Pegatron Unveils Next-Gen AI Server Platforms at NVIDIA GTC 2026

Pegatron unveiled four next-generation AI infrastructure platforms at NVIDIA GTC 2026, headlined by the RA4803-72N3 — a rack-scale liquid-cooled supercomputer powered by NVDA's new Vera Rubin NVL72 architecture. The flagship system integrates 72 NVIDIA Rubin GPUs, 36 Vera CPUs, and ConnectX-9 SuperNICs capable of up to 800Gb/s per port, delivering 3.6 exaflops of inference performance, 260 TB/s memory bandwidth, and 20.7 TB of HBM4 memory.

The product lineup spans multiple tiers: the AS210-2T1-8H3 is a 2U liquid-cooled enterprise server based on NVIDIA HGX Rubin NVL8 with dual Intel Xeon 6 processors, while the MS303-2A1G and AS402-based systems target generative AI and visualization workloads using the RTX PRO Blackwell Server Edition. The booths showcased SFP+ port connectivity and Front IO configurations across high-density rack designs, underlining Pegatron's positioning as a full-stack server OEM rather than a pure contract manufacturer.

For investors tracking the AI infrastructure supply chain, Pegatron's GTC presence signals continued enterprise buildout demand centered on NVIDIA's Vera Rubin and Blackwell platforms. The Taiwan-listed ODM (4938.TW / OTC: PGTRF, ~.9B market cap) competes with Supermicro and Quanta for AI server contracts, making NVDA ecosystem momentum a key revenue driver heading into 2026.