According to Jin10, Jalapeño leads Nvidia's GB300 in response speed and throughput per unit of power, but OpenAI has not marketed it as an "Nvidia killer" because it currently handles inference only, while training still relies heavily on GPUs.