As rumored for a year, OpenAI today finally launched its self-designed Jalapeño chip with the help of Broadcom. Jalapeño is specifically designed to perform LLM inference tasks and is OpenAI's first step towards developing its own infrastructure independent of NVIDIA.
Jalapeño is a chip designed for the specific tasks OpenAI needs and is more efficient than general-purpose GPUs. With its own designed chip, OpenAI can reduce costs and wait times by not having to compete for artificial intelligence (AI) chips that other AI companies are also using.
OpenAI says that the design process to production of the first chip only took 9 months. This process is faster than usual because OpenAI and Broadcom engineers are assisted by AI to perform the design and optimization process.
With Jalapeño, OpenAI is now in line with Google, which has developed its own Tensor chip, Microsoft with its Maia chip, and Meta with its MTIA chip.
