OpenAI Announces Performance Advances with Custom Inference Chip Developed in Partnership with Broadcom
On Tuesday, OpenAI announced that testing results for its custom inference chip called Jalapeño show a significant performance advance. The chip was developed in partnership with Broadcom and is part of OpenAI's strategy to build the full stack behind its AI models. According to OpenAI, Jalapeño can process more workloads per unit of power while delivering faster responses compared to existing hardware systems that often require trade-offs between these metrics. OpenAI plans to begin deploying Jalapeño in its compute infrastructure by year-end, which could bolster Broadcom's FY2027 AI revenue outlook. However, the news does not alter concerns about political pressures on data center construction or a recent decision to reduce holdings of Broadcom stock. Despite this development, OpenAI remains committed to using Nvidia hardware for training and inference workloads in its Ohio data center project. Nvidia is expected to address competition from custom-designed chips during its upcoming earnings call. Investors will also be watching the broader market reactions as well as key economic indicators like the PCE price index and GDP figures.
Topics: AI Broadcom Nvidia OpenAI semiconductors