AI Reference Designs to Enable Adoption:

A Collaboration Between Schneider Electric and NVIDIA

A Data Center Research and Strategy Report Version 3 - September 2025 Steven Carlini and Wendy Torell

LET’S GET STARTEDRATE THIS REPORT

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 12 se.com

Schneider Electric collaborates with NVIDIA, co-developing reference

designs for AI factories and data centers, in advance of when NVIDIA releases new

and more powerful GPUs and SuperPODs.

AI is moving from model development to working

models of real applications that are fundamentally

changing how industries operate, and consumers

interact.

Advancing these models to meet the demands of the future requires immense

computational power, relying on specialized accelerators.

Increasingly, more capable models are driving measurable shifts in productivity, decision- making, and customer engagement across sectors.

NVIDIA’s GPU-based compute remains the dominant accelerator type for training and deployment. NVIDIA’s strategy has evolved from GPU component supplier to a comprehensive supercomputing and software stack vendor. They now offer everything from server boards and servers to powerful SuperPODs (AI clusters). This integrated approach aims to fast-track AI deployment with lower risk.

These designs provide critical data center guidance and technical detail to enable, streamline and deploy these demanding, high density AI server clusters. Designs are available for retrofit and purpose-built scenarios.

Traditional data center power, cooling, and racks are insufficient.

K EY

T A

K EA

W AY

S

However, deploying these high-density, GPU-powered

AI clusters presents infrastructure challenges to

formerly accepted practices.

https://www.surveymonkey.com/r/GPSLF59 https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 13 se.com

For data center operators looking to deploy one or many 1.8 MW high-density AI clusters with a maximum rack density of 73 kW, using purpose-built heat rejection optimized for efficiency. This is demonstrated as one IT hall in reference design 99 (IEC) and reference design 100 (ANSI).

For data center operators looking to develop new purpose-built data centers, we have three designs:

Several reference designs are available, adhering to both IEC standards and ANSI standards. These designs support the following three retrofit data center scenarios:

AI Infrastructure and liquid-cooled controls design supplements the latest NVIDIA GB200 and GB300 NVL72-based cluster designs, providing a critical framework for intelligent control systems that manage complex AI infrastructure. This design focuses on seamless interoperability between building and cluster management software, establishes redundant systems for power and cooling, and introduces new guidance for measuring AI rack power profiles. Ultimately, this Controls Reference Design (CRD1) ensures the reliability and peak performance of AI deployments by enabling precise, real-time management of critical power and cooling resources.

For the data center operator looking to add a ~1 MW AI cluster(s) without implementing liquid-cooled servers. This design uses air-cooled servers and can accommodate 40 kW per IT rack. However, the cluster footprint is larger than the liquid-cooled versions.

For data center operators looking to add ~1 MW liquid-cooled AI clusters and remove heat using their existing (air-cooled) cooling system. This design can accommodate 73 kW per IT rack, balancing floor space savings with density.

For data center operators looking to add ~1 MW liquid-cooled AI clusters and remove heat by tapping into pipes of an existing chilled water-cooling system. This design can accommodate 73 kW per IT rack while minimizing floor space.

For data center operators looking to deploy one or many 7.4 MW high-density AI factories with NVIDIA GB200 NVL72-based cluster with a maximum rack density of 132 kW, using purpose-built heat rejection optimized for efficiency. This is demonstrated in reference design 108 (IEC) and reference design 109 (ANSI).

For data center operators looking to deploy one or many 7.4 MW high-density AI factories with NVIDIA GB300 NVL72-based cluster with a maximum rack density of 142 kW, using purpose-built heat rejection optimized for efficiency. This is demonstrated in reference design 110 (IEC) and reference design 111 (ANSI).

Retrofit 1 (air cooled)

Retrofit 2 (liquid-to-air

CDUs)

Retrofit 3 (liquid-to-liquid

CDUs)

New 1.8 MW data center

New 7.4 MW data center (NVIDIA GB200 NVL72- based cluster)

New 7.5 MW data center (NVIDIA GB300 NVL72- based cluster)

https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/CRD1DS/ https://www.se.com/us/en/download/document/RD108DSR0/ https://www.se.com/us/en/download/document/RD109DSR0_EN/ https://www.se.com/us/en/download/document/RD110DSR0/ https://www.se.com/us/en/download/document/RD110DSR0/ https://www.se.com/us/en/download/document/RD111DSR0 https://www.se.com/us/en/download/document/RD111DSR0

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 14 se.com

Exponential growth carries challenges

AI is experiencing exponential growth

Foundational AI models are crucial to this AI revolution

Training complex models pushes the technical limits of traditional data center infrastructure.

