SemiAnalysis reports up to 50% better performance per dollar for TPU inference

SemiAnalysis says efforts to bring the TPU inference stack to outside customers are accelerating. Its coverage highlights InferenceX, Ironwood and TPUv8i, alongside reported gains of up to 50% in performance per dollar and a growing customer base. The publisher also describes CUDA’s competitive advantage as weakening.