Schneider Electric is pushing modular power infrastructure deeper into the AI data center buildout with a new generation of prefabricated power modules and power skids capable of delivering up to 2.5MW per unit. The company introduced the solutions on September 16, positioning them for hyperscale, neocloud and colocation operators dealing with rapidly increasing compute density. The portfolio comes in 2MW, 2.25MW and 2.5MW configurations, with the largest version representing Schneider Electric’s highest-capacity single-container prefabricated power solution to date.

The move reflects a broader shift in AI infrastructure procurement, where operators increasingly need power systems that can reach deployment faster without treating electrical infrastructure as a bespoke construction project. Schneider Electric says its new standardized equipment can support AI training clusters, high-performance computing and GPU-intensive workloads, with deployment timelines as short as six months. The company manufactures and tests the modules at its expanded facility in Sant Boi de Llobregat, Barcelona, creating a controlled production path for equipment that would traditionally require more extensive site-level engineering. This approach makes power capacity itself a more modular component of AI infrastructure planning.

Schneider Electric Targets AI Power Density

The new platform runs on Schneider Electric’s Galaxy VXL uninterruptible power supply technology, extending a UPS architecture that the company has positioned for high-density AI and large-scale electrical workloads. Galaxy VXL supports modular configurations and high power density, while Schneider Electric’s product portfolio lists power ranges reaching 1.5MW for the UPS system itself. The new prefabricated architecture packages that power technology into larger deployment blocks, allowing operators to assemble capacity around the requirements of dense compute environments. That distinction matters as AI facilities move toward higher rack densities and more demanding electrical profiles.

Schneider Electric describes the products as “scalable, high-density power solutions designed for hyperscale, neocloud and colocation data center environments.” The wording points to a market broader than traditional hyperscalers, with neocloud providers and colocation companies increasingly building infrastructure around accelerated computing demand. These operators face similar constraints around grid capacity, equipment availability, construction schedules and increasingly complex cooling requirements. A standardized power block gives them a way to align electrical deployment with the pace at which AI hardware enters production.

The 2.5MW capacity is particularly significant because AI infrastructure is changing the relationship between IT equipment and facility power systems. NVIDIA’s GB300 NVL72 platform combines 72 Blackwell Ultra GPUs with 36 Grace CPUs in a fully liquid-cooled rack-scale architecture, creating a substantially different facility profile from conventional enterprise computing. NVIDIA lists the GB300 NVL72 as an architecture designed for AI reasoning workloads, with high-bandwidth networking and liquid cooling built into the platform. Schneider Electric’s latest infrastructure therefore targets a compute architecture where power delivery and thermal management must operate as interconnected facility systems.

NVIDIA Collaboration Connects Power And Cooling

Schneider Electric built the new power modules around reference designs developed with NVIDIA for its latest AI infrastructure platforms. Those designs include integrated power management and liquid cooling controls supporting NVIDIA’s GB300 NVL72 platform, extending the companies’ earlier work on AI-ready facility architectures. Schneider Electric previously introduced reference designs that connect electrical power monitoring, building management and liquid-cooling controls with NVIDIA Mission Control. The strategy moves infrastructure engineering closer to the GPU platform itself instead of leaving operators to integrate each facility layer independently.

The integration matters because liquid cooling changes more than the thermal system inside an AI facility. Electrical distribution, cooling controls, facility monitoring and compute orchestration increasingly need coordinated operating data as GPU clusters respond to changing workloads. Schneider Electric’s GB300 reference designs already cover facility power, facility cooling, IT space and lifecycle software, with dedicated designs for liquid-cooled AI clusters. Its reference architecture for GB300 NVL72 deployments targets facilities reaching roughly 7.5MW and rack densities around 142kW, illustrating how quickly infrastructure requirements are moving beyond conventional data center assumptions.

The new modules extend that engineering philosophy into a deployable physical product. Instead of treating the electrical room, UPS system and associated distribution equipment as separate site packages, Schneider Electric is offering a factory-engineered power block that can arrive with much of the integration work already completed. The company says the modules and skids are engineered, manufactured and tested before deployment, reducing the amount of construction activity required at the customer site. That model can shift more commissioning and integration work into a controlled manufacturing environment.

Prefabrication Becomes An AI Infrastructure Strategy

Schneider Electric’s announcement arrives as AI infrastructure developers face pressure to shorten the period between power availability and compute deployment. A facility can have land, network connectivity and GPU supply lined up while still waiting on electrical equipment and construction sequencing. Prefabricated power systems attack that bottleneck by standardizing portions of the infrastructure that previously required project-specific engineering and installation. For operators, the value lies not only in megawatts but in how predictably those megawatts can reach an operating AI cluster.

The company says the new solutions can support AI-ready infrastructure deployment in as little as six months, a timeline aimed at the accelerating cadence of GPU infrastructure expansion. Schneider Electric initially launched the power modules and skids in Europe, with North American availability planned later in 2026 and additional international markets and Asia-Pacific markets to follow. The initial European rollout gives the company a first market for the platform while expanding production around a product designed for repeatable deployment. This creates a potential bridge between standardized electrical design and the increasingly global footprint of AI compute operators.

Schneider Electric’s 2.5MW offering ultimately reflects where AI data center infrastructure is heading: higher density, tighter power-cooling integration and greater reliance on standardized deployment packages. The company is not simply adding capacity to its electrical portfolio; it is packaging facility infrastructure around the operating characteristics of next-generation GPU systems. With GB300 NVL72 deployments demanding liquid cooling and increasingly dense power architectures, that packaging can reduce the engineering gap between a GPU platform and the facility built to run it. At the same time, the expansion into Europe first and North America later gives Schneider Electric a staged path for scaling the product across the AI infrastructure market.