Advancements in machine learning and data analysis are fueling more sophisticated algorithms, capable of complex tasks once thought to be the exclusive domain of humans. This growth is having a ripple effect across all industries.

In healthcare, AI is assisting doctors in detecting diseases, analyzing medical images, and even developing personalized treatment plans. Manufacturing is witnessing the rise of intelligent robots for precision assembly and predictive maintenance.

AI is personalizing education by tailoring learning programs to individual student needs. In the entertainment industry, it’s powering realistic special effects and creating new forms of interactive content around sports and athlete performance. The potential applications for AI’s seem limitless.

These powerful models are “AI neural networks trained on massive unlabeled datasets to handle a wide variety of jobs.”1 They enable specific spinoff models fine-tuned to accomplish a narrower, more specialized set of tasks, from language translation to medical diagnosis. The GPT-series of large language models (LLMs) serves as a highly recognizable example. Their versatility has made them a de facto standard for widespread adoption of AI. Fine-tuning these models compresses the development time and reduces cost for specific applications.

Training these massive models pushes the boundaries of GPUs. NVIDIA’s GPUs are currently the frontrunners, offering unparalleled speed and adoption within the AI community. Deploying foundational AI training models utilizes a mix of colocation companies and large, cloud- based infrastructure providers. Cluster

In the Data Center Research and Strategy’s white paper, The AI Disruption: Challenges and Guidance for Data Center Design, we highlight the limitations of traditional physical infrastructure. We then offer insights on adapting data centers for AI clusters. Traditional designs weren’t equipped to handle the sheer power capacity and rack density demands of AI training clusters. These workloads require innovative solutions for power delivery, cooling strategies like liquid cooling, and fortified IT cabinets.

size drives the choice: colocation provides a flexible option for companies who need a single cluster or a few. Large-scale operations are typically being deployed by major cloud providers.

https://www.se.com/ https://www.se.com/ww/en/download/document/SPD_WP110_EN/?ssr=true https://www.se.com/ww/en/download/document/SPD_WP110_EN/?ssr=true https://www.se.com/ww/en/download/document/SPD_WP110_EN/?ssr=true

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 15 se.com

Schneider Electric and NVIDIA are collaborating

Reference designs offer a blueprint for a reliable and validated AI deployment

By combining strengths and experience, we address the data center challenges posed by the growing complexity of AI by assembling experts from both organizations. We pulled together the IT and physical infrastructure knowledge necessary to simplify and optimize data center infrastructure for AI workloads, enabling and accelerating AI adoption across industries. And we kept energy efficiency as a foundational principle in our designs, a major component of environmental sustainability (scope 2 and scope 3).2

A reference design is a system blueprint, a list of attributes including system-level performance specifications, engineering drawings and includes a detailed list of materials or components that comprise the system. While a reference design can be directly implemented, in most cases it acts as a baseline design that is adapted to meet specific user preferences or constraints.”3

“ Reference designs aren’t new. For many years, they’ve been applied to help data center owners shorten their planning cycle and reduce downtime risks once operational. However, these designs become crucial when the physical infrastructure (i.e., power and cooling) demands different and more complex solutions than owners and operators have previously deployed.

Through this collaboration, we produced comprehensive, validated reference designs for high-density AI clusters like NVIDIA’s NVL72- based clusters, supporting rack densities up to 73 kW. We then furthered the collaboration with purpose-built AI data center reference designs with three NVIDIA GB200 NVL72-based clusters, supporting rack densities up to 132 kW and most recently, with three NVIDIA GB300 NVL72-based clusters supporting rack densities up to 142 kW. These designs provide a necessary head start to a successful, reliable deployment. They’re crucial for data center designers who lack experience architecting solutions at extreme densities, and/or have not implemented liquid cooling (which is evolving as the industry matures).

Collaboration for innovation, leveraging strengths

https://www.se.com/

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 16 se.com

Simplify implementation feasibility analysis.

Reduce planning cycle time.

Leverage expertise in efficient design.

Increase confidence that your AI cluster can be deployed reliably.

Leverage our learning on key design considerations and obstacles.The extreme rack densities of AI

workloads make many traditional designs impractical. Additionally, when retrofitting a high-density AI cluster into an existing data center, certain constraints influence design choices. A feasibility study is simplified with these designs as a starting point.

