Early testing shows the first-generation chip delivers significantly better performance per watt than current state-of-the-art solutions. Built from the ground up for current and future large language models, the chip expands OpenAI's full-stack platform and will be deployed at gigawatt scale with data center partners across multiple generations.