CyberTAN Unveils Pallas Series Liquid-Cooled 102.4 Tbps Ethernet Switch Platform Blueprint at OCP Global Summit 2026

The collaboration between CyberTAN and Marvell focuses on addressing heat dissipation and energy consumption issues that arise with scaling generative AI and hyperscale workloads. CyberTAN's Pallas Series platform is built to accommodate Marvell's 3nm T100 architecture, which is reported to offer a 25% energy saving. This integration aims to enable high port density while maximizing the efficiency benefits of the silicon.

A core feature of the Pallas Series is its Direct-to-Chip liquid cooling. CyberTAN states this system is designed to manage over 1,000 Watts of localized heat dissipation. During a demonstration at OCP 2026, the platform exhibited a Power Usage Effectiveness (PUE) of 1.1, while safely handling 40W OSFP ZR/ZR+ optics under a 45°C liquid inlet temperature. This PUE figure indicates the system's efficiency in using power for computing versus cooling and other overheads.

The development targets global hyperscalers, offering network switch designs that prioritize high-density, low-latency, PUE optimization, and thermal control. CyberTAN, headquartered in Taiwan, positions its Pallas Series as a component for next-generation AI data center hardware, focusing on sustainable and efficient networking to support the demands of the AI era.