San Jose – October 13, 2026 -- CyberTAN Technology has unveiled a liquid-cooled Ethernet switch platform architecture built around Marvell's 102.4 Tbps Teralynx T100 silicon, and says a live demonstration at OCP Global Summit 2026 reached a Power Usage Effectiveness (PUE) of 1.1.
Pallas Series targets 1,000-watt heat loads with direct-to-chip cooling
The Pallas Series 1.6T platform is designed to accommodate the 3nm Marvell T100 and its 25% energy-saving advantage. CyberTAN pairs the silicon with its proprietary Direct-to-Chip (D2C) liquid-cooling engineering. The company says the design can manage more than 1,000 watts of localized heat dissipation while keeping port density high.
CyberTAN frames the work as a response to thermal and power limits in AI clusters, where generative AI and hyperscale workloads are driving demand for more efficient silicon and rack-level cooling.
Executives describe a production-oriented path for 1.6T connectivity
CyberTAN Chairman and CEO GwongYih Lee called the collaboration a milestone in bringing "production-ready 1.6T connectivity architectures" to market. Rishi Chugh, vice president and general manager of Marvell's Cloud Switch Business Unit, said the architecture shows how systems can use the T100's bandwidth and power-saving capabilities to ease networking bottlenecks. The source gives no commercial availability date or pricing.
Booth B29 demo validates 40W optics at 45°C inlet temperature
At Booth B29, CyberTAN is showing a pre-stage development switch platform. The prototype includes the schematic chassis, optical housing and internal cabling, plus a specialized thermal module that simulates T100 silicon workloads.
A Coolant Distribution Unit (CDU) drives a live liquid-flow demonstration monitored by multi-point temperature sensors. According to CyberTAN, the setup validates safe operation of 40W OSFP ZR/ZR+ optics at a 45°C liquid inlet temperature while achieving the PUE of 1.1.