AMD chips power Google Cloud's confidential TPU 8tx

AMD's 5th Gen EPYC processors are powering the host infrastructure for Google Cloud TPU 8tx, which is described as the first cloud TPU offering with hardware-enforced confidential computing.

AMD's server processors sit at the host layer of Google Cloud's TPU 8tx platform, handling workload orchestration, data preparation and communication with TPU accelerators. The setup is aimed at organisations running sensitive artificial intelligence workloads in the cloud, including inference, fine-tuning and reinforcement learning.

Confidential computing has become more prominent for cloud providers and chipmakers as businesses seek to deploy AI systems using proprietary models, prompts and internal data. In this case, AMD said its Secure Encrypted Virtualization technology in EPYC processors works with the TPU environment to protect data in CPU memory and as it moves between the host and accelerators.

According to the companies, the design verifies an accelerator's identity and security configuration before allowing it direct access to protected memory. This is intended to keep prompts, model weights and intermediate data encrypted in memory and in transit, while maintaining hardware isolation during accelerator execution.

AI host role

The announcement also highlights the role of central processors in AI systems that rely on dedicated accelerators. While much of the market focus has been on GPUs and custom AI chips, host CPUs still feed data, schedule workloads and link accelerator clusters to wider applications and services.

AMD said EPYC processors are now used as host CPUs across more than 10 hyperscalers and frontier AI labs, covering systems based on AMD Instinct GPUs, Nvidia GPUs and custom accelerator chips. That breadth reflects how cloud operators are building mixed AI infrastructure, often combining different accelerators with a standard server processor layer and a familiar x86 software base.

Google Cloud already uses AMD EPYC processors in several virtual machine families, including C4D, N4D, H4D, Z4D and G4. The latest deployment extends that relationship from general-purpose and compute-heavy cloud instances to confidential AI infrastructure tied to Google's TPU line.

Security has been a central theme of the collaboration. AMD said Google security teams have conducted deep reviews of AMD security processor firmware at a corporate level over several years. TPU 8tx adds a new element by extending confidential computing protections beyond the CPU to connected AI accelerators.

That matters because many AI workloads now involve sensitive commercial and operational information. Businesses using reasoning models or agentic AI systems may handle customer records, internal documents or proprietary code, making data protection during processing as important as protection at rest or in storage.

Dan McNamara, Senior Vice President and General Manager, Compute and Enterprise AI, AMD, said: "The most capable AI systems need a CPU foundation that can match the demands of the accelerators they host. Across generations of EPYC, we have built leadership capabilities in the performance, memory bandwidth and I/O that keep accelerated systems productive, backed by years of engineering collaboration with the world's leading cloud providers. Our work with Google Cloud brings that strength to TPU 8tx, helping customers scale AI with the performance and protection their most important workloads demand."

Google framed the move around privacy concerns as AI systems shift from development to operational use. It said customers want stronger assurances that model data and intellectual property remain shielded from the underlying cloud infrastructure.

Mark Lohmeyer, Vice President and General Manager for AI and Computing Infrastructure, Google Cloud, said: "When teams move AI agents into production, they need confidence that their data and intellectual property stay private, even from the underlying infrastructure. By pairing 5th Gen AMD EPYC processors with our 8th-generation TPU architecture inside the AI Hypercomputer, we are bringing hardware-enforced confidential computing to cloud TPUs for the first time. Customers get silicon-level data protection alongside the low-latency performance needed for modern reasoning models."

Beyond TPU 8tx, AMD said its SEV technologies are already used across parts of Google Cloud's confidential computing portfolio, including Confidential VMs, Confidential GKE, Confidential Space, Confidential Managed Service for Apache Spark and Confidential Dataflow.

The latest step gives AMD another reference point in a market where cloud providers are increasingly differentiating on AI infrastructure design as well as access to chips. It also shows how CPU vendors are seeking to remain central to AI deployments even as specialist accelerators dominate attention, with security and systems integration becoming a key part of that case.