GovExec TV: Five questions with HPE’s Jeff Lush and NVIDIA’s Ryan Simpson

As purpose-built infrastructure, AI factories offer government a fast and efficient means to ramp up AI-driven applications. Building and integrating AI infrastructure from scratch can be time-consuming. With an AI factory, agencies can cut down significantly on time-to-value and time-to-first-token, while supporting data security requirements.

To better understand AI factories, GovExec TV spoke with the experts: Jeff Lush, chief technologist for U.S. federal at Hewlett Packard Enterprise (HPE) and Ryan Simpson, chief technologist for federal partners at NVIDIA.

AI factories and security baselines

An AI factory isn’t just GPUs and racks. It’s security controls, sovereignty boundaries and observability tools that deliver reliable outputs. By analogy, a conventional factory isn’t just about machines.

“It’s about the infrastructure. It’s about the power. It’s about the people operating the machines. It’s how you pass one item from one machine to the next machine, like an assembly line,” Simpson said. “AI factories represent that.”

Beyond any one component, the true power of an AI factory comes from how each part of the system is assembled, co-designed and unified into one powerful production engine to help agencies efficiently develop and scale AI capabilities — with integrated security controls.

“There has to be security at all levels of the operation, from the silicon of the server that’s booting, all the way up through the storage and the network” — everything is bound by security controls, Lush said. “We use [NIST] as a framework … security provides accountability and structure.”

[subhead] Portability and sovereignty

Simpson compared AI factories to supercomputers of the past: A means of bringing to bear new levels of speed and efficiency. But unlike those machines, an HPE AI Factory offers portability — important to agencies managing evolving AI workloads and unpredictable environments.

“We can literally have one of these factories sitting on a data room floor, and when an emergency sets off, they scoop up the equipment, put it in some sort of modular or transportable type of container, and boom!” Lush said. “Within hours that AI is doing its job wherever it’s needed.”

At the same time, an AI factory can address the need for sovereignty in AI deployments, especially for workloads involving sensitive data. In particular, a sovereign AI factory delivers inferencing locally, where enterprise data resides. It’s focused on a specific mission, with all the needed controls in place.

Sovereignty here is “not just about what data you have access to, but what rules you want to operate in,” Simpson said. With a sovereign AI factory, agencies have “the controllability and the ownership of that system.”

Managing data, delivering accountability

As AI continues to scale, agencies increasingly must manage a deluge of data. Robust cybersecurity demands accountability and traceability of that data: Knowing which applications are using what data, when and why.

Agencies can look to tools like the NVIDIA Morpheus cybersecurity AI framework for digital fingerprinting workflows. “Morpheus can help with profiling and digital fingerprinting,” a behavior-based anomaly modeling system, Simpson said. “Now we’re extending digital fingerprints to not just the user and the computer, but to different agents. Agents need to have their own access and they might be accessing applications on my behalf.”

HPE Morpheus Enterprise Software and HPE OpsRamp support hybrid-cloud operations, infrastructure observability, and remediation workflows. With OpsRamp “we observe what’s going on, and then the next step is: How do we remediate what’s going on?” Lush said. “There are decisions that could be made through workflows that we can enable via Morpheus.”

These protections can take various forms. For example: If data goes back into the GPU unencrypted, NVIDIA Confidential Computing helps protect sensitive data and AI models while they are in use on supported NVIDIA GPUs.

All this will help to build confidence as AI operations become increasingly autonomous. With AI factories delivering both speed and consistency, trust will increase. Humans will need to be vigilant, but perhaps after verifying the first 1,000 iterations of a process, “maybe we’re going to trust it now to do #1001 by itself,” Lush said. Over time, “our comfort levels will increase.”

As AI moves from pilots to mission-critical deployments, a common thread is emerging: The technology is only as trustworthy as the systems built around it.

For government missions where protection of sensitive data is essential, an AI factory’s combination of speed, security and earned trust can accelerate deployments while ensuring agencies meet the need for strong compliance and accountability.

To learn more about the benefits of the HPE AI Factory with NVIDIA portfolio, watch the full episode above and visit our HPE-NVIDIA AI Factory content hub.