The designs include an equipment selection list with electrical and mechanical systems available on the market today. They also provide documented physical layouts and engineering drawings. This information strips out significant time from planning the project.

Reliably operating high-power, high-density AI loads is first and foremost, but the designs also follow best practices for energy efficiency to align with industry sustainability goals.

When new technologies are deployed, there is a greater risk that something will go wrong. We have iterated these designs using state-of-the-art electrical design and mechanical design software tools to ensure they are validated to operate reliably.

The designs provide best practices to overcome obstacles arising from the complexity of high density. Extreme densities and capacities in small footprints present challenges in electrical attributes, including short- circuit current, breaker coordination/ selectivity, and arc flash. Our designs solve for these realities.

New approaches to cooling are addressed: handling set points, airflow challenges, and key concerns for liquid cooling, CDUs, etc.

Alternative designs for high-density AI clusters The benefits of the resulting AI reference designs:

https://www.se.com/ https://etap.com/product/short-circuit-software https://etap.com/product/short-circuit-software https://en.wikipedia.org/wiki/Selectivity_(circuit_breakers)#%3A~%3Atext%3DSelectivity%2C%20also%20known%20as%20circuit%2Ca%20failure%20on%20the%20network https://www.se.com/us/en/download/document/SPD_VAVR-6KGRYW_EN/ https://www.se.com/us/en/work/solutions/data-centers-and-networks/reference-designs/

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 17 se.com

Retrofit scenario 1, air-cooled AI cluster

New 1.8 MW data center

Retrofit scenario 2,

liquid-cooled AI cluster with

liquid-to-air CDUs

New 7.4 MW data center

(GB200 NVL72-based

clusters)

Retrofit scenario 3:

liquid-cooled AI cluster with

liquid-to-liquid CDUs

New 7.5 MW data center

(GB300 NVL72- based

clusters)

1,050 kW AI cluster with maximum rack density of 40 kW, using traditional air cooling, with wider, contained hot aisles.

904 kW AI cluster with maximum rack density of 73 kW, using liquid-to-air coolant distribution units (CDUs) for heat rejection. This is ideal for scenarios where you cannot connect to the facility’s water systems.

904 kW AI cluster with maxi- mum rack density of 73 kW, using liquid-to-liquid CDUs for heat rejection. This scenario works when you can tap into facility water systems.

Six design scenarios to support AI training clusters, such as NVIDIA’s NVL72-based clusters, were engineered and documented, including three retrofit scenarios and three purpose-built or greenfield scenarios. In the retrofit scenarios, each design supports roughly a 1 MW AI cluster; and in the purpose-built designs, they support 1.8 MW, 7.4 MW and 7.5 MW workloads, respectively.

Here, we’ve included a high-level description of the AI clusters in each scenario as well as the physical infrastructure to support them. Table 1 and Table 2 then provide further details and a side-by-side comparison of the scenarios.

A purpose-built data center hall supporting 1,808 kW AI cluster with maximum rack density of 73 kW, using liquid-to-liquid CDUs for heat rejection, and optimized chillers for improved efficiency.

A purpose-built data center supporting 7,392 kW AI factory with maximum rack density of 132 kW, using liquid-to-liquid CDUs for heat rejection, and optimized chillers for improved efficiency.

A purpose-built data center supporting 7,536 kW AI factory with maximum rack density of 142 kW, using liquid-to-liquid CDUs for heat rejection, and optimized chillers for improved efficiency.

https://www.se.com/

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 18 se.com

Table 1 – Summary of retrofit reference designs

Retrofit design 1

Retrofit design 2

Retrofit design 3

Reference designs RD99 (IEC) & RD100 (ANSI) Scenario 1A

RD99 (IEC) & RD100 (ANSI) Scenario 1B

RD99 (IEC) & RD100 (ANSI) Scenario 1C

AI cluster IT capacity

AI server racks

1,050 kW 904 kW 904 kW

24 x 40 kW 8 x 73 kW 8 x 73 kW

AI networking racks 6 x 15 kW 8 x 40 kW 8 x 40 kW

Existing IT capacity in room 960 kW 960 kW 960 kW

Traditional IT racks 80 x 12 kW 80 x 12 kW 80 x 12 kW

Total IT capacity 2,010 kW 1,864 kW 1,864 kW

AI server rack type Standard on hard floor OCP power shelves on

hard floor OCP power shelves on

hard floor

AI server GPU cooling method Air-cooled Direct-to-chip Direct-to-chip

AI cluster heat rejection design Wide hot aisles with ducted

containment, chilled water fan walls, 30C (86F) chiller return

Wide hot aisles with ducted containment, liquid-to-air CDUs, chilled water fan walls, 30C (86F)

chiller return

Wide hot aisles with ducted containment, liquid-to-liquid

CDUs, chilled water fan walls, 30C (86F) chiller return

AI server racks power design

630A busway to 63A 400V rPDUs (IEC version)

