Confidential AI Solutions
NVIDIA Confidential Computing
Securing proprietary models and sensitive data
Overview
Protect Models and Data in Use
As organizations turn to agentic AI to enhance productivity and improve business outcomes, a critical concern emerges: safeguarding their sensitive data and models during inference and fine-tuning. Enterprise data is vulnerable to potential threats that could compromise its integrity and confidentiality. NVIDIA Confidential Computing, now in its third generation with the NVIDIA Rubin architecture, continues to advance AI—delivering strong data and model security with performance comparable to unencrypted models. With NVIDIA Confidential Computing, businesses can unlock the value of their sensitive data with AI to deliver revolutionary insights, with the assurance that data and models remain secure, compliant, and uncompromised.
Benefits
Confidential Computing Secures Every AI Workload
Explore how enterprises can secure the AI stack with NVIDIA Confidential Computing.
Safeguarding Models and Data in Use
NVIDIA Vera Rubin NVL72 unlocks rack-scale confidential computing across NVLink for high-speed, low-latency scaling to create a unified security domain spanning 72 NVIDIA Rubin GPUs, 36 NVIDIA Vera CPUs, and interconnects. It protects GPU execution, memory, and register states while keeping models, training data, and inference prompts isolated across the entire AI lifecycle.
High-Performance AI Security
NVIDIA Confidential Computing preserves the confidentiality and integrity of AI models deployed on Rubin, Blackwell, and Hopper GPUs. Vera Rubin NVL72 extends near-unencrypted performance to rack-scale, enabled by NVLink and NVLink-C2C. This design enables sensitive AI workloads to run securely at scale with near-native performance, even in shared or cloud environments.
No Code Changes
NVIDIA Confidential Computing on Hopper, Blackwell, and Rubin GPUs allows companies to quickly move any model into a protected enclave without code changes. Just identify the workloads that need to be protected and with a few keystrokes, those workloads are secure.
Verifiability With Device Attestation
A zero-trust architecture with hardware-rooted attestation to verify the trustworthiness of compute assets. Maintain compliance and ensure that apps and data are protected within the trusted execution environment (TEE) with the Vera Rubin Platform, Blackwell, and Hopper GPUs, regardless of where the platform or workload is running.
Products
Confidential Computing With NVIDIA
AI security and meeting the imperatives around data sovereignty is possible today, regardless of where your data resides and where infrastructure is deployed. NVIDIA has provided the trusted foundation to safeguard AI—whether running on a dedicated bare-metal host, kubernetes-native confidential containers, or confidential virtual machines.
Use Cases
Explore Customer and Partner Success Stories
Discover how leading organizations are revolutionizing security with NVIDIA Confidential Computing—cutting-edge technology that keeps sensitive workloads safe while data is in use.
-
Encrypted Compute
-
Data Protection
-
Model Sovereignty
Adopters
Advancing Every Industry With Confidential Computing
-
Independent Software Vendors
-
Model Providers
-
Cloud Partners
Resources
Advance Your Understanding
Learn more about NVIDIA Confidential Computing by exploring our blogs, training videos, Meet the Expert recordings, and GTC sessions.
Confidential Computing FAQs
NVIDIA Confidential Computing encrypts and protects data during execution. For AI workloads, this allows for encrypting and protecting proprietary AI models and sensitive data in use, including training data, fine-tuning datasets, inference prompts, and model weights, within a hardware-enforced trusted execution environment (TEE). This protection extends to on-premises model deployments, ensuring that proprietary models and confidential data remain secure even within shared or multi-tenant infrastructure, giving organizations full control over their AI workloads without compromising security.
NVIDIA Confidential Computing is supported across the NVIDIA Rubin, NVIDIA Blackwell, and NVIDIA Hopper™ platforms. For detailed information on supported GPU, CPU, and deployment configurations, please refer to the Supported Platforms page.
Attestation leverages hardware-rooted evidence to verify the identity and security posture of a confidential computing environment, ensuring its integrity before any sensitive code or data is released to it.
Supported configurations are designed to run at near-native performance. Actual results depend on the workload, platform, and configuration. Explore the confidential computing performance blog for more details on performance.
Many supported workloads can use confidential computing without application-code changes. Infrastructure and deployment settings still need to follow the supported configuration.
It can support data-protection and isolation controls, but it does not certify a workload by itself. Confirm your full architecture and operating process with your security, privacy, and compliance teams.
Next Steps
Ready to Get Started?
Secure your data today in a trusted execution environment powered by NVIDIA Confidential Computing.
Azure Cloud Services
If your data is stored in Microsoft Azure cloud, protect your data in use by accessing confidential computing.
Google Cloud Services
Cloud, protect your data in use by accessing confidential computing in the Google cloud.