Verifiable and Private AI: Google Cloud Expands Confidential Computing Territory

Google Cloud is expanding its Confidential Computing technology to guarantee privacy for AI inference and fine-tuning workloads. This allows AI models and prompts that handle sensitive data to be encrypted and protected within a hardware-based Trusted Execution Environment (TEE).

The newly announced Confidential G4 VM and Confidential GKE Nodes are equipped with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs to support high-performance AI computation. This service leverages AMD SEV technology on AMD EPYC Turin CPUs to encrypt even the data transfer process between the CPU and GPU, providing strong hardware-level security.

Key features are as follows:

• Global scalability: Available in all Google Cloud regions where the standard G4 series is offered
• Flexible models: Supports various consumption models including On Demand, Reservations, and Spot/Preemptible
• Performance optimization: Minimizes performance degradation even during high-performance tasks such as AI inference, fine-tuning, and HPC

With this update, enterprises can now focus on developing AI agents and models using sensitive business data without worrying about data leaks.