Google has reportedly selected AMD EPYC processors over its own Axion CPUs for its first confidential Tensor Processing Unit (TPU). This decision highlights Google’s efforts to strengthen AI security while continuing to expand its custom chip infrastructure.
The move stands out because Google has invested heavily in developing its own Axion Arm processors for cloud computing. However, the company has chosen AMD’s established EPYC platform to support this security-focused TPU deployment.
Google TPU 8tx Combines Custom AI Chips With AMD EPYC Processors
Google’s new TPU 8tx system combines its eighth-generation Tensor Processing Units with fifth-generation AMD EPYC processors. The platform introduces hardware-based confidential computing to protect sensitive information during AI processing.
AMD announced on November 9 that its EPYC processors power the host infrastructure behind Google Cloud TPU 8tx. This partnership combines Google’s AI acceleration technology with AMD’s security features and broad x86 software compatibility.
Google TPU 8tx targets businesses requiring stronger protection for sensitive AI workloads. These include government agencies, regulated industries and enterprises managing proprietary AI models or confidential data.
The integration also highlights the growing role of confidential computing in cloud-based AI infrastructure. It allows organisations to run demanding AI workloads while maintaining stronger control over sensitive information. The partnership brings together two different processor technologies to address the security needs of enterprise AI applications.
AMD EPYC Protects AI Data During Processing
Google TPU 8tx uses AMD’s Secure Encrypted Virtualization (SEV) technology to extend confidential computing protections beyond traditional CPU security. This integration combines AMD EPYC security features with Google’s TPU architecture to protect information throughout AI processing.
AMD EPYC processors manage workload orchestration, data preparation and communication with Google’s TPU accelerators. The system also encrypts and isolates prompts, model weights and intermediate data during AI operations.
This security approach allows businesses to process sensitive AI workloads without exposing information to underlying cloud infrastructure. Google says TPU 8tx protects confidential data and model weights even while AI models actively process them.
The technology also helps prevent unauthorised access to sensitive information during cloud-based AI workloads. It creates a secure environment where data remains protected across different stages of model execution. This is particularly important for organisations handling confidential business records, financial information and proprietary AI models.
ALSO READ: Google Runs AI Model in Space for First Time, Sends Poem to Earth
Why Google Chose AMD EPYC Instead of Its Axion CPUs
Google has not publicly explained why it selected AMD EPYC processors instead of its own Axion CPUs for TPU 8tx. However, Google’s existing confidential computing infrastructure may help explain the processor choice.
Google’s confidential computing services already support AMD Secure Encrypted Virtualization and Intel Trust Domain Extensions. Its documented configurations currently do not include Axion-based instances for Confidential VMs.
This existing security infrastructure may have influenced Google’s processor selection for TPU 8tx. AMD’s established security technology and compatibility with Google’s confidential computing services could have made EPYC a practical option. However, Google has not officially confirmed this explanation.
ALSO READ: AI Infrastructure Spending Statistics
TPU 8tx Expands Google’s Confidential AI Infrastructure
Google’s TPU 8tx joins its eighth-generation TPU lineup, which supports AI training and inference workloads. The company introduced TPU 8tx and TPU 8i Max on November 9, expanding its AI computing offerings.
TPU 8tx also uses Google’s Boardfly networking architecture, which reportedly reduces latency up to twice as much as TPU 8t. The system focuses on confidential AI inference rather than general-purpose model training.
Google is also expanding TPU access beyond its own data centers through customer-operated deployments. Selected customers can install TPU 8tx systems within their facilities, supporting organizations with strict data security requirements.
The AMD partnership marks an important development in Google’s AI hardware strategy. Google continues developing custom processors while using AMD EPYC technology for security-focused TPU deployments.
The agreement also strengthens AMD’s role in AI infrastructure, where CPUs manage workloads and connect accelerators. However, AMD’s involvement does not mean it designed Google’s TPU or confirmed a broader accelerator partnership.
