"SOF units already use generative AI 'heavily' for things like resource allocation and force deployment, and are 'delving' into its use for tactical operations," said Rob McClintock, the program manager for intelligence for the program executive office for digital applications.
From cloud centers to the tactical edge: the operational demand
At the Global SOF Foundation’s SOF Week in Tampa, Florida, officials framed a practical problem: powerful generative models live in cloud-connected data centers, but special operators do not always operate where those networks reach. Physical proximity to the "tactical edge," event speakers said, would enable faster use of mission‑critical data and faster decision‑making—capabilities operators say they need when disconnected from continental infrastructure.
Special Operations Command and the search for "fog computing"
Special Operations Command is pursuing frameworks that push the computational power of cloud systems closer to where data is collected and used, a concept attendees described as "fog computing." That approach is intended to bridge the gap between massive centralized processing and the disconnected, bandwidth‑constrained environments special operators encounter.
Col. Robert "Ramsey" Oliver on voice, cognitive load, and autonomy
"In that conversation about managing the cognitive load on operators, voice command is a logical step," said Col. Robert "Ramsey" Oliver, PEO of SOCOM’s SOF Warrior. Officials emphasized simpler interfaces—spoken or even gestured commands—so operators can plan and execute missions without being diverted into complex digital workflows. Lt. Col. Aaron Davidson, program manager for the unmanned systems autonomy and Interoperability portfolio, highlighted practical use cases: getting different types of drones to work together, and carrying out mission plans with minimal input.
Melissa Johnson on acquisition: smaller businesses over big tech
SOCOM’s acquisition executive, Melissa Johnson, told attendees that the solutions SOCOM needs may come from smaller startups rather than the largest consumer tech firms. "From an acquisition perspective, we're not just limited to the bigger companies with their own mindset, because AI is very dynamic," she said. "Sometimes the smaller organizations, smaller businesses bring those solution sets."
Rob McClintock on agents and lighter language models
McClintock said work is under way on AI "agents" that can plan, revise and execute strategies—autonomous software able to manage aspects of a mission workflow. In parallel, the command is seeking versions of large language models that require less computing power but still "understand human intent with less instruction," reducing the processing and bandwidth burden at the edge.
What this means for technologists, procurement leaders, and operators
- Technologists and security teams: expect requirements that prioritize compact compute footprints, low‑latency local inference, and interfaces that accept voice and gesture inputs—engineering that operates without always relying on cloud connectivity.
- Procurement leaders and acquisition executives: opportunities favor smaller businesses and startups that can tailor niche, tactical solutions; acquisition decisions will weigh dynamism and fit to specific operational constraints over scale alone, per Melissa Johnson.
- Operators and program managers: the immediate aim is tools that make existing tasks easier—coordinating heterogeneous drones, planning missions with fewer manual steps, and offloading routine cognitive load to AI agents, as Col. Oliver and Lt. Col. Davidson described.
The recurring thread at SOF Week was pragmatic: special operations units want the computational power of data centers, but packaged for the pack—smaller, easier to use, and capable of functioning where networks do not. The next steps cited by officials are technical: build lighter models, enable agentic behaviors, and move cloud capabilities closer to the tactical edge. Whether those pieces come together from large firms or the smaller businesses Melissa Johnson singled out will shape how quickly operators see the tools they asked for.
