OpenAI's New Chip Makes AI Faster and Cheaper

OpenAI is making its own computer chip, called Jalapeño, with help from Broadcom. In early tests, the chip used electricity much more efficiently than Nvidia's latest systems — delivering up to 1.9 times more work for the same amount of power. That's important because AI is very energy-hungry. Lower power use can mean lower costs, and those savings might eventually reach you through cheaper subscriptions or more affordable AI-powered apps.

Jalapeño isn't just for OpenAI's own artificial intelligence models. The company tested it with three models: its own GPT-OSS 120B, plus DeepSeek R1 670B and Kimi K2.5 1T — both made by other companies. This shows the chip is flexible and can handle different kinds of AI work. It's designed to process your questions quickly and generate answers smoothly, even when many people are using it at the same time.

One smart feature is how it handles memory. AI models need to move data around constantly, which takes time and power. Jalapeño keeps important data close to the computing parts, reducing wasted effort. This helps the chip respond fast without sacrificing quality, something many systems struggle with.

OpenAI isn't dumping Nvidia. In fact, it still plans to use chips from Nvidia, Cerebras, and other partners for both training and running AI. Jalapeño will be an extra tool in the toolbox. For businesses, this could mean faster AI responses and lower bills. For everyday users, it might mean smoother chatbots, quicker answers, and eventually cheaper AI services. But keep in mind: the test results were published by OpenAI itself, not an independent lab, so the real-world performance is still to be confirmed.