Large-scale GPU supercomputing cluster architecture connecting hundreds of thousands of NVIDIA GPUs for AI training workloads, launched in 2025.