Inference Runs Where Data Lives: From Drone to Rack, One Inference Platform

Inference has left the data center. It now runs on a drone in flight, a pole at an intersection, a haul road, a drive-thru. In those places, there is no round trip to a data center, and the power budget is fixed before the workload is chosen.

This session covers what that costs. Latency budgets measured in milliseconds. Accelerators that pay a switching penalty every time a real scene needs detection, depth, thermal, and fusion running at once. And what breaks at scale, where tens of thousands of streams defeat centralized and edge-only architectures alike.

Real deployments converge on the same shape: inference at the source, aggregation and retraining in the rack, one platform rather than two. Includes deployment data from live programs across air, surface, and undersea platforms, and rack-scale benchmarks for the aggregation tier.