The Data Center of the Future According to Huawei: A Complete Ecosystem of Hardware, Software, and Cooling Systems

Huawei updates its data center strategy for the era of artificial intelligence and presents a portfolio that covers multiple levels of infrastructure: from computing to networking to cooling. With the growth of AI applications, especially agent-based ones, and models tailored to different sectors, businesses need to rethink data centers in light of more stringent requirements for density, consumption, and operational continuity.

Huawei’s Solutions for the Data Center of the Future

Among the announced innovations, the most important is probably the Huawei AI Data Platform, a solution designed to support AI agents in the enterprise sector and tackle some of the most common bottlenecks, such as inefficient memory usage and slow inference.

This proposal integrates three components: knowledge generation and retrieval, memory extraction and recall, and inference acceleration through the Unified Cache Manager.

On the networking front, Huawei has introduced Xinghe AI Fabric 2.0, an evolution of the AI Fabric solution introduced in 2018. The architecture is built on three levels: Intelligent Brain, Intelligent Connectivity, and Intelligent Network Elements, and incorporates four main functions:

Rock-Solid Architecture 2.0, which ensures three levels of high reliability thanks to the Eagle-Eye Engine;
StarryWing Digital Map 2.0, offering three levels of automation with NetMaster;
Xinghuan AI Turbo 2.0, which provides capabilities for Network Packet Load Balancing (NPLB) and Network Stream Load Balancing (NSLB) technologies;
iFlashboot 2.0, enabling ultra-fast reboots in 5 seconds to ensure operational continuity.

Along with Xinghe AI Fabric 2.0, Huawei also announced a new Scenario Driven Foundation Model Solution, designed to support the implementation and development of AI data centers with a scenario-based approach. The goal is to build data center networks for agent-based AI applications.

As for networking hardware, the company has made available the CloudEngine XH9230-128DQ-LC, the first commercially available fixed switch featuring liquid cooling and equipped with 128 400GE ports. This system uses a proprietary cooling structure and a new thermal conduction material, allowing for liquid cooling of 100% of the optical modules. The result, according to Huawei, is a heat dissipation double that of the industry average and the ability to install up to eight switches per cabinet, improving space utilization.

This is complemented by the optical module family StarryLink 800GE and 400GE, designed for high-density environments and long-distance transmissions. The 800GE portfolio also includes modular switches CloudEngine XH16800, with up to 768 800GE ports, in addition to fixed switches CloudEngine XH9330 with 128 800GE ports and CloudEngine XH9320 with 64 800GE ports.

Additional innovations are related to cooling systems: Huawei introduced the Large Temperature Difference Cooling System, developed to address the rise of data centers and increasingly stringent constraints on PUE (Power Usage Effectiveness), which for new facilities is indicated at 1.25 or less. According to the company, the traditional water-based approach is reaching a physical limit, necessitating an architectural change.

Huawei's proposal aims to optimize the thermal exchange network and, based on the data released by the company, to improve cooling efficiency by over 30%. The benefits are threefold: an architectural revision for better utilization of free cooling, a progressive improvement in PUE through advanced simulations and component optimization, and increased operational reliability, with more stable temperatures and greater safety for the equipment.