HPE expands Nvidia AI Factory with Vera CPU & security

HPE and Nvidia have expanded the HPE AI Factory with Nvidia, adding new processor, software and security elements to the portfolio.

The changes include support for the Nvidia Vera CPU in HPE Private Cloud AI, the addition of Nvidia Agent Toolkit, and the extension of Nvidia Confidential Computing across the HPE AI Factory range.

The move comes as suppliers push to turn agentic AI from trial projects into production systems in corporate and public sector environments. HPE and Nvidia are positioning the AI Factory range as infrastructure for those workloads across private, sovereign and large-scale deployments.

One of the main hardware changes is the planned availability in 2027 of the HPE ProLiant Compute DL394 Gen12 with Nvidia Vera CPU for HPE Private Cloud AI. Nvidia describes Vera as its first CPU built for AI agents, designed to handle tool calls, orchestration and real-time data processing within agent-based workflows.

The Vera CPU also forms part of the Nvidia Vera Rubin platform. HPE plans to offer the rack-scale Nvidia Vera Rubin NVL72 system, while the HPE Compute XD700, based on Nvidia HGX Rubin NVL8, is being added to the HPE AI Factory portfolio with support for up to 128 Rubin GPUs per rack.

An early enterprise customer is the New York Stock Exchange, which is working with Redpanda and HPE to explore the Vera CPU with the HPE ProLiant Compute DL394 Gen12 server. That gives the announcement a reference point beyond product roadmap detail, although no commercial terms were disclosed.

Agent tools

HPE Private Cloud AI will also include Nvidia Agent Toolkit. The software package includes Nvidia Nemotron open models, the Nvidia OpenShell runtime and Nvidia NemoClaw blueprints.

The toolkit is intended to help customers monitor agent behaviour, apply governance policies, and build and run autonomous multi-agent systems. HPE Private Cloud AI is also adding local agent registration so customers can approve AI models, skills and tools against central governance and security rules before deployment.

HPE is pairing those additions with data protection and storage changes. New HPE Zerto software features are designed to detect rogue agent actions and let systems be rewound to a clean state through continuous data protection.

On the storage side, HPE Alletra Storage MP X10000 has reached foundation-level status under Nvidia-Certified Storage. The system applies metadata and governance policies to unstructured data used in AI pipelines.

Security push

The expansion also places greater emphasis on confidential computing, an area that has drawn attention as businesses use proprietary data and models in AI systems. Nvidia Confidential Computing is now available across HPE AI Factory at Scale, HPE Sovereign AI Factory and HPE Private Cloud AI through HPE Services.

Confidential computing protects models and private data during execution in on-premises and sovereign deployments through cryptographic attestation and encryption. The HPE ProLiant Compute DL380a has also been certified under the Nvidia-Certified Systems program for Nvidia Confidential Computing.

Across the product line, Nvidia BlueField DPUs and Nvidia DOCA software are being used for policy enforcement, threat detection and network encryption. These features reflect rising customer concern over how AI agents access sensitive systems and datasets, particularly when workloads run across multiple users, tools and models.

Broader integration

HPE is also broadening the use of Nvidia components across the AI Factory portfolio. HPE AI Factory at Scale, HPE Sovereign AI Factory and HPE Private Cloud AI are now available with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs, Nvidia Spectrum-X Ethernet networking, Nvidia BlueField-3 DPUs and Nvidia ConnectX-8 SuperNICs.

For the Vera Rubin NVL72 systems, Nvidia networking will be integrated as standard, including Nvidia Vera BlueField-4 DPUs, Nvidia ConnectX-9 SuperNICs and Nvidia Spectrum-X Ethernet. Nvidia says Spectrum-6 switching delivers 1.6x higher networking performance for AI communication than standard Ethernet, although the announcement did not include independent benchmarks.

For larger-scale and sovereign workloads, HPE has also been adding Nvidia InfiniBand networking options, including Nvidia Quantum-X800 InfiniBand with the HPE Cray Supercomputing GX5000. These configurations are based on Nvidia reference architectures and are intended to support AI work from model development through production deployment.

The software ecosystem around the platform is expanding as well. HPE's Unleash AI partner program has added nearly a dozen software partners, including Aizen, BridgeTEK, deepset, Deliverance, Faclon Labs, Gallop, Rocket, Supervity, Thales, Trustwise and Vortiqx.

The breadth of the announcement shows how competition in enterprise AI infrastructure is shifting from individual chips and servers to more integrated stacks that combine compute, networking, security, storage and operational controls for AI systems expected to run continuously inside regulated organizations.

Spectrum-X Ethernet networking is now the standard approach across the HPE AI Factory with Nvidia portfolio, covering at-scale, sovereign and turnkey AI factory offerings available now.