SoftBank taps ZutaCore, Foxconn for robot AI datacentre first
AI servers from Foxconn can be managed by robots as the result of rack development by Sofbank with direct chip cooling.
The development is significant as Softbank is funding the $100bn Stargate project in the US with Opena AI that aims to roll out AI datacentres across the country with Nvidia Blackwell GPU chips and highlights Foxconn as a potential equipment supplier. The rack design is also ‘robot friendly’, allowing automated replacement of servers which are rollout out in clusters that can have 100,000 GPUs.
Softbank says this is the first demonstration of ZutaCore’s two-phase DLC (Direct Liquid Cooling) technology in an AI server using the current generation Nvidia H200 Hopper GPUs at the rack scale. This cools the cold plate on the GPU inside the Foxconn server by circulating an insulating refrigerant without water in a sealed two-phase system with liquid and gas.
However Pegatron has already certified the ZutaCore technology for its AI server boards, and Iceotope in the UK has also launched a direct cooling system for AI server racks.
- The art of cooling
- Liquid cooling for H200 GPUs
- Pegatron certifies ZutaCore DLC technology for AI servers
SoftBank has also designed and developed a rack-integrated system that integrates each component of the server, including cooling equipment with two-phase DLC technology, on a rack scale, and conducted an operational demonstration and performance evaluation at its data centre in February 2025.
“As we work to develop one of the largest AI infrastructures in Japan and homegrown largescale language models (LLMs), improving the energy efficiency of data centers is essential to accelerate AI development and realize a sustainable society. The operational demonstration of this rack-integrated solution with optimized two-phase DLC technology has confirmed stable operation and improved energy efficiency of high-density GPU servers with liquid cooling. In the future, we will work on the practical application of this solution for use in AI data centers and AI factories to build a sustainable AI infrastructure and maximize computational efficiency,” said Hironobu Tamba, Vice President and Head of the Data Platform Strategy Division at SoftBank.
- Iceotope launches AI cooling system
- Nvidia pushes Blackwell as the start of the AI factory
- Nvidia readies Blackwell Ultra, Vera, Rubin
This plug and play system simplifies the installation and rollout of AI racks in datacentres with 21-inch servers and conventional 19-inch servers compliant with the ORV3 standard from the Open Compute Project (OCP). The rack has airflow for improved cooling efficiency, while the power supply and wiring are concentrated in the rear of the rack for operational safety and, interestingly, for robot-friendly operation. The system also automatically recognizes peripherals and makes the necessary settings when peripherals are connected to the main unit of a device.
The design of the rack with the direct cooling allows the automated robot management of the datacentre that is key to the rollout of AI factories.
The demonstration results indicated the implementation passed Nvidia’s temperature test (NVQual), confirming the compatibility, stability and reliability of the rack, which achieved pPUE (partial Power Usage Effectiveness) of 1.03 (actual measured value) for cooling efficiency.
The Hopper H200 GPU has a thermal envelope of 700W, while Blackwell marks a step up in cooling requirements in the datacentre with the need to handle 1200W. With the spread and increased adoption of AI, demand for AI servers and other computing resources is expected to expand significantly, further increasing power consumption at datacentres and the need for the direct chip cooling.
“This groundbreaking collaboration highlights the transformative potential of our waterless DLC cooling technology to deliver unmatched efficiency while addressing the growing need for sustainable energy solutions. Together, we are not just meeting today’s challenges but setting a new standard for energy-efficient innovation in next-generation AI data centers and factories,” said Erez Freibach, Co-founder and CEO at ZutaCore said,
SoftBank, ZutaCore and Foxconn are working on on the development of rack-integrated systems with two-phase DLC technology for practical use and deployment in the global market, which is expected to include a Blackwell version of the rack.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
