OpenAI's new chip makes AI faster with less power
OpenAI showcased its custom AI chip, Jalapeno, at the Hot Chips conference, revealing that it can accelerate AI responses while consuming less power. The processor delivered between 1.5 and 1.9 times more AI work per unit of power used and reduced response generation times by between 1.7 and 3.6 times across three tested AI models.
For highly interactive workloads, performance increased by 2.1 to 4.1 times. This announcement reflects OpenAI's ongoing effort to build its own computing infrastructure to support growing AI demand. Jalapeno is an inference chip, designed to run AI models and generate answers after training. OpenAI tailored the chip to modern AI models' unique requirements, aiming to improve speed and efficiency.
Tested on GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T models, Jalapeno demonstrated up to 3.4 times lower end-to-end latency and 1.5 times higher performance per watt for the largest model. Its power consumption is rated at 700 watts, with sustained power at or below 550 watts during tests. OpenAI credits its AI models for accelerating the chip's development from design to manufacturing in nine months.
Despite this breakthrough, OpenAI will continue using Nvidia accelerators for training and inference, using Jalapeno as an additional computing resource and gaining more control over its AI infrastructure.
Written by urgent.news from The Economic Times's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
