OpenAI Says Its Jalapeño Chip Can Power Faster AI Responses Than the Competition
At the Hot Chips conference on 25 August 2026, OpenAI published the first benchmark results for Jalapeño, the custom inference chip it built with Broadcom. On SemiAnalysis’s InferenceX benchmark, OpenAI reported 1.5x to 1.9x more work per watt and 1.7x to 3.6x lower end-to-end latency than an Nvidia Blackwell system, rising to 4.1x faster on highly interactive workloads. Here are the numbers, the architecture behind them, what OpenAI still has not disclosed, and what it means for anyone buying AI capacity.