OpenAI Unveils Its First Processor: Jalapeño Created in Partnership with Broadcom

OpenAI, considered a leader in the world of AI, has officially announced its first proprietary chip. Developed in collaboration with Broadcom, the new processor is named Jalapeño. This step is a strategic shift aimed at reducing the company's heavy dependence on NVIDIA GPUs and increasing the efficiency of its own systems. This is reported by Techcrunch.com news reports.

The Jalapeño chip is specifically designed for the unique needs of OpenAI's systems. Interestingly, the company's own AI models were used during the processor's creation process. Although the chip is currently in the testing phase, initial results indicate that it is significantly superior to existing market alternatives in terms of energy consumption and performance ratio.

A specialized solution for the inference process

The new processor is primarily intended for "inference" processes, namely running ready-made models in response to user queries. OpenAI President Greg Brockman noted in a company podcast that they understand their workloads very well and have focused on accelerating areas that are currently underserved.

According to data, Jalapeño provides a very low cost when using real-time code generation models. While heavy tasks such as pre-training models still remain on NVIDIA equipment, even a slight reduction in inference costs will have a positive impact on the company's financial stability.

A step toward technological independence

With this move, OpenAI joins the ranks of giants like Google and Amazon. These companies have also developed their own specialized "AI accelerator" chips to speed up machine learning processes. For users, this news could mean faster and more stable operation of ChatGPT and other services in the future.

The company stated in its announcement that it is not limiting itself to creating models, but is optimizing the entire underlying infrastructure — from chip architecture to memory systems and network technologies. Such a complex approach serves to make models not only faster but also cheaper for users.

According to ixbt.com, OpenAI is currently actively developing agentic products like Codex and the data centers that power them. Having its own chips allows the company to perfectly integrate software and hardware toward a common goal.