Twice the Cost-Efficiency of NVIDIA Blackwell: AMD MI355X Reshapes the AI Landscape with GLM5.2 Benchmarks
As of July 2026, AMD's Instinct MI355X has delivered a seismic shock to the AI semiconductor market. New benchmarks using the GLM5.2 model reveal a throughput of 2626 tokens per second per node, achieving over double the cost-efficiency of NVIDIA’s Blackwell architecture. This shift signals a transition from NVIDIA’s era of absolute dominance to a competitive, multi-polar AI infrastructure landscape.