This OCP Global Summit 2025 executive session features a Wiwynn spokesperson presenting the company's strategic position in the AI infrastructure buildout. The talk covers the explosive growth in AI compute requirements since the Transformer architecture's emergence, the resulting hardware challenges in cooling, power, and signal integrity, and how Wiwynn is evolving from a rack-level ODM to a full AI cluster deployment partner. Key themes include the diversification of AI cloud customers (from hyperscalers to neoclouds and sovereign operators) and Wiwynn's smart manufacturing capabilities.

Key Takeaways

• AI computing power requirements have grown from 2x per six months to 256x every two years since 2018, driven by the Transformer architecture scaling AI models to billions of parameters.
• Leading AI accelerators (Nvidia Blackwell, AMD MI350) now exceed 1,000W TDP per chip; rack-level power is expected to surpass 1 MW within a few years, necessitating HVDC power distribution and advanced liquid cooling.
• Cooling technology roadmap: heatsink solutions grew 20% per two years; cold plate and immersion cooling now doubles every two years; IC-package-integrated cooling is projected to grow 3x every two years.
• Wiwynn is expanding from rack-level ODM to full AI cluster deployment, covering Level 12 architecture design, on-site installation, and LLM workload validation — as demonstrated with YTL and Nvidia in Malaysia.
• The AI data center customer base is diversifying: traditional hyperscalers (AWS, Google, Microsoft) are now joined by AI model providers (OpenAI, Anthropic), GPU neocloud giants (CoreWeave, Nebius), and ~20 sovereign neocloud operators (YTL, Tata, G42).
• Wiwynn's smart manufacturing integrates AI and robotics, with automated testing simultaneously validating over 1,000 servers; automation spans warehousing, assembly, testing, and quality control.
• Wiwynn and Wistron are co-developing the GB200 NVL72-based rack; OCP 2025 booth B13 demos include single/double-side cold plate, DLC, two-phase and immersion cooling, next-gen double-wide rack with in-row CDU (AMD-based), and Nvidia NVDS servers.