Report: OpenAI's In-House 'Jalapeño' Chip Could Outpace Nvidia's Blackwell
A detailed report from SemiAnalysis outlines OpenAI's push to design its own AI accelerator, codenamed Jalapeño, built in partnership with Broadcom. According to the analysis, early architectural details suggest the chip could match or exceed Nvidia's Blackwell GPUs in performance for the specific inference and training workloads OpenAI cares most about.
The move follows a broader industry trend of major AI labs and cloud providers designing custom silicon to reduce dependence on Nvidia, which currently dominates the GPU market with steep pricing and long lead times. Google, Amazon, and Meta have pursued similar in-house chip efforts for years.
While Jalapeño isn't shipping yet, the report suggests OpenAI is serious about vertically integrating its hardware stack as compute costs balloon alongside model scale.