“NVIDIA's GB300 and VL72 systems achieve up to 50 times better performance per watt and a 35 times lower cost per token compared to its Opera platform, according to results from Inference X.”