OpenAI is incorporating Google's Tensor Processing Units (TPUs) for AI inference, diversifying its chip supply beyond Nvidia and aiming for cost efficiency.