Jane Edwards 
- Booz Allen SVP Graham Gilmer has discussed agencies’ shift to hybrid AI infrastructure
- GPU advances are enabling AI deployment in edge environments
- The 2026 FedCiv Summit will cover AI and cloud modernization
Graham Gilmer, a senior vice president at Booz Allen Hamilton, said federal agencies are adopting a hybrid infrastructure model as artificial intelligence moves from experimentation into operational use.
The infrastructure, data and AI delivery challenges Gilmer discussed continue to shape modernization priorities across the federal government. At the Potomac Officers Club’s 2026 FedCiv Summit on Oct. 29, executives and agency officials will examine how civilian agencies are scaling AI, modernizing cloud and compute infrastructure, advancing cybersecurity initiatives and shaping procurement priorities across government missions. Sign up now!
Speaking during a panel discussion for Federal News Network’s “Delivering the tech that delivers for government” series, Gilmer said agencies and systems integrators are designing environments that combine cloud, on-prem and edge systems to support AI workloads across mission settings.
“Depending on the use case, you may want to go with one or both or a hybrid combination for redundancy,” Gilmer said.
Why Are Agencies Moving Toward Hybrid AI Architectures?
Gilmer said federal environments increasingly require systems that can operate across classification levels, geographies and disconnected environments, driving agencies toward distributed architectures.
“We are constantly monitoring the gap between commercial — what’s available on your phone or outside of the SCIF where you may work — to what’s available within government,” he said. “I’m pleased to say that that is actually closing.”
According to Gilmer, agencies are placing greater emphasis on designing AI environments that support operational speed, resiliency and flexibility across cloud, on-prem and edge deployments.
How Are GPUs Expanding AI Capabilities at the Edge?
Gilmer said advances in graphics processing unit-based computing have accelerated the deployment of AI tools in on-prem and edge environments.
“It used to be only small, very small, LLMs,” he said. “Now, I’d say it’s up to medium, very capable LLMs — things that can change how missions are run, or how enterprises operate, or certainly sensitive or classified workloads are processed in the government.”
Gilmer also said some government workloads require localized processing and tighter control over data movement.
“There are reasons we wouldn’t want data to leave a certain area,” he said.
Why Is Inference Becoming an AI Infrastructure Challenge?
Gilmer said inference workloads now account for most AI compute demand as agencies scale operational AI use cases, with inference representing approximately 80 percent to 90 percent of activity. He described inference as a significant variable cost driven by real-time queries, agent interactions and continuously operating AI systems.
According to Gilmer, AI is reshaping software engineering workflows by improving productivity through development agents that assist with coding, refactoring and testing.
He noted that agencies and contractors are prioritizing rapid deployment and continuous iteration as AI capabilities evolve.
