At HPE Discover Las Vegas on June 16, HPE and NVIDIA framed their expanded AI Factory platform around four words: security, governance, scale, and sovereignty. Not one mention of raw throughput in that list. That's a real shift from the "bigger and faster" pitch that's dominated AI infrastructure announcements for years, and it tells you where enterprise buyers' actual objections to agentic AI currently sit — not "can it work," but "can we trust it and keep it compliant while running it at scale."
What's actually new
The HPE ProLiant Compute DL394 Gen12 is built around NVIDIA's Vera CPU — notably a CPU announcement, not another GPU story, positioned for agent orchestration, reinforcement learning, and data processing rather than raw model-serving throughput. That's a signal in itself: as agentic workloads mature, the orchestration and data-processing layers around the model are becoming their own infrastructure bottleneck.
The NVIDIA Agent Toolkit bundles Nemotron open models, NemoClaw, and the NVIDIA OpenShell secure runtime into what the companies are calling an "agent operating system" — software meant to monitor agent behavior, enforce policy on what agents are allowed to do, and reduce deployment risk, instead of treating governance as something bolted onto infrastructure that wasn't built for it. NVIDIA Confidential Computing rounds it out, protecting sensitive workloads even from privileged access at the infrastructure level — relevant if you're in a regulated industry or running agentic workloads against sensitive data on shared infrastructure.
The timeline is the part people will skip past
New HPE Private Cloud AI features from this announcement land in July 2026. Agentic observability, data intelligence, HPE Alletra Storage MP X10000, and Agent Toolkit support with NemoClaw arrive in Q4 2026. The flagship piece — Private Cloud AI with the DL394 Gen12 and Vera CPU — doesn't ship until 2027. If you're planning infrastructure around this announcement, sequence your expectations to those actual dates, not the announcement date, or you'll be budgeting for capability that isn't available yet.
Why this lines up with the Five Eyes guidance
This lands in the same window as the joint Five Eyes guidance on careful agentic AI adoption, and the two reinforce each other in a useful way: regulators are telling enterprises to deploy agentic AI incrementally with real control points, and infrastructure vendors are now building the products that make that kind of controlled, observable deployment actually possible at the data-center layer instead of leaving it to application-level code.
If you're building out agentic AI infrastructure, put agent governance — policy enforcement, observability, secure runtime — on the vendor evaluation checklist as a first-class requirement, not a follow-up question after compute specs. And if your organization has data residency requirements, ask about sovereignty and confidential computing support explicitly now, because that's clearly where the roadmap is headed and it's easier to get answers before you're locked into a contract than after.