400A modular RPPs to 100A 415V rPDUs (ANSI version)

800A busway to 63A 400V power feeds (IEC version)

400A modular RPPs to 60A 415V power feeds (ANSI version)

800A busway to 63A 400V power feeds (IEC version)

400A modular RPPs to 60A 415V power feeds (ANSI version)

https://www.se.com/ https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 19 se.com

Partners and engineers in early planning of AI clusters can obtain detailed documentation of the reference designs, including specifications, layouts and system drawings, equipment lists, and computational fluid dynamics (CFD) analysis.

While the detailed designs reference specific NVIDIA AI factories, they may be applied to other high-density AI clusters. In doing so, they should be tailored based on the IT design temperature set points, server configuration, etc.

Table 2 – Summary of purpose-built reference designs

Purpose-built design 1

Purpose-built design 2

Purpose-built design 3

Reference designs RD99 (IEC) & RD100 (ANSI)

Room 2

RD108 (IEC) & RD109 (ANSI)

AI cluster IT capacity

AI server racks

1,808 kW 7,392 kW

16 x 73 kW 48 x 132 kW

AI networking racks 16 x 40 kW 48 x 22 kW

Existing IT capacity in room 0 kW 0 kW

Traditional IT racks 0 0

Total IT capacity 1,808 kW 7,392 kW

AI server rack type OCP power shelves on hard floor

NetShelter Open Architecture Rack MGX

configured with NetShelter Open Architecture

Power Shelves

AI server GPU cooling method Direct-to-chip Direct-to-chip

AI cluster heat rejection design

Wide hot aisles with ducted containment, liquid-to-liquid CDUs, chilled water fan walls, 40C (104F) chiller return (liquid-cooled IT) and

30C (86F) chiller return (air-cooled IT)

Wide hot aisles with ducted containment, liquid-to-liquid CDUs, chilled water fan walls, 47C (117F)

chiller return (liquid-cooled IT) and 33C (91F) chiller return (air-cooled IT)

AI server racks power design

800A busway to 63A 400V power feeds (IEC version)

2,000A RPPs to 60A 415V power feeds (ANSI version)

800A busway to 63A 400V power feeds (IEC version)

800A RPPs to 60A 415V power feeds (ANSI version)

RD110 (IEC) & RD111 (ANSI)

7,536 kW

48 x 142 kW

48 x 15 kW

0 kW

0

7,536 kW

NetShelter Open Architecture MGX racks

configured with NetShelter Open Architecture

Power Shelves

Direct-to-chip

Wide hot aisles with ducted containment, liquid-to-liquid CDUs, chilled water fan walls, 47C (117F)

chiller return (liquid-cooled IT) and 33C (91F) chiller return (air-cooled IT)

800A buswayto 63A 400V power feeds (IEC version)

800A RPPs to 60A 415V power feeds (ANSI version)

https://www.se.com/ https://www.se.com/us/en/download/document/RD99DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD100DSR0_EN/?ssr=true https://www.se.com/us/en/download/document/RD108DSR0/ https://www.se.com/us/en/download/document/RD109DSR0_EN/ https://www.se.com/us/en/download/document/RD110DSR0/ https://www.se.com/us/en/download/document/RD111DSR0

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 110 se.com

Impact of future generations of AI GPUs and high-density AI clusters

NVIDIA is dedicated to staying ahead of the market. They’re continuing to advance their GPUs and AI “server boards” (which integrate GPUs and CPUs on a circuit board connected through NVLink), and providing blueprints to package them into supercomputers (SuperPODs). The most recent evolution is from the DGX GH200, the “Hopper,” to the DGX GB200 and DGX GB300, both part of the “Blackwell” platform (named after the famous scientist/mathematician). The Blackwell platform “enables organizations everywhere to build and run real-time generative AI on trillion-parameter large language models at up to 25x less cost and energy consumption than its predecessor.”4

Servers based on NVIDIA’s design include the NVIDIA GB200 NVL72 with 72 GB200 Superchips (36 CPUs and 72 GPUs, 132 kW per rack),5 and now the GB300 NVL72. The GB300 NVL72 delivers 1.5x more FP4 performance per GPU, an increase in HBM3e memory per GPU from 192GB to 288GB, and upgraded networking with ConnectX-8 supporting 1.6T optical modules (up from 800G). These systems, at the time of this writing, have multi-million-dollar price points.

