HPE Introduces Liquid-Cooled Vera CPU Rack for Agentic AI Factories
HPE announced a liquid-cooled NVIDIA Vera CPU rack designed for agentic AI and reinforcement-learning workloads, alongside expanded governance and security support.
HPE introduced a high-density, liquid-cooled NVIDIA Vera CPU Rack as part of an October 21 expansion of its AI Factory with NVIDIA portfolio. The system is designed for CPU-intensive agentic AI, reinforcement-learning evaluations and data-processing pipelines that operate alongside GPU compute.
HPE says the rack will use 256 NVIDIA Vera CPUs and support 22,500 sandboxes. Its stated performance improvement of 80% for agentic CPU work is a vendor claim, not an independently verified benchmark. The company also plans support for NVIDIA’s OpenShell reference architecture, BlueField-4 security enforcement and Vera Rubin inference systems.
Private Cloud AI governance updates are rolling out in late 2026, while parts of the developer experience are planned for 2027. HPE did not disclose pricing, delivery volumes or customer deployments for the new rack.
Official HPE announcement. Development announced October 21, 2026; source reviewed October 23, 2026.