NVIDIA and HPE Expand AI Factory for the Agentic Enterprise
At HPE Discover Las Vegas, NVIDIA and HPE announced a major expansion of their AI Factory partnership, adding NVIDIA's Vera CPU, Agent Toolkit, and Confidential Computing to move enterprise agentic AI from proof of concept into production.
NVIDIA and Hewlett Packard Enterprise used HPE Discover Las Vegas to announce the most substantive expansion of their AI Factory partnership yet — one aimed squarely at getting autonomous agent workloads out of the lab and into enterprise production.
The headline addition is the NVIDIA Vera CPU, now included in HPE Private Cloud AI. Vera is not a GPU. It is NVIDIA’s first processor designed from the ground up around the orchestration demands of agentic systems — tool calls, real-time data processing, and the coordination overhead of running multi-agent pipelines. That kind of work is architecturally different from training or dense inference, and existing CPU designs were not built for it. Vera ships in the HPE ProLiant Compute DL394 Gen12 server, with general availability set for 2027.
The second major piece is the NVIDIA Agent Toolkit, which lands with HPE Private Cloud AI as a software layer for safely running long-horizon autonomous agents. The toolkit bundles:
- NVIDIA Nemotron open models — compact, enterprise-tunable LLMs optimized for agent reasoning
- NVIDIA NemoClaw — a set of blueprints for multi-agent orchestration patterns
- NVIDIA OpenShell — a secure runtime that monitors agent behavior, enforces governance policies, and contains blast radius if an agent goes off-script
The third component is NVIDIA Confidential Computing, which keeps workload data encrypted even while in use on the GPU — a requirement for regulated industries like finance and healthcare that are otherwise blocked from deploying sensitive data in agent workflows.
HPE is framing this as an “agentic enterprise operating system.” That description has weight given the technical specifics: the combination of a CPU optimized for orchestration overhead, a governance-aware agent runtime, and confidential compute is genuinely new architecture, not just a rebrand.
Immediate availability: HPE AI Factory with NVIDIA RTX PRO Blackwell Server Edition GPUs, Spectrum-X Ethernet, BlueField-3 DPUs, and ConnectX-8 SuperNICs is available now. The Nemotron/NemoClaw software stack arrives in Q4 2026. The Vera-based DL394 hardware ships in 2027.
The timing is deliberate. HPE CEO Antonio Neri used the keynote to argue that most enterprise AI deployments are stalled at the PoC phase not because the models are wrong but because the infrastructure around them — governance, security, compute separation — is missing. That is the problem this stack is solving.
For developers and platform engineers: if your organization is evaluating on-premises agentic AI infrastructure, this partnership raises the bar on what “enterprise-ready” looks like. The NVIDIA OpenShell governance layer in particular — monitoring agent behavior at runtime rather than just at design time — is a pattern that cloud-native deployments currently lack.