According to CoinMeta, NVIDIA announced that its new generation of rack-level AI computing platform Rubin has begun mass production and delivery, with Microsoft's first batch of production devices having arrived. Microsoft CEO Satya Nadella shared images from the data center site, and NVIDIA confirmed that Rubin is now entering full-scale mass production. This platform integrates 72 Rubin GPU and 36 Vera CPU, and compared to GB200 NVL72, it is able to reduce the inference cost per million token by about one-tenth, while also reducing the number of GPU required to train MOE models by a quarter.