Scaling for AI: The Rise of Modular Data Centers

As AI workloads continue to surge, organizations are struggling to keep up with rising demands for power, cooling and compute capacity in their data centers. Generative AI (GenAI), in particular, requires dense, high-capacity infrastructure with extreme power and cooling needs, especially when it comes to large language models (LLMs).  Organizations operating aging data centers or facilities in repurposed buildings face even greater challenges. 

AI already uses up to 20% of global data center power demand, according to a recent study published by Elsevier, Inc.  And it will only get worse as organizations increase their adoption of GenAI. IDC predicts that AI data center capacity will have a compound annual growth rate (CAGR) of 40.5% through 2027, and AI data center energy consumption will grow at 44.7% over the same period, reaching 146.2 Terawatt hours by 2028.

 These extreme AI-driven demands have caused a strategic shift in how organizations approach infrastructure, location, and power. Organizations will need to scale for higher power density, assessing building constraints in terms of power, square footage, and space optimization to quickly find ways to meet power demands.

Modular data centers are a powerful solution for a new era of computing

Modular designs offer a scalable, efficient way to expand infrastructure quickly and meet modern AI demands. 

 Modular data centers are essentially data centers in a box: portable, scalable facilities that combine power, cooling, networking, and storage into a containerized solution that is pre-manufactured, pre-engineered and pre-tested. They are designed to help you scale faster and far more easily than building a new data center or even retrofitting an existing one. 

 AI-capable modular data centers are ultrahigh density and have "liquid to the chip" cooling built in, managing liquid down to the racks.

 To meet this demand, Schneider Electric offers standardized, pre-engineered modular data center solutions that integrate uninterrupted power supply (UPS), cooling, racks, and management software in factory-assembled, weatherproof structures. To facilitate speed-to-market, scalability, and cost-efficiency, they come in four designs: All-in-One, Single, Dual, and Multi-Bay Configurations, all of which are purpose-built for maximum space and aesthetics. These containerized modules are easy to transport internationally and include ruggedized applications. Schneider also offers modular rooms, which are field-assembled and combine the benefits of modularity and traditional IT space.

 What's more, Schneider partners with WWT to integrate IT stacks and software onto the empty racks Schneider builds. 

Benefits of a modular data center approach 

Modular data centers offer many significant benefits:

• Time. Building or retrofitting a data center is extremely complex, expensive, and time-consuming. Modular data centers can be deployed quickly to add needed data center capacity, as opposed to a traditional construction.
• Scalability. When built for AI, modular data centers give organizations the ability to quickly and efficiently scale further and faster than retrofitting traditional data centers.
• Modernization. It's far easier to modernize as a whole with a modular data center than to retrofit an existing complex data center piece by piece.
• Efficiency and simplicity. Schneider offers four different modular data centers: 120 kilowatt, 70 kilowatt, 40 kilowatt and 20 kilowatt, each entire data center on one SKU as one part in addition to their customized ETO solutions for hyperscale applications.

Use cases for modular data centers 

A wide range of use cases can benefit from leveraging modular data centers for AI workloads. These include:

• Organizations housed in buildings that were not designed to support data centers, such as an older hospital, school, manufacturing facility, or a small corporate headquarters.
• Hospitals that don't want to give up space for income-producing beds in exchange for a data center. Modular data centers can be set up right outside.
• Pharmaceutical companies that need to spin up research quickly. For example, an epidemic makes it necessary for speedy vaccine development.
• 911 centers that can't wait to scale.
• Federal agencies that need agile, mobile, tactical data centers, perhaps to move with troops in a sandy location or for disaster recovery. Schneider offers custom-built modular data centers for the DOD that are often ballistic-rated and RMI/RFI shielded, which means that they have a Faraday cage on the inside, so communications are not usurped by enemy intelligence.
• Service providers/hyperscalers—Large "aaS" data centers rely heavily on modular data centers to scale quickly and serve their customers' needs.

4 ways to successfully leverage modular data centers for AI workloads

1. Find a single point of contact to simplify and accelerate the process of deploying a modular data center.

Together, WWT and Schneider Electric provide a useful single point of contact, bringing the best of breed from both facilities and IT into a single solution. Schneider builds the infrastructure as empty racks, and WWT integrates the IT stack and software onto the racks and sends it on to the client, either already in the container or to be installed into the container at the customer location. 

2. Engage as early as possible 

When you realize a modular data center may meet your needs, don't wait to get the process started.  Timelines can be long for certain parts of the process, such as sourcing certain types of electrical equipment, which could take a year or more. Prioritize early identification and proactive resolution of potential obstacles to optimize project velocity and minimize delays. Resist the urge to over-customize.

The beauty of a modular data center is its speed and simplicity. The more you stay with the standard design, the easier it will be to continue to scale, and the quicker the process will be. If you move a rack or a door, it will cost more, take longer, and inject complexity.

3. Consider facilities and utilities from the beginning.

A common mistake that organizations make is neglecting to factor in power and other facility issues until it is too late. They have already designed and purchased a technology solution, but the facility can't fit it or provide the needed power or cooling.

4. Make sure all the right people are involved (IT, Facilities) before making key decisions.

Another pitfall to avoid is making big decisions without all the key players. IT and Facilities have very different perspectives that both need to be considered throughout the process. Effective business and technical decisions can only be made with experts from both angles.

About the Authors:

Mike Parham is a Technical Solutions Architect at World Wide Technology, where he has been delivering innovative IT solutions since 2011. With a strong technical background and a focus on aligning technology with business goals, Mike helps clients design and implement scalable, efficient architectures. He is known for his collaborative approach, deep expertise, and commitment to driving successful outcomes.

Thomas Humphrey is the Business Development Manager - Modular Data Centers for Schneider Electric North America. His current role includes developing strategy and thought leadership, driving revenue and orders growth, and new customer acquisition. Thomas has been with Schneider Electric for over 22 years and has worked in many facets of the data center business including technical, sales, and managerial roles primarily focused on mission-critical power and cooling solutions for data centers.