OpenAI Unveils Custom Inference Chip

San FranciscoOpenAI’s first custom chip runs inference without Nvidia.

Jalapeño, co-designed with Broadcom and built by Taiwan Semiconductor, matches Nvidia Blackwell performance, Broadcom’s CEO said; deployment is planned by year-end.

Meta, Google and Amazon have already moved to custom silicon; Anthropic is now weighing the same.

Sources: Reuters