Huawei Digital Power has introduced a grid-interactive strategy for AI Data Centres (AIDC) to address the surging energy demands of large-scale AI models.

The strategy focuses on four technical areas: power supply (watt), thermal management (heat), digital operations (bit), and modular construction. Huawei aims to create a “grid-friendly” architecture that allows data centres to interact dynamically with power systems, particularly as renewable energy penetration increases.

The company’s “watt innovation” involves a multiple-input multiple-output (MIMO) power architecture. This system uses grid-forming energy storage and is designed to evolve toward solid-state transformer technology. For cooling, Huawei introduced a megawatt-level liquid cooling system to manage the high-density heat loads generated by AI chips.

“The upper limit of AI computing power is ultimately defined by the upper limit of energy supply,” said Bob He, Vice President of Huawei Digital Power. He proposed a new “TokEnergy Index” to measure the ratio of energy consumed to the number of AI tokens produced, shifting the industry focus from simple energy efficiency to value-based metrics.

Huawei stated that the strategy also includes prefabrication and modular construction methods. This allows for the testing and verification of data centre components in a factory setting before on-site installation, which the company claims reduces delivery times.