“The maximum number of NVIDIA Hopper GPUs that can be run as a coherent cluster for AI training is approximately 200,000.”