San Jose – October 12, 2026 -- Wistron Corporation and Wiwynn Corporation say they are among the first manufacturers ready to build NVIDIA's Vera Rubin NVL72 rack-scale platform, which unifies 72 Rubin GPUs and 36 Vera CPUs in a fully liquid-cooled design. The two companies are showing the system at the 2026 OCP Global Summit in San Jose under the theme "Create AI Beyond Limits."
Vera Rubin NVL72 combines 72 GPUs, 36 CPUs and BlueField-4 DPUs
The NVL72 platform links 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs through NVIDIA NVLink. It also integrates NVIDIA ConnectX-9 SuperNICs and NVIDIA BlueField-4 DPUs. Wistron and Wiwynn describe it as built for the resiliency, energy efficiency and scalability that AI factories demand, and optimized for frontier AI model training, inference and reasoning.
Vera CPU rack supports more than 22,500 concurrent environments
The NVIDIA Vera CPU rack targets reinforcement learning and agentic AI. It is built on the NVIDIA MGX architecture and integrated with NVIDIA Spectrum-X Ethernet networking. Each rack supports more than 22,500 concurrent execution environments and completes workloads twice as fast with twice the efficiency of traditional CPU infrastructure, according to the source.
Security runs through the NVIDIA Open Agent Safety Platform. It pairs OpenShell runtime controls on Vera CPUs with NVIDIA Sentry on BlueField-4 DPUs, providing host-independent monitoring and policy enforcement.
NVIDIA CMX storage platform extends GPU KV cache capacity
NVIDIA CMX is a context memory storage platform powered by the BlueField-4 STX Storage Processor, NVIDIA DOCA and NVIDIA AI software. It extends effective GPU KV cache capacity and enables high-bandwidth sharing of context across the AI pod. It also reclaims compute and electrical power for agentic AI inference.
Spectrum-X Photonics integrates 200G co-packaged optics for 800 GbE
NVIDIA Spectrum-X Ethernet Photonics, based on Spectrum-6, integrates 200G co-packaged optics to support 800 GbE networking. It was designed for the Rubin generation of NVIDIA's accelerated computing platforms. The platform uses direct liquid cooling to raise data center efficiency and addresses congestion at scale.
HGX Rubin NVL8 targets enterprise AI factories
The NVIDIA HGX Rubin NVL8 platform combines Rubin GPUs, the NVLink 6 switch, ConnectX-9 SuperNICs, BlueField-4 DPUs and Spectrum-X Ethernet. It is aimed at enterprise AI and high-performance computing workloads.
The 2026 OCP Global Summit runs October 13-15 at the San Jose Convention Center. Wistron is at Booth C74.