Google unveils new TPUs, challenging Nvidia's AI dominance

Google has previewed its eighth-generation Tensor Processing Units (TPUs), designed to address the high costs and demands of AI model training and inference. The new TPUs, TPU 8t for training and TPU 8i for inference, aim to offer superior cost-efficiency by vertically integrating Google's AI stack. This move directly challenges Nvidia's market position, as many AI labs rely on Nvidia's expensive silicon. Google claims its custom-designed chips, built from the ground up, provide a significant cost advantage per token, a benefit it can pass on to enterprise clients using Google Cloud and Vertex AI.