Google Bets on the Agentic AI Era With Its AI Hypercomputer, Merging 8th-Gen TPUs, NVIDIA Rubin, and Axion CPUs

Google unveiled AI Hypercomputer at Cloud Next 26, a new datacenter architecture that unifies compute, storage, networking, software, and ML frameworks.

The platform adds 8th-gen TPUs for training and inference, including TPU 8t and TPU 8i, plus Axion CPUs and NVIDIA Rubin GPUs.

Google says the stack is designed to improve performance and scalability for frontier model training and serving.

For customers, the update could mean more efficient AI infrastructure and faster deployment of agentic AI applications.