Every conversation about decision superiority in the Pacific collides with the same phrase: "at the edge." Ask around about what that means, though, and you get different answers. Is the edge a forward base, a ship, a dismounted Marine, or a UAS in contested airspace? Each answer implies different assumptions about compute, power, bandwidth, and where the data lives.
Modern AI is power-hungry, compute-hungry, and data-hungry. The tactical edge is defined by the opposite: constrained size, weight, and power; contested and intermittent connectivity that must be engineered for resilience, not assumed away; and data that may be unreachable when it matters most. In a Denied, Disrupted, Intermittent, and Limited (DDIL) Pacific, the enterprise cloud is not a fallback and the network cannot be taken for granted.
This panel explores moving past buzzwords to talking about how to build for the edge. Where should inference happen, and where must training stay? What gets pre-positioned versus streamed? What does a resilient comms architecture look like when SATCOM, terrestrial, and mesh links are contested simultaneously, and how does the AI layer above it adapt? How much data compression can we tolerate before an AI product becomes untrustworthy?