AI Data Center Reference Architecture

The rapid growth of AI is creating unprecedented power density and infrastructure requirements. Many data centers are evaluating new architectures involving advanced cooling, battery energy storage, grid integration, and alternative power distribution strategies. However, navigating these technologies while balancing deployment speed, power availability, efficiency, and reliability can be complex.

MEPPI developed these reference designs to provide customers with a practical roadmap for building next-generation AI infrastructure. The designs combine proven technologies and scalable architectures to help organizations accelerate decision-making, reduce design risk, and prepare for future AI workloads.

Key areas of focus include:

1. Power Availability – Addressing growing utility constraints through solutions such as battery energy storage systems (BESS), onsite generation integration, optimized power architectures, and ability to connect to the grid in the future.
2. Time to Market – Providing validated design frameworks that help customers streamline planning and deployment for multiple current and future GPU generations.
3. Efficiency – Enabling more efficient power delivery and advanced cooling strategies to support increasing rack densities.
4. Reliability and Safety – Incorporating resilient architectures and proven technologies designed to support mission-critical operations.

While these AI reference architecture designs were initially developed with North American market requirements in mind, many of the underlying technologies and principles can be applied globally. Regional adaptations may be required to address local regulations, standards, and utility requirements.

Customers worldwide can submit inquiries and MEPPI will coordinate support through Mitsubishi Electric’s global network, including Mitsubishi Electric Corporation, regional group companies, and qualified partners where appropriate.

• Hyperscale data center operators
• Colocation providers
• AI infrastructure developers and neocloud providers
• Utilities and energy developers supporting AI campuses
• Engineering firms and infrastructure partners

MEPPI supports customers as both a technology provider and infrastructure advisor, helping organizations evaluate, plan, and implement AI-ready power and cooling infrastructure solutions.

Technology Provider

MEPPI delivers a broad portfolio of critical infrastructure solutions represented within the reference designs, including:

• Uninterruptible Power Supply (UPS) systems
• IT Cooling Equipment, such as chillers, thermal walls, direct expansion units, and CDUs
• Switchgear and electrical distribution equipment
• High voltage and medium voltage circuit breakers
• STATCOM
• Transformers
• DCIM
• Batteries and other energy storage systems
• Energy management systems
• Optical communication
• Additional power and thermal infrastructure technologies from Mitsubishi Electric and affiliated companies

MEPPI works closely with Mitsubishi Electric Group companies and qualified ecosystem partners to help customers align technology selections with their performance, scalability, efficiency, and reliability objectives. While our portfolio’s coverage of solutions in the designs is extensive, we depend on third party organizations so we can coordinate full system builds.

Explore our product portfolio https://www.meppi.com/ 

Infrastructure Advisor

MEPPI helps customers understand and adapt reference designs to their specific site, operational, and growth requirements through:

• Reference design consultation and architecture guidance
• Power and cooling infrastructure planning
• Utility interconnection engineering and grid integration services
• Power system studies and electrical infrastructure planning
• Coordination across Mitsubishi Electric Group companies and partner organizations to support integrated infrastructure strategies

Whether customers are evaluating future AI infrastructure concepts, adapting designs for a specific deployment, or planning long-term capacity growth, MEPPI can help assess requirements and identify appropriate technology and engineering resources.

Yes. The data center reference architecture is included as a starting point rather than a fixed solution. MEPPI can work with customers to evaluate:

1. Site-specific requirements
2. Utility constraints
3. Cooling strategies
4. GPU roadmap assumptions
5. Scalability requirements

The reference designs support AI environments ranging from approximately 200 kW to more than 1 MW per rack for future-generation architectures. This scalable approach reduces redesign work as future GPU generations are introduced.

MEPPI’s Chip-to-Grid DSX Reference Designs were developed to support AI infrastructure associated with current and future NVIDIA GPU platforms, including the following rack configurations:

1. MGX 1.2 (supporting NVIDIA Vera Rubin NVL72 GPUs)
2. MGX 1.3 (supporting NVIDIA Vera Rubin Ultra GPUs)
3. Future High-Density DC Platforms (supporting future NVIDIA GPUs)

The reference designs are intended to help data centers plan infrastructure that can support increasing rack densities and evolving power and cooling requirements across multiple GPU generations.

Yes. The long-term vision is to help data centers optimize infrastructure from energy source to AI workload. The reference designs are intended to evolve in conjunction with:

1. Future GPU generations
2. New power and cooling technologies
3. Customer requirements

MEPPI expects to periodically update the designs as technology and market needs change.