PEGATRON Unveils Future-Ready AI Solutions at NVIDIA GTC 2026

PEGATRON Unveils Advanced AI Platforms at NVIDIA GTC 2026

PEGATRON, a premier global provider of server solutions, is set to make waves by showcasing its latest AI infrastructure at NVIDIA's GTC 2026 event. This exciting reveal includes platforms powered by the cutting-edge NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8, promising revolutionary performance for both hyperscale and enterprise applications.

Next-Generation AI Supercomputing

One of the stars of the presentation is the Pegatron RA4803-72N3, a liquid-cooled rack-scale AI supercomputer specifically engineered for demanding workloads. Equipped with an impressive 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs, this system utilizes NVIDIA's NVLink 6 networking, ConnectX-9 SuperNICs, and BlueField-4 DPUs. Together, they deliver a staggering 3.6 EFLOPS inference capability alongside a bandwidth of 260 TB/s and an extensive 20.7 TB of HBM4 memory. Compared to its predecessor, the NVIDIA Blackwell architecture, this new system requires only a quarter of the GPUs to effectively train mixture-of-experts models, thereby slashing inference costs by up to 10 times while achieving five times the throughput. This dramatic enhancement not only improves performance efficiency but also minimizes the physical footprint and overall costs associated with AI operations.

Enhanced AI Infrastructure Scaling

Accompanying the RA4803 system is the Pegatron AS210-2T1-8H3, a compact 2U liquid-cooled server designed to supercharge enterprise AI workloads. By integrating NVIDIA RGX Rubin GPUs and dual Intel Xeon 6 processors, it enhances performance across AI and high-performance computing (HPC) tasks. The architecture can scale up significantly, with models like the RA4800-64H3 and RA4800-72H3 achieving ultra-high-density configurations. The RA4800-64H3 supports multi-system integration, incorporating up to 64 NVIDIA HGX Rubin GPUs and 16 Intel Xeon 6 processors, while the RA4800-72H3 can fit as many as nine 2U systems in a single 48U rack, further elevating compute density and accelerating insight generation for large-scale AI initiatives.

Exceptional Performance for Diverse Applications

The launch also features Pegatron’s servers that deploy NVIDIA RTX PRO Blackwell Server Edition GPUs, optimized for a range of AI and digital workloads. The MS303-2A1G-P60 server, for instance, facilitates up to four liquid-cooled NVIDIA RTX PRO 6000 GPUs complemented by dual AMD EPYC™ 9005 processors. With rapid 96GB GDDR7 memory, it assures high performance across multiple demanding tasks. Additionally, the AS205-2T1 model maintains efficiency by supporting up to four NVIDIA RTX PRO 4500 GPUs designed for enterprise and edge applications.

Dr. James Shue, Senior Vice President and Chief Technology Officer of Pegatron, emphasized the need for comprehensive system integration expertise in constructing AI factories. He noted that Pegatron's advanced platforms enable businesses to transition from infrastructure development to operational AI productivity swiftly, efficiently, and with increased confidence.

Discover the Next Wave of AI Technology

Attendees at NVIDIA GTC 2026 are encouraged to visit Pegatron's Booth #830. They will see firsthand how the integration of rack-scale systems, cutting-edge cooling technologies, and large-scale manufacturing processes are driving the future of AI infrastructure.

For more details, you can check out the PEGATRON SVR website and follow them on LinkedIn and YouTube for updates:

About PEGATRON

PEGATRON Corporation specializes in delivering advanced manufacturing and design services. With extensive experience in server development, it is committed to innovation through vertical integration and a strong focus on meeting the ever-evolving needs of cloud service providers and enterprise data centers. Whether it’s through liquid-cooled, air-cooled, or AI-focused server solutions, PEGATRON is a name synonymous with excellence in the server industry.