Representative End User Clients
Representative Automation Clients
Representative Software Clients
Designing, powering, and operating the next generation of AI data centers.
From Data Centers to AI Factories
In a recent analyst briefing, Schneider Electric, with participation from AVEVA, outlined its approach to designing, building, and operating what it refers to as “AI factories”—high-density data centers optimized for AI workloads.
The discussion highlighted how infrastructure requirements are shifting as AI scales. Rather than incremental improvements to traditional data centers, the focus is moving toward tightly integrated systems spanning power, cooling, simulation, and operations across the full lifecycle.
Schneider Electric frames this lifecycle across five stages—design, simulate, build, operate, and maintain—with increasing emphasis on continuity between them.
Reference Designs Reduce Deployment Risk at Scale
A central theme was the role of engineering-grade reference designs in supporting deployment of next-generation AI platforms.
Schneider Electric develops detailed designs—including electrical one-lines, piping systems, and controls—to help operators deploy high-density AI infrastructure with lower risk. A new reference design aligned with NVIDIA’s Vera Rubin platform was introduced ahead of broader rollout.
As rack densities increase—reaching approximately 227 kW per rack—engineering precision becomes more critical. Failures can propagate quickly, and the margin for error is significantly reduced compared to traditional environments.
At the facility level, this is also changing how space is allocated:
A smaller proportion of space is dedicated to IT racks.
More space is allocated to supporting infrastructure (“gray space”).
Liquid cooling is becoming dominant, with the majority of heat removed through liquid systems.
800 VDC: A Response to Density Constraints
A major focus of the briefing was the shift toward 800 VDC power architectures.
This transition is being driven primarily by increasing rack density rather than incremental efficiency gains. At higher power levels, traditional approaches—such as 48/54V distribution—become physically difficult to scale due to space constraints and conductor requirements.
Two key changes are emerging:
Moving AC-to-DC conversion out of the compute rack.
Delivering power at higher voltages to reduce current and cabling complexity.
Two architectural directions were discussed:
Distributed (“sidecar”) architectures, where conversion is handled in an adjacent power rack.
Centralized DC architectures, where conversion is moved further upstream to maximize compute space.
Schneider Electric emphasized that there is no single optimal architecture. Instead, deployments will vary depending on:
Greenfield vs. retrofit scenarios.
Availability and redundancy requirements.
Protection and selectivity strategies.
Supply chain readiness.
Maintainability considerations.
The broader ecosystem around 800 VDC components, including overlap with electric vehicle supply chains, is also contributing to momentum.
Digital Twins Expand Beyond Design
The role of digital twins and simulation is also expanding significantly.
Schneider Electric and AVEVA discussed efforts to enable a shared simulation environment based on OpenUSD and Omniverse, with the goal of creating reusable, simulation-ready models rather than bespoke builds for each facility.
This approach enables:
Multi-domain simulation (electrical, thermal, airflow) within a unified environment.
Consistent interpretation of assets through shared ontologies and semantic mapping.
Integration of IT, OT, and energy data into a common framework.
The longer-term direction is toward continuous simulation, where operational data feeds back into the digital twin to support ongoing optimization and scenario analysis.
Agentic AI Supports Operations and Maintenance
Operational complexity is increasing as AI infrastructure scales, particularly given the volume of data and alarms generated by modern systems.
Schneider Electric highlighted its use of agentic AI within its services organization to address these challenges and help manage industry-wide skill shortages.
These systems support:
Alarm interpretation and root-cause identification.
Guided troubleshooting workflows.
Work-order creation and customer communication.
Smarter dispatch decisions for field technicians.
Rather than replacing human expertise, these systems are designed to augment it—improving efficiency and enabling more targeted interventions.
Conclusion: Toward Integrated AI Infrastructure Systems
The briefing reflects a broader shift in how AI infrastructure is being designed and operated.
Schneider Electric’s approach—working with partners including AVEVA and collaborating closely with NVIDIA technologies—highlights a move toward integrated systems that combine:
Reference-driven design.
New power distribution architectures.
Lifecycle digital twins.
AI-enabled operations.
As AI workloads continue to scale, the ability to coordinate these elements as a unified system will become increasingly important for reliable and efficient deployment of AI factories.
Follow ARC Advisory Group for the latest trends in sustainability, automation, and next generation technology.