Following Nvidia's announcement at the end of August that its Rubin architecture chips were scheduled for mass production next year, the company further discussed Rubin's progress at the Goldman Sachs Technology Conference on September 8th. Nvidia CFO Colette Kress stated that Rubin chips are ready for market entry, with six Rubin chips already taped out. She stated that Nvidia's data center revenue for the second quarter of fiscal 2026 will be comprised of various components, including the Blackwell architecture's GB200 and B200 processors, and the Blackwell Ultra architecture's GB300 processors. The GB200 and GB300 processors are still shipping in the third quarter. Looking further ahead, Nvidia has already seen several gigawatts of demand for Rubin before it even enters the market. Kress stated that a year ago, the discussion was still focused on the transition between Blackwell and other architectures. Now that Blackwell, along with Blackwell Ultra, has entered the market, Nvidia has been on a relatively smooth path this year. Another frequently discussed topic is whether significant computation will still be required during pre-training, post-training, and inference. As we can see, there has been huge demand for computing over the past year, a very important part of which is generated by inference models.