Inference chip

OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom ASIC targeting 50% cheaper inference than Nvidia GPUs. Here's what the chip is, how it compares, and what it means for API costs and the broader silicon war.