Delivering Tailored AI Infrastructure: Pegatron's Rack Innovation and Manufacturing Expertise

At OCP Global Summit 2025, Pegatron SVR presented its end-to-end AI infrastructure capabilities, from board-level design and smart manufacturing to full rack-scale liquid-cooled cluster deployment. The presentation covered Pegatron's scale as a Fortune Global 500 ODM/OEM manufacturer (2023 revenue $40.3B, ~100,000 employees), its global expansion to Mexico and Texas, and the three core AI infrastructure challenges of power, thermal, and scaling cost. Concrete deployments included a 1,024-GPU NVIDIA HGX B200 installation delivering 9,216 petaFLOPS in a US generative AI data center.

Key Takeaways

• Pegatron achieved a GB300 assembly yield improvement from 57% to over 90% using simulation-synchronized robotic arms in its smart manufacturing system.
• Pegatron's Texas facility (FATP + rack-and-stack) is planned for Q1 2026, complementing existing full-scale manufacturing in Taiwan and Mexico.
• The liquid cooling service chain covers cold plate design through L12 cluster CDU simulation and onsite deployment, with CDU redundancy options including N+1 and N+N.
• A production deployment of 1,024 NVIDIA HGX B200 GPUs providing 9,216 petaFLOPS FP4 was completed at a US generative AI software data center approximately four to five months before the October 2025 presentation.
• Pegatron's AI compute portfolio spans NVIDIA GB300/GB200 NVL72, HGX B200/B300 (liquid and air cooled), AMD Instinct MI355X/MI325X, and MGX add-in card configurations including Vera Rubin on the roadmap.
• Data center power footprints have escalated from ~10-15 MW a decade ago to gigawatt scale in 2025, with power, thermal, and scaling identified as the three top AI infrastructure challenges.