Analyst Guo Mingyi said that its latest industry survey shows that Nvidia has restarted the AI inference pre-filling acceleration GPU “Rubin CPX” project, but its design has been drastically adjusted, and production is expected to begin in Q1 2027. At the chip level, the new Rubin CPX has computing power close to that of a standard Rubin GPU, and the pre-filling performance has been further improved. The maximum power consumption of the single unit is 2,300 W, and it is equipped with 168 GB of HBM4 memory.

Analyst Guo Mingyi said that its latest industry survey shows that Nvidia has restarted the AI inference pre-filling acceleration GPU “Rubin CPX” project, but its design has been drastically adjusted, and production is expected to begin in Q1 2027. At the chip level, the new Rubin CPX has computing power close to that of a standard Rubin GPU, and the pre-filling performance has been further improved. The maximum power consumption of the single unit is 2,300 W, and it is equipped with 168 GB of HBM4 memory.