Scott Data Embraces AI Transformation to Offer GPU as a Service

Located in Omaha, Nebraska, Scott Data (SD) is one of about 20 Tier III, multi-tenant data centers operating in the U.S. The company's 110,000 square feet of facilities are designed to meet the advanced data storage and computing needs of a global customer base that spans highly regulated industries such as healthcare, financial services, big data analytics and the media.

Until recently, SD chiefly operated as a traditional colocation facility that customers leveraged as a primary data center site or for disaster recovery backups. However, following an industry conference where he learned that most data centers were unable to support large AI workloads due to the densities of power and cooling required to support the hardware, SD CEO Ken Morano recognized a unique opportunity for innovation.

- Ken Moreano, co-founder, President & CEO of Scott Data

GPU as a Service

Why would an industry leader in colocation data center services like SD pivot to GPU as a Service (GaaS or GPUaaS)? GaaS is an emerging cloud-based offering that gives organizations on-demand access to the powerful graphics processing units (GPUs) increasingly required for accelerated computational tasks. It enables organizations to leverage high-performance computing (HPC) and high-performance architecture (HPA) resources to train AI models without the need to invest in physical hardware — a common roadblock for those just starting the AI journey.

Unlike most data centers, SD was in a perfect position to pivot to GaaS as it could already supply the power and cooling needed to support large AI workloads. In fact, SD had been ready for something like generative AI (GenAI) since 2011 when it repurposed its facilities with a 20-megawatt central power plant capable of delivering 60 kilowatts (kW) per rack.

While the vast majority of SD's customers had expressed interest in AI, few wanted to make the upfront capital investment or navigate the lifecycle management issues that come with procuring their own servers and chips. This is understandable given the complexity of modern AI solutions, many of which require some level of specialized integration for the new hardware, software, cooling and power.

By leveraging its electrical power and cooling advantages, SD envisioned a new revenue stream in which customers could easily access on-demand or long-term leased GPU resources to accelerate AI research and development on their own terms.

Having already invested in NVIDIA DGX™ H100 infrastructure and NVIDIA AI Enterprise, SD was ready to begin transformation.

Challenge

Though SD had an ambitious vision for transformation in hand, it needed help understanding the art of the possible when it came to refining and executing its strategic GaaS vision.

 - Ken Moreano, co-founder, President & CEO of Scott Data

To maximize its mechanical and electrical density capabilities, SD needed a custom architectural design that would accommodate the many complexities of running large AI workloads, including:

• High-density rack support: High-density racks have become the new standard for AI applications. Depending on the physical data center footprint, these racks may need to be custom-built.
• Airflow management: Proper airflow allows heat to dissipate, which is crucial for preventing hotspots and maintaining uniform temperature distribution.
• Cooling capacity and infrastructure: Traditional rack cooling may not provide enough capacity to handle the heat loads of densely racked GPUs. Moreover, data center cooling solutions often require complex infrastructure, such as plumbing for liquid cooling systems that are integrated with custom airflow management solutions.
• Flexible power architecture: Power distribution systems must be engineered to support the elevated energy requirements of next-generation AI hardware, ensuring scalability for future growth.
• Potential for heat recycling: The facility's design should consider the possibility of capturing and repurposing waste heat, aligning with sustainable data center practices.
• Expertise: Data center operators need the right skills to manage and operate today's HPA environments.

These were some of the challenges that SD engaged WWT to help solve.

Solution

SD engaged WWT for AI consulting and to facilitate a proof of concept for the design and installation of a GaaS design featuring technology from NVIDIA and Supermicro at its core. Experts from WWT's facilities infrastructure and high-performance architecture (HPA) teams worked closely with SD leadership to understand their vision and goals.

Using lessons learned from a decade of delivering advanced AI/ML solutions for our clients, we helped SD address the design unique challenges of building a GaaS-ready data center.

- Ken Moreano, co-founder, President & CEO of Scott Data

In addition to working closely with strategic partner NVIDIA, WWT brought in APC by Schneider Electric, who proposed a power solution using pre-configured modular whips and rows of power distribution. WWT engineers then designed a custom rack configuration for the NVIDIA DGX H100 systems. This entailed collaborating with another trusted partner, Motivair, on designing a suitable rear-door heat exchanger that could handle the extreme heat output of SD's high-density racks.

For implementation, due to the large size of the custom rack compared to the entryway measurements of SD's physical data center in Omaha, WWT oversaw an on-site proof-of-concept build with the help of Infinite Networks, another strategic partner who provided critical cabling and logical configuration services for the install.

Outcomes

SD's GaaS proof-of-concept was a huge success, marking a key milestone in the company's transformation from leader in traditional colocation services to leader in GPU as a Service.

SD clients are one step closer to harnessing the power of accelerated computing resources and supporting AI infrastructure on demand, eliminating the burdensome need for upfront investment in AI hardware, software, power, cooling and expertise. This pivot should accelerate the ability of SD clients to develop and deploy their own AI applications while SD expertly handles the complexities of maintaining and optimizing the AI-ready environment.

SD expects its GaaS offering will be popular with customers across a variety of verticals and use cases such as:

• Agriculture: Using computer vision and machine learning (ML) models trained on GPUs to analyze images of livestock for potential health issues.
• Healthcare: Using ML models to accelerate medical image analysis and genomic research.
• Financial Services: Using AI applications for real-time fraud detection and investment analysis.

By recognizing and seizing this opportunity for innovation, SD has only reinforced its position as an industry frontrunner dedicated to efficiently supporting its customers' most demanding AI workloads.

SD is now focused on scaling its AI infrastructure investments and capabilities with WWT's guidance. A second phase of this project is planned to build out high-density racks for complementary high-performance technologies and architectures, to support a wider variety of customer use cases, requirements and AI goals.