Inference-Layer De-Nvidia Goes Structural — OpenAI+Broadcom Jalapeño ASIC (50% cheaper inference, deploy by year-end) + DeepSeek V4-Pro permanent 75% price cut ($0.87/M output tokens, backed by Huawei Ascend 950). World's two largest inference providers building non-Nvidia supply chains simultaneously — de-Nvidia shift from tactical to structural. Validates ENTRY 83 telco-ification + ENTRY 78 '80% domestic'. Training moat intact; inference pricing power eroding. App layer is the real winner.