Google turns to AMD for agent-focused chip security

Google Cloud unveiled the TPU 8tx and TPU 8i MAX, expanding its 8th-generation accelerator lineup for increasingly specialized AI workloads. The TPU 8tx targets agents that handle sensitive information, supporting small-scale training and inference with built-in protections for personal data. AMD’s 5th-generation EPYC processors, code-named Turin, will power its host infrastructure, coordinating data movement and processing.

The partnership brings hardware-based confidential computing to Cloud TPUs for the first time. Google Cloud said customers deploying agents need confidence that their data remains protected. AMD said the combination offers EPYC performance, AMD Infinity Guard security and x86 software compatibility. The choice contrasts with Google’s push to deploy its own Arm-based Axion CPUs alongside TPUs in its data centers, reducing reliance on outside chip suppliers.

The TPU 8i MAX links two TPU 8i chips to increase computing power, memory capacity and inference speed, delivering twice the computing performance and memory bandwidth of the TPU 8i. The agreement also strengthens AMD’s position as it challenges Intel in server processors. Mercury Research put Intel’s share of x86 server CPU revenue at 53.8% in the first quarter of this year, while AMD reached an all-time high of 46.2%.