broadcom and Marvell often help with the work. The aim is to lower cost per token and reduce dependence on one supplier. Custom chips trade flexibility for efficiency, and they need their own software, which can slow their adoption by outside customers.