The NVIDIA Vera Rubin platform, designed for agentic AI, prioritizes data privacy and compliance through a comprehensive, hardware-accelerated security architecture. A core tenet is its third generation of Confidential Computing, which expands secure execution to a full-rack scale with the NVIDIA Vera Rubin NVL72. This creates a unified, trusted execution environment spanning all NVIDIA Vera CPUs, Rubin GPUs, and the NVLink™ fabric that connects them. The platform maintains data security across CPU, GPU, and NVLink domains, offering cryptographic proof of compliance via attestation services. This robust framework is specifically engineered to protect the world's largest proprietary models, sensitive training data, and inference workloads, ensuring that data remains secure even when in use.
To achieve this level of protection, the Vera Rubin platform integrates several key technical components and features. The BlueField-4 DPU (Data Processing Unit) is instrumental in offloading critical tasks such as networking, storage, encryption, virtual switching, telemetry, and security enforcement from the main compute path. It functions as a software-defined control plane for the AI factory, enforcing security, isolation, and operational determinism independently of host CPUs and GPUs. The platform also accelerates data-in-transit encryption, including IP Security (IPsec) and Platform Security Protocol (PSP for GPU-to-GPU communications), and provides data-at-rest encryption acceleration for securing storage platforms. Further enhancing security are features like secure boot, firmware authentication, and device attestation. The Vera Rubin NVL72 establishes a unified security domain across its components, protecting GPU execution, memory, and register states while keeping models, training data, and inference prompts isolated throughout the AI lifecycle.
These integrated security measures are critical for meeting stringent compliance requirements, especially for sensitive applications and sovereign AI initiatives. The platform's hardware-level data protection ensures that sensitive data and proprietary models can remain securely within specific jurisdictions, supporting air-gapped configurations for fully isolated environments. This capability is vital for industries such as healthcare, finance, and government, where data sovereignty and regulatory compliance are paramount. Furthermore, NVIDIA’s Agent Toolkit includes OpenShell, an open-source runtime designed to enforce security and privacy guardrails for autonomous AI, providing an additional layer of software-defined protection. By building security into every layer, from silicon to software, Vera Rubin aims to provide a resilient and compliant foundation for deploying and operating complex AI systems at scale.