Many organizations are expected to adopt Blackwell, including internet giants such as, Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla and xAI.

NVIDIA’s goal with DGX is to offer their IT reference architectures broadly to all of its partners. These reference designs represent a perfect companion piece in planning for the data center as it represents the culmination of both NVIDIA’s and Schneider Electric’s R&D investments. As NVIDIA develops the next GPUs, server boards, and clusters, their ambition is to create designs that operate within the power and cooling constraints of previous generations. However, you can expect, as new designs launch, Schneider will develop accompanying solutions and reference designs to support them.

We’re working to advance industry standards around architectural changes including higher voltage DC powering a single rack. We will be ready with new designs when these changes are ready for data center designers.

https://www.se.com/ https://www.nvidia.com/en-us/data-center/nvlink/ https://www.nvidia.com/en-us/data-center/dgx-superpod/ https://developer.nvidia.com/blog/nvidia-gb200-nvl72-delivers-trillion-parameter-llm-training-and-real-time-inference/ https://www.nvidia.com/en-us/data-center/gb300-nvl72/ https://resources.nvidia.com/en-us-dgx-platform/nvidia-dgx-platform-solution-overview-web-us?_gl=1%2A1gk5kh1%2A_gcl_au%2AMTMxMTgzMjI1NS4xNzEzMjY5MzM3 https://resources.nvidia.com/en-us-dgx-platform/nvidia-dgx-platform-solution-overview-web-us?_gl=1%2A1gk5kh1%2A_gcl_au%2AMTMxMTgzMjI1NS4xNzEzMjY5MzM3

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 111 se.com

AI infrastructure and liquid-cooled controls design

Controls Reference Design (CRD1), developed in close collaboration with NVIDIA, provides essential guidance for managing the complex infrastructure supporting AI workloads. This collaboration has been instrumental in creating a reference design that specifically addresses the critical need for robust and intelligent control systems to deliver the reliability, efficiency, and optimal performance of AI deployments. The design outlines a comprehensive strategy to integrate advanced controls across key technical areas, from power delivery to cooling, providing infrastructure that can meet the demanding requirements of AI.

This reference design focuses on several vital control aspects. It details the interoperability between Building Management Systems (BMS/ SCADA) and NVIDIA Mission Control software, enabling real-time communication to protect servers and racks. The design also includes a standardized interface for EPMS/BMS data, crucial for feeding information to digital twins, AI/ML analytics, and other enterprise systems.

Furthermore, the reference design outlines redundant control designs for cooling infrastructure, including coolant distribution units (CDUs), and for electrical distribution panels (RPPs), enabling continuous operation. The design also provides new guidance for measuring AI rack power profiles. It includes a focus on understanding peak power and power quality, which is vital for efficient resource allocation and preventing overloads.

Implementing the control strategies outlined in CRD1 will yield significant operational benefits. This proactive approach to infrastructure control minimizes downtime. It also extends the lifespan of hardware, and enhances operational efficiency, allowing our AI systems to consistently perform at peak capacity and effectively support our strategic AI initiatives.

https://www.se.com/ https://www.se.com/us/en/download/document/CRD1DS/

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 112 se.com

Next steps

Schneider Electric and NVIDIA’s collaborative reference designs add significant value for data center operators looking to add AI clusters to an existing facility or build a new one. By producing the first publicly available, comprehensive blueprints for AI data centers, and continuing that co-development over time, this partnership goes beyond optimizing infrastructure – it empowers companies to deploy AI with greater efficiency, ease, and confidence.

To get the most out of the reference designs, we recommend you:

Looking forward, this collaboration positions data center design teams and the companies they work with to deliver reliable, sustainable AI infrastructure, ensuring that future advancements can build upon a strong foundation.

Download/share the design summaries with your data center design team, and then request the relevant engineering package(s) including technical documentation and details from referencedesigns@ se.com.

Get specific about infrastructure and material specification in your design or redesign. Planning now increases confidence in later design choices.

Assess the pre-selected equipment and documented layouts to save significant time in the planning cycle.

Ask your design team to implement and or document the best practices as applied to your design. This translates to reliable deployment and reduced risk of unforeseen issues.

1 2 3 4

https://www.se.com/ https://www.se.com/ww/en/work/solutions/data-centers-and-networks/reference-designs/ https://www.se.com/ww/en/work/solutions/data-centers-and-networks/reference-designs/ mailto:mailto:referencedesigns@se.com?subject=Reference%20Design%2099 mailto:mailto:referencedesigns@se.com?subject=Reference%20Design%2099

RATE THIS REPORT

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 113 se.com

Endnotes

1 NVIDIA blog, What Are Foundation Models?, accessed on August 21, 2025 2 Schneider Electric White Paper 99, Quantifying Data Center Scope 3 GHG Emissions to Prioritize Reduction Efforts 3 Schneider Electric White Paper 147, Data Center Projects: Advantages of Using a Reference Design 4 NVIDIA, NVIDIA Blackwell platform arrives to power a new era of computing, accessed on August 21, 2025 5 NVIDIA, 8th Annual DGX User Group Meeting, accessed on August 21, 2025, Slide 32

https://www.surveymonkey.com/r/GPSLF59 https://www.se.com/ https://blogs.nvidia.com/blog/what-are-foundation-models/ https://www.se.com/ww/en/download/document/SPD_WP99_EN/ https://www.se.com/us/en/download/document/SPD_VAVR-7XMT5M_EN/ https://www.power-and-beyond.com/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing-a-9a880794793c0c2051b40d4b213e3082/ https://nvdam.widen.net/s/z797gd8qjf/dgx-user-group-meeting-spring-2024

AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

Property of Schneider Electric | Executive Report 114 se.com

THIS DOCUMENT IS TO BE CONSIDERED AS AN OPINION PAPER PRESENTING GENERAL AND NON-BINDING INFORMATION ON A PARTICULAR SUBJECT. THE ANALYSIS, HYPOTHESIS AND CONCLUSIONS PRESENTED THEREIN ARE PROVIDED AS IS WITH ALL FAULTS AND WITHOUT ANY REPRESENTATION OR WARRANTY OF ANY KIND OR NATURE, EITHER EXPRESS, IMPLIED OR OTHERWISE.

© 2025 Schneider Electric. All rights reserved.

Steven Carlini is the Chief Advocate for AI and Data Centers for Schneider Electric. Steven is responsible for developing integrated solutions and communicating the value proposition for Schneider Electric’s data center segment including enterprise and cloud data centers. A frequent speaker at industry conferences and forums, Steven is an expert on the foundation layer of data centers which includes power & power distribution, cooling, rack systems, physical security, and software solutions that improve availability and maximize performance. Steven has been responsible for guiding the direction of many industry changing products and solutions that solve real customer problems or give businesses competitive advantages. Steven holds a BS in Electrical Engineering from the University of Oklahoma, and an MBA in International Business from the CT Bauer School at the University of Houston.

Wendy Torell is a Senior Research Analyst in Schneider Electric’s Data Center Research & Strategy group bringing 30 years of data center experience. Her focus is analyzing and measuring the value of emerging technologies and trends: providing practical, best practice guidance in data center design and operation. Beyond traditional thought leadership, she championed and leads development of interactive, web-based TradeOff Tools. These calculators help clients quantify business decisions, while optimizing their availability, sustainability, and cost of their data center environments. Her deep background in availability science approaches and design practices helps clients meet their current and future data center performance objectives. She brings a wealth of experience across Schneider Electric’s broad portfolio and with the market at large. She holds a BS in Mechanical Engineering from Union College and an MBA from University of Rhode Island. Wendy is an ASQ Certified Reliability Engineer.

Authors

Steven Carlini

Wendy Torell

Vice President, Innovation and Data Center Schneider Electric LinkedIn Forbes

Senior Research Analyst Data Center Research & Strategy Schneider Electric LinkedIn

https://www.se.com/ https://www.linkedin.com/in/stevencarlini/ https://councils.forbes.com/profile/Steven-Carlini-Chief-Advocate-Data-Centers-AI-Energy-Management-Business-Unit-Schneider-Electric-Sch/9166354b-b7fd-4357-a2b5-2b807bccb424 https://www.linkedin.com/in/wendytorell