OpenAI and Broadcom unveiled a jointly designed AI inference chip called Jalapeño on Wednesday in San Francisco [1, 2, 3].
The partnership marks a significant shift in AI infrastructure as OpenAI seeks to build its own hardware to support large-scale language models. By developing a custom accelerator, the company intends to compete directly with the dominant AI chips produced by Nvidia [4, 5].
Broadcom CEO Hock Tan and OpenAI President Greg Brockman presented the new hardware during the unveiling [1, 2]. The Jalapeño chip is specifically designed for inference, which is the process of using a trained AI model to make predictions or generate text based on new data [3, 5].
Market reaction to the announcement was immediate. Broadcom stock rose more than 1% following the news [6]. This uptick comes after a volatile period for the company, as Broadcom shares had lost nearly 20% since early June [6].
The move toward custom silicon allows OpenAI to optimize the hardware for its specific software requirements. This approach can potentially lead to greater efficiency, and lower operational costs, compared to using general-purpose AI hardware [5].
While the companies did not disclose the full technical specifications of the Jalapeño chip, the collaboration highlights a growing trend of major AI developers partnering with semiconductor firms to secure their own supply chains [3, 5].
“OpenAI and Broadcom unveiled a jointly designed AI inference chip called Jalapeño”
The development of the Jalapeño chip signals OpenAI's transition from a software-centric company to one that controls its full technology stack. By reducing its dependency on Nvidia, OpenAI can better manage the massive compute requirements of its models while potentially lowering the cost of AI services. For Broadcom, this partnership provides a critical victory in the AI hardware race and a potential catalyst for stock recovery after a month of declines.
