Modular, mobile data centers bring powerful AI capabilities to the mission’s edge

As federal agencies accelerate and scale AI adoption, they face critical challenges: Sophisticated AI processing requires massive compute power and advanced cooling, both of which can quickly incur skyrocketing costs. Furthermore, government agencies operate across disparate locations and environments, and some models need to be deployed on-site to meet performance and control demands.

Legacy infrastructure and traditional data centers often fall short in supporting these requirements. “They don't have the power and cooling to support the compute, the processing that is needed in large areas and the capability for being mobile,” said Tracy Mills, federal distinguished chief technologist at HPE.

Bridging the gap between operational urgency and infrastructure limitations calls for an entirely new approach to data center architecture. By transforming massive compute power into an agile, mobile asset, agencies can deliver the full power of an AI factory wherever the mission demands.

Fixed, distant facilities slow mission decision-making

When it comes to certain workloads and models, proximity matters. Agency leaders relying on AI for real-time decision-making need the speed and low latency that being close to a data center provides. In this scenario, investing capital to build new traditional, permanent data centers may not offer the expected return on investment.

“In the amount of time it takes for them to build a data center, to break ground they’re getting further and further behind the ball,” Mills said. “But if they can pivot and basically be given a data center in a box, get that certified and online and ready to go, it moves them a lot closer to the goalpost.”

Beyond construction delays, large and fixed facilities pose security risks. Large-scale centers are emerging as significant targets for threat actors, and they aren’t difficult to identify.

“It takes time to lay foundations and build buildings, and it’s not a secret,” said Jeff Lush, U.S. federal chief technologist at HPE. “We don’t have a big umbrella over those sites, so when they come up, they already have a target on them.” 

By transforming that data center into a modular unit, agencies can access the same power and capacity within a discreet and mobile footprint.

Delivering full-scale AI factories to the mission edge

The demand for proximity, computing power and flexibility in a small package is the driving force behind the HPE AI Mod POD (modular data center). Engineered to support complex HPC and AI workloads, the unit can be deployed almost anywhere to deliver customized computing power directly to users.

“We’ve seen some procurements already where they're requesting deployable data centers that can be rolled in and then rolled out when they need to,” said Trent Boyer, North America OEM sales director at NVIDIA. “HPE and NVIDIA have been working together to where we can essentially deploy a data center in a modular container that has all the power and cooling needed, and they can essentially bring in a generator, hook that thing up and deploy a full AI factory into any part of the world.”

“Picture the back of a semi,” added Lush. “You can literally put it in the back of a transport aircraft and move it to a different place in hours.”

Beyond speed and mobility, the HPE AI Mod POD provides high-density computing, scalability, optimized cooling, power efficiency and rapid deployment. Agencies can train and deploy advanced models on-premises using the power of the HPE AI Factory with NVIDIA, achieving enterprise-grade processing at significantly reduced costs to budget and time over traditional fixed data centers.

Key capabilities and metrics

• Direct liquid cooling (DLC): HPE is the only vendor that offers DLC technology for modular data centers, enabling units to support high GPU densities and high-wattage processors without thermal throttling.
• Cost efficiency: The AI Mod POD runs approximately 30% less expensive in terms of $/kW IT compared to equivalent fixed data centers for the same total capacity and tier level. 
• Sustainability: Cooling systems and energy-efficient designs typically reduce the infrastructure carbon footprint by 30-40%
• Rapid deployment: The pre-configured models can be tailored to specific requirements and delivered within a few months. 

As AI requirements evolve, agencies can no longer afford to let legacy infrastructure timelines impact mission timelines. Modular data centers reduce tradeoffs between processing power, deployment speed, flexibility and security by packing high-powered compute into a mobile footprint. HPE AI Mod POD configurations that incorporate NVIDIA accelerated computing can help provide federal agencies with a flexible path to deploy AI closer to where data is generated.

Learn more about the power and agility HPE and NVIDIA can bring to your most advanced AI workloads.