Pegatron introduces Huida DSX AI factory architecture to strengthen mass production of advanced servers

According to Pegatron's plan, the company uses the workflow of the Huida DSX AI factory platform architecture to connect digital twin simulation and optimization to the physical production line at the manufacturing base. (File photo)

[Reporter Fang Weijie/Taipei Report] Pegatron (4938) hosted the 2026 Open Computing Project (OCP) Global Summit to announce the layout of the next-generation AI infrastructure. Through the NVIDIA DSX AI factory platform architecture, digital twin simulation and physical manufacturing processes are connected in series, covering automated production, heat dissipation verification and cabinet-level stability testing, strengthening the integration capabilities of AI servers from design, manufacturing to large-scale deployment.

Zheng Guangzhi, General Manager and Co-CEO of Pegatron, said that the key criterion for AI infrastructure is how efficiently it can move from engineering development to actual deployment. Pegatron combines system design, vertically integrated manufacturing, OCP-compliant architecture, and cabinet-level verification capabilities before deployment to help customers accelerate the transition of AI infrastructure from design to mass production deployment.

According to Pegatron's plan, the company uses the workflow of the Huida DSX AI factory platform architecture to connect digital twin simulation and optimization to the physical production line at the manufacturing base, covering automated production, heat dissipation verification, and stability testing from the system to the cabinet level, promoting the AI ​​infrastructure from engineering development to actual deployment.

In terms of product layout, Pegatron is exhibiting the Vera Rubin NVL72 cabinet-level AI platform, which integrates 72 Rubin graphics processors (GPUs), 36 Vera central processing units (CPUs) and sixth-generation NVLink high-speed interconnect technology. It adopts a fully liquid-cooled cabinet design to demonstrate the integration and manufacturing capabilities of high-density AI computing systems.

In addition, Pegatron simultaneously demonstrated a full liquid-cooled system using Huida's HGX Rubin NVL8 architecture. A single system integrates 8 Rubin GPUs and is paired with a 48U liquid-cooled cabinet. It can integrate 64 or 72 GPUs according to different configurations, expanding the layout of cabinet-level AI computing products.

In response to the application needs of agent-based AI (Agentic AI), Pegatron also showcased the Vera CPU liquid cooling system and Groq 3 LPX, which support the computing coordination, data processing and low-latency inference applications required by AI agents, further expanding the AI ​​infrastructure product line.

Pegatron mentioned that its platform adopts the OCP open architecture standard, covering AI cabinets, accelerated computing, general computing and storage infrastructure, and demonstrated computing and storage platforms using AMD EPYC 9006 series processors, continuing to strengthen the integrated deployment capabilities from a single system to AI clusters and data centers.