NVIDIA Vera CPU Targets Agentic AI with HPE Factory Expansion

2027 is the year agentic AI gets its own silicon. NVIDIA and HPE just expanded the HPE AI Factory lineup with the Vera CPU, purpose-built for the tool calls, orchestration, and real-time data processing that multi-agent systems demand. No more shoving agent workloads onto general-purpose cores that weren't designed for this loop.

Vera CPU: First Silicon for Agent Loops

Vera is the first CPU explicitly built for agents. It lands inside the HPE ProLiant Compute DL394 Gen12 server, part of HPE Private Cloud AI, in 2027. The New York Stock Exchange, working with Redpanda and HPE, is already an early enterprise customer evaluating Vera. Vera is a component of the broader Vera Rubin platform, which includes the NVL72 rack-scale system shipping with full-stack NVIDIA Confidential Computing across every chip. Vera Rubin targets frontier models above 1 trillion parameters. HPE is also bringing the Compute XD700, based on HGX Rubin NVL8, supporting up to 128 Rubin GPUs per rack.

Agent Toolkit Goes Live Now

You don't have to wait for Vera to start building safe agents. NVIDIA Agent Toolkit, including Nemotron open models, the OpenShell secure runtime, and NemoClaw blueprints, is available now with HPE Private Cloud AI. This gives enterprises an agentic AI operating system: monitor behavior, enforce governance policies, run autonomous multi-agent systems. HPE Private Cloud AI adds secure local agent registration, so you approve models, skills, and tools against centralized policies before they execute. New HPE Zerto capabilities detect rogue agent actions and can rewind state continuously. On the storage side, the HPE Alletra MP X10000 (NVIDIA-Certified at foundation level) auto-applies metadata and governance policies to unstructured data, improving token throughput.

Confidential Computing Covers the Whole Factory

NVIDIA Confidential Computing now spans every HPE AI Factory solution within the portfolio - HPE AI Factory at Scale, Sovereign AI Factory, and Private Cloud AI - via HPE Services. The HPE ProLiant DL380a also achieved NVIDIA-Certified Systems for confidential computing, validating application performance under encryption. BlueField DPUs and DOCA provide in-silicon zero-trust enforcement and runtime threat detection without performance tradeoffs. For the next-gen factories, every Vera Rubin NVL72 ships with Vera BlueField-4 DPUs, ConnectX-9 SuperNICs, and Spectrum-X Ethernet. Spectrum-6 switching delivers 1.6x higher networking performance versus off-the-shelf Ethernet. For large-scale sovereign workloads, HPE is also adding Quantum-X800 InfiniBand via the Cray GX5000.

Agentic AI infrastructure is shifting from bolted-on security to baked-in hardware trust. Vera's 2027 arrival gives the industry a clear target for purpose-built agent compute.

Source: HPE AI Factory With NVIDIA Expands for the Era of Agents
Domain: blogs.nvidia.com