Broadcom: The Performance And Economic Benefits Of OpenAI's JalapeñO Inference Chip

OpenAI partnered with Broadcom to develop its first custom inference chip: Jalapeño. It's a much more cost and power efficient way to monetize AI as compared to using NVIDIA GPUs. Jalapeño was designed using an AI-optimized architecture. It achieves lower latency, higher throughput, industry-leading performance, and uses less power than existing solutions, at less than half the cost of GPUs. Phase I of the AI-boom focused on GPU compute capacity for LLM training. Phase II will be focused on inference - a far larger market. Broadcom's XPU business will thrive.