Wistron and Wiwynn Show NVIDIA Vera Rubin AI Racks at 2026 OCP Summit

Rack-scale platforms built around NVIDIA Rubin

According to Wistron, the company and Wiwynn are among the first with readiness for NVIDIA's latest rack-scale platform, the NVIDIA Vera Rubin NVL72. The system unifies 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs with NVIDIA NVLink, NVIDIA ConnectX-9 SuperNICs and NVIDIA BlueField-4 DPU.

Wistron describes the platform as fully liquid-cooled and built for the resiliency, energy efficiency and scalability that AI factories demand. It is optimized for frontier AI model training, inference and reasoning, the company says.

Also on show is the NVIDIA HGX Rubin NVL8 platform, which is aimed at scaling AI factories for enterprise. It combines NVIDIA Rubin GPUs, the NVIDIA NVLink 6 switch, NVIDIA ConnectX-9 SuperNICs, NVIDIA BlueField-4 DPUs and NVIDIA Spectrum-X Ethernet. Wistron says the design delivers optimized AI and high-performance computing (HPC) application performance with the fastest time to insights.

CPU racks and context storage for agentic AI

The NVIDIA Vera CPU rack targets reinforcement learning and agentic AI at scale. It is built on the NVIDIA MGX architecture and integrated with NVIDIA Spectrum-X Ethernet networking. According to the release, each rack supports over 22,500 concurrent execution environments and completes workloads twice as fast with twice the efficiency of traditional CPU infrastructure.

For security, the NVIDIA Open Agent Safety Platform combines OpenShell runtime controls on Vera CPUs with NVIDIA Sentry on BlueField-4 DPUs. This pairing provides host-independent monitoring and policy enforcement.

The NVIDIA CMX context memory storage platform is described as a new class of AI-native storage infrastructure designed to accelerate agentic reasoning at scale. It is powered by NVIDIA BlueField-4 STX Storage Processor, NVIDIA DOCA and NVIDIA AI software. According to the release, it:

• extends effective GPU KV cache capacity;
• enables high-bandwidth sharing of context across the AI pod;
• reclaims essential compute and electrical power to support agentic AI inference.

Co-packaged optics in Ethernet networking

Networking is covered by NVIDIA Spectrum-X Ethernet Photonics, based on Spectrum-6. It integrates 200G co-packaged optics (CPO) to support 800 GbE networking and was designed for the NVIDIA Rubin generation of NVIDIA's accelerated computing platforms.

The Spectrum-X Ethernet platforms address congestion at scale and improve reliability through co-packaged optics integration, according to the release. They also use direct liquid cooling to increase data center efficiency and provide the capabilities needed for peak AI network performance.

Event details and what it signals

The 2026 OCP Global Summit takes place October 13 to 15, 2026, at the San Jose Convention Center, 150 W San Carlos Street, San Jose, CA 95113. Wistron is at booth C74 and invites industry partners to explore its AI solutions and exchange insights on next-generation AI infrastructure.

Wistron describes itself as a global technology service provider delivering ICT products, AI infrastructure and manufacturing solutions, with more than 63,000 employees across North America, Europe and Asia. The showcase indicates that the company is preparing server and storage hardware for NVIDIA's Rubin generation, spanning compute, CPU, storage and networking. The performance and efficiency figures come from the company's own release and were not independently verified.