Richard Ho on OpenAI Jalapeño Chip's Early Performance

Richard Ho, OpenAI’s VP of Hardware, shares early results from Jalapeño, OpenAI’s first custom inference chip. He explains how it is already proven to make ChatGPT responses faster and Codex sessions and agents more responsive. He also discusses how OpenAI models helped the team design and improve the chip, and why developing the models, software, and infrastructure together was so important. Finally, Richard looks ahead to deploying Jalapeño across OpenAI’s compute infrastructure and building future generations of the chip.