Summary
AMD says its 5th Gen EPYC processors provide the host infrastructure for Google Cloud TPU 8tx. The companies describe the system as bringing hardware-enforced confidential computing to cloud TPUs.
AMD announced on November 9, 2026, that 5th Gen AMD EPYC processors are powering the host infrastructure for Google Cloud TPU 8tx. The companies describe the system, which pairs EPYC CPUs with Google’s 8th-generation TPU architecture, as the first cloud TPU system with hardware-enforced confidential computing.
The EPYC processors serve as host CPUs: they prepare data, coordinate workloads and manage communication with the TPU accelerators. AMD says the combined system is intended to support inference, fine-tuning and reinforcement-learning workloads, including applications that handle sensitive prompts, model weights or intermediate data.
How the confidential-computing design works
Confidential computing uses hardware-backed protections to help keep data protected while it is being processed. That matters for AI workloads because inputs and model data must be available to the system during computation, not just while stored or travelling over a network.
According to AMD, TPU 8tx uses Secure Encrypted Virtualization (SEV) capabilities in the EPYC processor alongside the TPU’s security features. The design extends confidential-computing protections from the CPU to connected accelerators. Before an accelerator receives direct access to protected memory, the system verifies its identity and security configuration. AMD says prompts, model weights and intermediate data remain encrypted in CPU memory and while moving between the host and accelerators, with hardware isolation during accelerator execution.
Google Cloud says the arrangement is designed to protect customer data and intellectual property even from the underlying infrastructure. The protection described by the companies is aimed at enabling sensitive AI work to run in the cloud; it is not a change to the TPU’s role as the accelerator.
The host processor’s role in AI systems
AI accelerators perform specialised computation, but they rely on host processors to supply data, schedule work and connect accelerator activity to the rest of a system. In this deployment, EPYC handles those host tasks while Google’s TPU provides AI acceleration. The pairing illustrates how CPU capabilities and accelerator security features can be combined in a cloud system rather than treating the accelerator as a standalone component.
AMD says EPYC processors also support host infrastructure across systems using GPUs and custom AI accelerator chips, and that its security technologies are used in other parts of Google Cloud’s confidential-computing portfolio. TPU 8tx extends that collaboration to confidential AI acceleration. The announcement focuses on the architecture and intended workloads; it does not report quantitative performance or security-evaluation results.