“The models are giving superpowers to our engineers,” says Richard Ho, vice president of hardware at OpenAI.
OpenAI claims its chip beats Nvidia’s GB300 on latency. Every benchmark here comes from OpenAI, and the article admits real-world gains remain to be seen. The story is also a sales pitch. A company that sells models says its models designed its chip. Wake me when it runs the inference fleet for a year without Nvidia.