CyberTAN Debuts Liquid-Cooled Design for Marvell’s AI Networking Chip
CyberTAN introduced a liquid-cooled prototype called Pallas for Marvell’s Teralynx T100 networking chip, aiming to address heat in dense AI data-center racks. The announcement did not include customer orders or shipment volumes. For Marvell investors, the key question is whether the thermal design can move from prototype to qualified, profitable production.
CyberTAN’s Pallas prototype targets Marvell’s Teralynx T100, a switch chip rated at 102.4 Tbps. The cooling setup is said to manage over 1,000 watts of concentrated heat, using a coolant distribution unit, temperature sensors, a 45°C liquid inlet, and 40-watt optical modules at an OCP Global Summit display. No orders or shipment figures were disclosed.
Marvell introduced the T100 on June 1, with sampling planned for that quarter. It claimed typical power under 1,000 watts and up to 25% lower consumption than rival offerings. A dated quote listed MRVL at $265.33 and a $232.66 billion market value, though the feed was delayed and not live.
If liquid-cooled networking designs advance beyond prototypes, data-center operators, chip suppliers, and cloud users could see denser AI racks run more reliably. Engineers and equipment makers may gain new work, while surrounding communities might face changed energy and resource demands. Benefits would depend on qualification, deployment scale, and operating costs; without orders, broader effects remain uncertain. Investors may treat the design as a possible step toward production rather than a proven social or environmental shift.