Huawei Unveils 3D Data Center: AI Clusters Could Scale Beyond 100,000 Cards
Huawei unveiled a four-layer 3D data center designed for massive AI clusters as it pushes prefabrication and standardized infrastructure. The post Huawei Unveils 3D Data Center: AI Clusters Could Scale Beyond 100,000 Cards appeared first on TechRepublic .
Huawei has introduced a four-layer 3D data center model in Wuhu, Anhui province, China, which can potentially host clusters with over 100,000 accelerator cards, according to the Chinese tech giant. This innovative architecture, designed for massive AI clusters, features cooling, computing equipment, power systems, and backup power stacked vertically.
Huawei's Senior Vice President Ren Shulu explained that server rack power has risen from several kilowatts to 100 kW and even up to 200 kW, with expectations of further increases. Deputy Chairman and Rotating Chairman Wang Tao highlighted that "massive AI models with trillions of parameters and the growing demand for tokens" are driving the need for higher-capacity AI infrastructure.
Huawei's approach to data center construction emphasizes prefabrication and standardization, aiming to reduce construction time from 18-24 months to just three months. The Wuhu facility took about nine months to build, showcasing the efficiency of this method. Huawei's design separates different infrastructure domains, which may isolate failures and enable hardware upgrades without rebuilding the facility.
The company is treating the data center as a standardized, replicable product rather than a one-off engineering project. Huawei's 3D data center is not just a showcase; it is being deployed in Huawei's cloud business sites in Gui’an, Inner Mongolia, and Wuhu. However, for this architecture to become an industry standard, operators need to validate cooling reliability, power efficiency, maintenance procedures, and compatibility with various hardware vendors.
Huawei's announcement indicates a shift in how AI capacity planning should be approached, focusing not just on the number of accelerators a facility can accommodate but also on the adaptability of surrounding infrastructure as computing requirements evolve.
Written by urgent.news from TechRepublic's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.