Hewlett Packard Enterprise introduced a liquid-cooled rack containing 256 Nvidia Vera CPUs at Nvidia GTC Berlin 2026 on 21 October, as it expanded its HPE AI Factory with Nvidia portfolio for enterprise AI deployments, StreetInsider reported. The additions span computing hardware, controls for AI agents and support for workloads in regulated or sovereign environments.
Key points
- HPE introduced a rack with 256 Vera CPUs for CPU-intensive agent and reinforcement-learning workloads.
- HPE says its Private Cloud AI platform can now run H200 and RTX PRO 6000 Blackwell Server Edition GPUs in one cluster.
- HPE plans to roll out new agent-governance tools in late 2026 and a self-service developer experience in 2027.
- HPE is expanding confidential-computing support for sovereign and regulated environments.
Vera rack joins HPE’s mixed-GPU clusters
HPE designed the Nvidia Vera CPU Rack by HPE for work that places heavy demands on CPUs, including AI agents and reinforcement learning. Its liquid-cooled configuration gives the company a distinct hardware addition to a portfolio that also includes GPUs and software for enterprise AI systems.
Within HPE Private Cloud AI, a single cluster can combine Nvidia H200 GPUs with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs. HPE says that mixed configuration is generally available. It gives customers a way to deploy the two specified GPU types together rather than confining each type to a separate cluster.
Nvidia’s open-source KAI Scheduler is also generally available on the platform. It allocates workloads according to quotas and priorities, providing a mechanism for assigning computing capacity when different users or applications share the system. Those scheduling rules sit alongside the new hardware options in HPE’s enterprise offering.
HPE Private Cloud AI adds agent controls
HPE is expanding Private Cloud AI through HPE AI Essentials, with governed APIs, Model Context Protocol servers for connecting agents to external tools, and development tools based on OpenAPI. The planned capabilities also include enterprise identity and access controls. HPE intends them to let developers, software vendors and agents use enterprise AI services while retaining central governance, StreetInsider reported.
The company also announced support for Nvidia’s OpenShell Reference Architecture for Safe AI Agents. That design brings together secure environments for running agents, hardware-based security enforcement through Nvidia BlueField-4 and Nvidia Vera Rubin NVL72 for model inference. AI Affairs previously reported Nvidia’s introduction of OpenShell as an architecture for governing agents on enterprise systems.
HPE said it will support Nvidia DSX POD as well. The modular, prefabricated reference design is intended to reduce integration work at the installation site for large AI factory deployments. AI Affairs reported Nvidia’s 14-day deployment target for DSX POD when the design was introduced.
HPE extends confidential computing for regulated workloads
For sovereign and regulated environments, HPE is expanding support for Nvidia Confidential Computing on HPE ProLiant Compute DL380a Gen12 servers fitted with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs, StreetInsider reported. The server update places another hardware configuration within the portfolio HPE is offering for deployments with data-control requirements.
Schneider Electric separately announced plans to combine HPE Private Cloud AI with its industrial technology expertise for AI agents deployed at the edge. The planned combination takes HPE’s platform into an industrial setting, alongside the data-centre and enterprise configurations covered by the other announcements.
HPE plans to roll out the new HPE AI Essentials capabilities in late 2026. HPE plans a self-service developer experience in 2027.
