OpenAI Shows First Results for Jalapeño Inference Chip

At the Hot Chips conference on August 25, 2026, OpenAI published its first test results for Jalapeño, its inference ASIC developed with Broadcom: according to SemiAnalysis's InferenceX benchmark, the chip delivered 1.5–1.9 times more AI work per watt and 1.7–3.6 times lower end-to-end latency than the best previously recorded results on NVIDIA GB200/GB300.

Compiled by Sergey KostenchukPublished 2026-08-25Updated 2026-09-18

2026-08-25 Coding OpenAI

💻 OpenAI Shows First Results for Jalapeño Inference Chip

At Hot Chips on August 25, OpenAI announced figures for Jalapeño, its inference ASIC developed with Broadcom. According to SemiAnalysis's InferenceX, the chip delivers 1.5–1.9 times more AI work per watt and 1.7–3.6 times lower end-to-end latency than the best results on NVIDIA GB200/GB300.

🌍 The ASIC targets the most expensive part of the AI economy — inference costs. If the figures are confirmed, OpenAI's request costs will fall, and its negotiating position against NVIDIA will strengthen. Mass production — 2027.

👤 In the long run, this hardware will determine response latency in ChatGPT and the API: more responsive agents and voice modes. But the figures are currently manufacturer-claimed, and there are no mass shipments before 2027.

Source 1: https://openai.com/index/jalapeno-first-results/ Source 2: https://techcrunch.com/2026/08/25/openais-jalapeno-chip-is-built-for-fast-inference-at-scale-benchmarks-show/

Sources