tl;dr: OpenAI announced "Jalapeño," its first custom inference chip, developed with Broadcom and reportedly designed with help from OpenAI's own AI models. The chip targets inference workloads (not training) and claims significantly better performance-per-watt than current alternatives, following a similar custom-silicon path taken by Google and Amazon to reduce Nvidia dependence. Training-heavy workloads will likely remain on Nvidia GPUs, but cutting inference costs could meaningfully improve OpenAI's economics.