$NVDA has announced that it is supplying its AI platform 'Vera Rubin' to customers including Google "In CoreWeave operations, megawatts per throughput is 10× higher than NVIDIA Blackwell." "Vera Rubin is a rack-scale AI infrastructure. It is co-designed with 32 NVIDIA Vera CPUs, 72 Rubin GPUs, NVLink 6, and storage. There are no internal cables, fans, or hoses, reducing assembly time from several hours to 1 minute." "Rubin CPU is NVIDIA's first self-designed server CPU. It applies 88 Olympus cores based on the company’s architecture." "Olympus cores deliver 1.8× higher CPU performance compared to x86 architecture-based CPUs. Inter-core bandwidth is 3× higher. Memory latency is 40% lower." "The number of tokens per megawatt is a key indicator for AI infrastructure. The more tokens, the more workloads can be processed with the same power budget and less power." $MU $SKHY $DRAM $TSM