AI demands more power

The rise of generative artificial intelligence is causing a deep transformation of data center infrastructures. Servers equipped with GPUs require more electrical power and appropriate cooling systems, which complicates the construction of facilities.

Prefabricated solutions for efficiency

To meet this constraint, Schneider Electric is proposing an approach based on prefabricated equipment, tested in the factory and deployable in stages. Prefabricated electrical infrastructures to save time.

Standardized designs for faster deployment

Schneider Electric presented two solutions: prefabricated power modules and electrical skids. The objective is to reduce the construction and deployment time of the electrical infrastructure required for AI training clusters and high-performance computing environments. The company is relying on a reference design developed with NVIDIA.

High power needs for AI servers

The NVIDIA GB300 NVL72 architecture serves as a reference for the design. It allows reaching a power of up to 142 kW per rack and integrates the electrical power supply and liquid cooling systems necessary for the operation of high-density computing infrastructures. This combination meets the requirements of AI servers, whose GPUs generate energy and thermal needs significantly higher than those of traditional computing equipment.

Integrated power and cooling systems

Power supply and cooling must therefore be considered together from the design of the data center. Schneider Electric's solutions include the Galaxy VXL, a compact three-phase inverter designed for high-energy density infrastructures. The prefabricated power sets are available in three power levels: 2 MW, 2.25 MW and 2.5 MW.

Modular approach for cost control

A gradual increase in power to limit costs. Beyond the timeframes, modularity should allow operators to better control their investments. Instead of immediately building an infrastructure sized for future needs, they can add new modules as demand increases.

Targeting hyperscale and cloud providers

This logic concerns in particular hyperscale data center operators, cloud providers specialized in AI, sometimes called Neoclouds, as well as colocation actors who host the infrastructures of several clients. A European launch before an international expansion.

European production and global rollout

The new equipment must be designed, manufactured and tested at Schneider Electric's site in Sant Boi de Llobregat, south of Barcelona. The company plans to prioritize the European market first, before commercializing in North America in 2026, and then expanding to the Asia-Pacific region and other regions.

Meeting the needs of AI infrastructure

Schneider Electric thus aims to address a central challenge for operators: having a reliable power supply that meets the requirements of AI infrastructures. The ability to deploy and expand these installations is indeed a determining factor, as computing and energy needs continue to grow.