OpenAI unveils Jalapeño inference chip, challenging Nvidia's dominance

OpenAI's custom Jalapeño chip achieves 1.5-1.9x perf/watt and 3.6x lower latency vs Nvidia GB300 on real models like DeepSeek R1. Designed in 9 months with AI assistance, 700W TDP. Deployment by year-end, with Gen 2/3 in pipeline. This directly challenges Nvidia's inference monopoly and impacts infrastructure financing dynamics.