AIdataNewsGlobal General: How NVIDIA OpenShell Reference Architecture Frames Secure Enterprise Agent Workloads
This AIdataNewsGlobal General overview reads the announcement made today through three lenses: infrastructure, runtime control and adoption. The release name is the NVIDIA OpenShell Reference Architecture, a full-stack reference design whose stated goals are to build, contain, observe and enforce safe agents on enterprise infrastructure and edge systems. In simple terms, the project moves NVIDIA OpenShell from a secure runtime software layer into a broader industrial-scale pattern for agent development and operations.
Stack view: where compute, runtime and enforcement sit
The design pairs NVIDIA Vera Rubin infrastructure for running enterprise workloads, including closed and open models built with NVIDIA Nemotron, with NVIDIA Vera CPUs for operating agent harnesses. It also includes open source NVIDIA OpenShell software and NVIDIA BlueField DPUs for running the NVIDIA Sentry unified security software framework, which provides continuous out-of-band monitoring of agent behavior and in-silicon policy enforcement.
For compute, NVIDIA Vera Rubin is centered on NVIDIA Vera Rubin NVL72 infrastructure and is described as usable across pretraining, post-training and inference. For agentic workloads, the cited system-level figure is 30x higher throughput per megawatt from NVIDIA Vera Rubin NVL72. For agent-host compute, NVIDIA Vera CPUs are said to deliver 2x the performance of x86 processors while running NVIDIA OpenShell to create a secure runtime boundary for autonomous agents.
On enforcement, NVIDIA BlueField-4 DPUs use NVIDIA Sentry, built on NVIDIA DOCA software, to watch agent behavior continuously. The pairing of BlueField and NVIDIA Sentry is presented as in-silicon security enforcement: if an agent attempts to move outside its software boundary, NVIDIA Sentry can quarantine and stop it in milliseconds. Running on BlueField-4 and powered by DOCA, NVIDIA Sentry is also said to boost agent capacity by up to 20%.
| Layer | Anchored components | Function in the reference design |
|---|---|---|
| Model and workload compute | NVIDIA Vera Rubin, NVIDIA Vera Rubin NVL72, NVIDIA Nemotron | Runs enterprise workloads across pretraining, post-training and inference; agentic throughput is cited as 30x per megawatt. |
| Agent harness compute | NVIDIA Vera CPUs | Runs agent harnesses and NVIDIA OpenShell; cited as 2x the performance of x86 processors. |
| Secure runtime | NVIDIA OpenShell, open source NVIDIA OpenShell software | Provides a secure runtime boundary for autonomous agents. |
| Out-of-band enforcement | NVIDIA BlueField DPUs, NVIDIA BlueField-4 DPUs, NVIDIA Sentry, NVIDIA DOCA | Monitors agent behavior continuously and enforces policy in silicon; BlueField-4 with DOCA supports up to 20% higher agent capacity. |
| Edge reach | NVIDIA RTX Spark, NVIDIA DGX Spark, NVIDIA DGX Station | Extends the OpenShell Reference Architecture to agents running at the edge. |
Ecosystem checklist: who is named in the deployment map
NVIDIA is working with industry leaders that build infrastructure, models and software so industries can safely deploy agents in production. For platform compatibility, the reference design supports leading enterprise operating systems and container orchestration layers. Those include Linux and Kubernetes infrastructure powered by Canonical, SUSE and Red Hat, plus Microsoft Windows using Microsoft Execution Containers, abbreviated MXC. Cloud operating-system support is named for Amazon Web Services, Google Cloud and Microsoft Azure.
- Model providers: Enterprises can run agents from model builders including Anthropic, Cohere, Google, Meta, OpenAI, Perplexity, Reflection and SpaceX Super Intelligence. NVIDIA is also working with model builder partners on OpenShell deployments for advanced model research.
- Cybersecurity integration: Support for the OpenShell Reference Architecture is being integrated into agent security solutions by cybersecurity leaders 1Password, Check Point, CrowdStrike, Cisco, F5, Fortinet, Microsoft, Okta, Palo Alto Networks, TrendAI and Zscaler.
- Bank governance interest: BBVA and Santander in Europe are working with NVIDIA to explore the OpenShell Reference Architecture as a framework for strengthening agent safety, security and governance.
- Enterprise software platform builders: Cadence, Dassault Systèmes, IBM, SAP, Siemens and Synopsys, among enterprise software leaders, are working with OpenShell to build agent software platforms. Palantir and NVIDIA are working with customers to deploy secure agents in production.
- System manufacturers: Dell Technologies, Cisco, HPE, Lenovo and Supermicro are among NVIDIA system manufacturers building infrastructure with the OpenShell Reference Architecture.
- Cloud service providers: CoreWeave, Crusoe, Firebird, Firmus, GMI Cloud, IREN, Lambda, Nebius, Nscale, Oracle Cloud Infrastructure and Vultr are working with NVIDIA to deploy OpenShell-ready infrastructure.
- Inference and post-training service providers: Baseten, Deepinfra, Fireworks and Together AI are building OpenShell Reference Architecture integrations into their model development pipelines.
Named product patterns: SAP, ServiceNow and edge systems
SAP connects business governance with secure agent execution by integrating the NVIDIA OpenShell Reference Architecture with Joule Studio runtime, part of SAP Business AI Platform.
ServiceNow today announced the general availability of ServiceNow Arc, a personal long-running AI agent whose actions are governed at runtime by the ServiceNow Super Intelligence Control Tower and enforced by NVIDIA OpenShell. The recently announced AI Passport authority layer from ServiceNow carries those policies to NVIDIA OpenShell and NVIDIA Sentry, so enterprises can revoke an AI agent’s access across software and hardware within seconds.
The OpenShell Reference Architecture can extend to agents running at the edge across industries such as retail, healthcare, manufacturing, robotics, software development and telecommunications with NVIDIA RTX Spark, NVIDIA DGX Spark and NVIDIA DGX Station systems.
Availability
The NVIDIA OpenShell Reference Architecture is now available.
FAQ for AIdataNewsGlobal readers
What is the main idea in one sentence?
The NVIDIA OpenShell Reference Architecture is a full-stack design intended to build, contain, observe and enforce safe agents on enterprise infrastructure and edge systems.
Which components handle runtime security?
NVIDIA OpenShell provides the secure runtime boundary, while NVIDIA Sentry running on NVIDIA BlueField DPUs and NVIDIA BlueField-4 DPUs performs continuous out-of-band monitoring and in-silicon policy enforcement.
Which performance figures are explicitly cited?
- 30x higher throughput per megawatt for NVIDIA Vera Rubin NVL72 in agentic workloads.
- 2x the performance of x86 processors for NVIDIA Vera CPUs.
- up to 20% higher agent capacity when NVIDIA Sentry is powered by DOCA and running on BlueField-4.
Does it cover only large data centers?
The design covers enterprise infrastructure and also extends to edge agents with NVIDIA RTX Spark, NVIDIA DGX Spark and NVIDIA DGX Station.
What is the adoption status?
The architecture is now available, and the partner list spans operating systems, clouds, model builders, cybersecurity vendors, banks, enterprise software, system manufacturers and inference providers.