Huawei Debuts Multi-Layer 3D Architecture to Support Massive AI Hardware Arrays

As the hardware required for artificial intelligence becomes increasingly power-intensive, Huawei has introduced a novel vertical solution for infrastructure design. At an event in Wuhu, located in China's Anhui province, the technology firm revealed a "3D data center" featuring a four-story layout. This configuration stacks essential components—including power supplies, backup systems, cooling mechanisms, and computing hardware—on top of one another rather than spreading them across a single floor. The design is specifically engineered to accommodate supernode clusters that utilize more than 100,000 accelerator cards.

This shift toward verticality is a direct response to the skyrocketing energy needs of modern AI servers. According to Huawei's leadership, the power density of server racks has surged from just a few kilowatts to as much as 200 kW. With AI models now incorporating trillions of parameters, the demand for tokens is placing unprecedented pressure on physical facilities. By utilizing this stacked approach, Huawei aims to provide a more efficient environment for its Ascend supernode clusters, which the company expects will become a standard requirement for high-level AI development.

A central pillar of this new strategy is the transition from traditional construction to a product-based manufacturing model. Huawei reports that roughly 90% of the mechanical and electrical components in these facilities are prefabricated. This modular method significantly accelerates deployment; the Wuhu site was completed in approximately nine months, a sharp reduction from the typical 18-to-24-month timeline associated with standard builds. Furthermore, by isolating different infrastructure domains, the company allows for computing hardware to be upgraded or replaced without necessitating a total structural overhaul of the building.

Beyond the physical facility in Wuhu, Huawei is promoting this architecture as a repeatable industry standard. The company has released a comprehensive design library and a dedicated book on the subject to encourage broader adoption. While Huawei is already implementing this 3D model at its core cloud hubs in regions like Inner Mongolia and Gui’an, the long-term success of the strategy will depend on real-world performance metrics. Industry observers will be monitoring the system’s cooling reliability, energy efficiency, and its ability to remain compatible with hardware from various vendors as AI technology continues to evolve.

---

Read original at TechRepublic AI.