
OpenAI CEO Sam Altman and Broadcom President and CEO Hock Tan. (Photo: OpenAI)
According to the two companies, Jalapeño is the first generation of chips in a long-term collaborative project and will continue to be improved in the coming years. The design and manufacturing process for this chip took nine months.
Jalapeño belongs to a group of specialized chips, developed from scratch to serve the request processing of large language models. Broadcom says the chip's design is based on in-depth exchanges with OpenAI researchers, while also taking into account the company's future model and product development plans.
OpenAI says initial tests show Jalapeño can achieve significantly higher processing performance per unit of power than current leading products. However, the evaluation process is not yet complete, and a detailed technical report is expected to be published in the coming months.
Currently, the two companies have not disclosed many specific product specifications. Jalapeño is expected to be better suited to the operational needs of AI models compared to the general-purpose chips currently used in many data centers.
Developing its own chips is also part of OpenAI's plan to give it more control over the infrastructure supporting models and products like ChatGPT and Codex. This could help the company reduce its dependence on external suppliers like Nvidia, while improving performance and energy efficiency.
Amidst rapidly growing global demand for AI computing, many technology companies have accelerated the development of specialized chips to better leverage the limited capacity of data centers.
Broadcom has also recently expanded its supply of custom chips to major technology corporations and advanced AI model developers. OpenAI and Broadcom plan to deploy the Jalapeño chip in their data centers before the end of 2026.
Source: https://vtv.vn/openai-va-broadcom-cong-bo-chip-ai-jalapeo-100260625151028717.htm







