- OpenAI named the chip 'Jalapeño' and said it was designed with Broadcom for inference workloads.
- TSMC manufactured chips; Celestica will build server systems for OpenAI deployments.
- OpenAI reported samples running at target power and performance with its GPT-5.3-Codex-Spark model.
- OpenAI plans to deploy Jalapeño by the end of the year.
- The chip is intended for inference (serving queries) rather than training.
- OpenAI described this as the first step in a multi-generation chip program.
- Broadcom CEO Hock Tan said margins on custom AI chips are pressured by high HBM memory demand.
- Article names SK Hynix and Samsung as memory suppliers cited by Broadcom.
- The deal underscores Broadcom's role as a custom-accelerator supplier to major AI labs.
Useful?