Regarding SemiAnalysis's tweet about "massive latency in Nvidia Kyber NVL144 rack architecture," many people may ignore the differences between the two chip models, Rubin and Rubin Ultra, and the Oberon and Kyber rack models? Points to pay attention to: 1. Rubin is a GPU architecture (chip level), paired with Vera CPU, and will begin mass production and shipment in the second half of 2026 (mass production models have already been showcased at the Taipei Computer Show in early June) using Oberon racks (NVL72 scale). 2. Rubin Ultra is an upgraded and enhanced version of Rubin (with more memory and higher performance configuration), planned to be launched in 2027, using Kyber racks. The Kyber rack is a new generation high-density rack specifically designed for Rubin Ultra, with the goal of increasing the number of single rack GPUs from 72 to 144, and achieving a larger "scale up domain" through more advanced NVLink (tightly connecting more GPUs like a large GPU). The delay mentioned by Semi is that the Kyber rack may experience delays due to the manufacturability of the PCB board. 3. In Nvidia's plan, the Rubber Ultra NVL144 (Kyber cabinet architecture) itself will not be launched until the second half of the year 27, and mass production may only begin in the second half of the year. After all, Rubin's NVL 72 will only be mass-produced in half a year this year. Actually, there's no need to pay too much attention to the progress of Kyber. Everything will be pushed according to Dazi's expected pace, and Kyber's mass production will not be implemented until the second half of next year. 4. Then SemiAnalysis said that the Kyber NVL144 rack architecture, which was originally planned to be mass-produced and implemented in the second half of 2017, will be delayed by 28 years. Everyone is saying that Nvidia will be delayed The focus is still on the pace of mass production and landing of Rubin NV72 (Oberon rack) this year.