OpenAI Develops Inference Chip with Help from Broadcom (AVGO)
OpenAI has developed its first custom inference chip, Jalapeño, which delivers up to 1.9 times more AI work per watt than comparable systems and achieved tape-out in just nine months—faster than typical industry timelines. The project was supported by Broadcom Inc. (NASDAQ:AVGO), which handled the physical implementation of the design and provided networking components like Tomahawk switches, while TSMC manufactured the chip. Jalapeño outperformed Nvidia’s GB200 and GB300 systems in terms of work per watt and latency across several open models. However, it is an inference-specific chip and not a training accelerator; its performance was also tested under conditions that may favor single-token prediction over multi-token techniques used by Nvidia GPUs. OpenAI plans to deploy Jalapeño at scale starting late 2026. The development signals broader strategic implications for both Broadcom and Nvidia. While OpenAI still relies on Nvidia hardware, the success of Jalapeño could shift some workloads away from premium GPUs toward custom inference solutions. This has potential long-term impacts on market dynamics in AI chip design.
Tickers: $AVGO $CLS $NVDA $TSM
Topics: AI chip Broadcom partnership inference hardware OpenAI