August 26, 2026 OpenAI's Jalapeño ASIC: 1.9x Nvidia GB300 Throughput at Half the Power AI ChipOpenAINVIDIAInference Optimization