OpenAI chip challenges Nvidia in inference tests
OpenAI has disclosed performance results for its first custom artificial intelligence chip, saying the processor beat Nvidia’s leading systems on key measures of inference speed and energy efficiency. The chip, called Jalapeño, was developed with Broadcom as part of OpenAI’s push to gain greater control over the computing infrastructure powering ChatGPT and its other artificial intelligence…
OpenAI has unveiled performance data for its inaugural custom AI chip, Jalapeño, which outperforms Nvidia's top systems in key inference speed and energy efficiency metrics. The chip, co-developed with Broadcom, aims to give OpenAI more control over the infrastructure for its AI services. Jalapeño delivered between 1.5 and 1.9 times more AI inference per watt than Nvidia's GB200 and GB300 technology and cut response latency by 1.7 to 3.6 times across various large language models.
The benchmarking, which covered models like GPT-OSS 120B, DeepSeek R1, and Kimi K2.5, focused solely on running trained models, not the training process itself. OpenAI plans to roll out Jalapeño in its infrastructure by late 2026, using it to lower costs and lessen the strain on energy and data-center resources, while continuing to rely on Nvidia hardware for model training.
Written by urgent.news from Arabian Post's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
