OpenAI's custom Jalapeño chip has been benchmarked using SemiAnalysis' InferenceX test, showing superior token-per-user throughput and energy efficiency per kilowatt compared to existing state-of-the-art inference hardware.
Background
OpenAI has been developing custom silicon to optimize its AI inference workloads, reducing reliance on third-party GPU vendors like NVIDIA. This move is part of a broader industry trend of AI companies building tailored hardware for their models.
- Source
- TechCrunch
- Published
- Aug 25, 2026 at 10:22 PM
- Score
- 7.0 / 10