How Compute and Storage Infrastructure Support Federal AI Adoption

Government officials must realize the essential elements for scaling artificial intelligence solutions from the start.

Adam Stone writes on technology trends from Annapolis, Md., with a focus on government IT, military and first-responder technologies.

In government, there’s a lot more to delivering on artificial intelligence initiatives than simply plugging in a graphics processing unit and going to work.

“How are you training those models? How are you getting those models distributed out? How are you maintaining those models?” asks Ken Rollins, chief AI technology strategist at Dell Federal.

A misstep here can derail an AI pilot — and often does. In federal agencies, there’s an urgent need to scale up AI efforts, and that requires a new strategic approach, one that ensures alignment of data and infrastructure.

Face Challenges to Scaling AI for Federal Agencies

Infrastructure is key to supporting scalable AI. In defense agencies, for example, that AI application may need to work with data collected at the tactical edge.

“When you’re dealing with huge sensor feeds, you’re not able to bring that all the way back to a central location; you have to process it locally,” Rollins says. “But heavy AI processing out in the field is really hard because of size, weight and power constraints.”

DISCOVER: Dell and CDW can support your agency’s technology advancements.

When data and infrastructure in an AI pilot are out of alignment, “many of those AI projects never fully make it to production,” he says. And if they do go live, the applications often can’t scale. The initial setup will prove insufficient to the need over time.

A recent survey by Dell found that only 33% of federal civilian agencies have embedded AI into multiple workflows agencywide, and 81% admit they are more likely to launch a new AI pilot than scale an existing one. Some 41% cite integration hurdles as the reason AI projects fail.

To maximize the value of their AI investments, IT leaders need to rethink their strategy.

Adopt a New Mindset for AI Problems

To ensure an AI-driven application can deliver in actual operational scenarios, “you need to really understand the solution set, and then make sure you’re buying systems that are scalable,” Rollins says. “You may have thought about five or 10 users, but when 100 people are using it, that gets more complicated.”

This requires not a tech fix, but a mindset shift. Knowing that compute and storage needs are inevitably going to expand, IT leaders need to be thinking about scalable infrastructure from the start, “as opposed to having to go through that hardware procurement cycle again,” he says.

Click the banner below to learn how to build a frictionless enterprise with AI.

Even in the sandbox stage of an AI project, IT leaders need to be looking ahead to ensure their data and infrastructure are aligned in such a way as to support future expansion.

“You want to understand what model you’re going to run, what your data flows look like,” he says. “That means thinking about the model, the data, the infrastructure and the ecosystem.”

Because there is an ecosystem: Beyond just the model, there are tool calls and memory subsystems. “You’re almost creating an AI operating system,” Rollins says. To ensure an AI pilot can scale, IT leaders need to think about that big picture, right from the start.

How Dell Helps Federal Officials Achieve AI Goals

Dell’s Infrastructure Services Group can help federal agencies realize that vision. “Before day zero, we can help define that initial use case, addressing the data challenges part. We will help to actually design the system, to stand it up and to operationalize it,” Rollins says.

When it comes to building the flexible infrastructure that ensures scalability, Dell can deliver large systems and help to cluster them, working with partners such as NVIDIA to combine multiple GPUs to create an inferencing powerhouse.

LEARN MORE: How to empower your workforce with AI solutions.

From rack-scale systems to ruggedized servers, “we can stand up an AI data platform relatively quickly and help them migrate onto that,” he says. Along with NVIDIA, Dell brings to the table partners such as Intel and AMD, “companies that understand the federal space.”

Agencies can start now to move in this direction. The first step, Rollins says: “You have to understand that use case and define where that data lives.”

From there, assemble the stakeholders. “You need the mission owner, the head data officer, security architects, infrastructure folks and that end user, who will have some very valuable domain knowledge,” he adds.

From there, it makes sense to run a pilot, not only to see if the AI will work but also to understand what kind of infrastructure will be needed to scale it over time. “Building just a rough version will teach you what’s true about your data and the models,” he says.

Then, agencies can build on what they’ve learned, aligning the data and the infrastructure to ensure the AI application can deliver even as the mission and the AI capabilities expand over time.

Brought to you by: