Industry Dynamics - Broadcom is enabling OpenAI's custom inference chip named Jalapeño [1] - The new custom inference chip is positioned to change the inference hardware competition by evolving beyond raw performance [1] Market Trends & Performance - Jalapeño achieves approximately 4.2x performance improvements on the GPT-OSS-120B model [1] - The chip sits on the throughput-per-kW frontier across a wide range of decoding speeds [1]