VIXA Wildfire AI is an integrated wildfire detection, monitoring, and spread prediction platform powered by custom NPU hardware. The system establishes an automated disaster response workflow combining edge AI, Vision Language Model (VLM) verification, spatial-temporal forecasting, and autonomous drone sensing. Edge cameras run optimized YOLOv7/v9 models on NPU AI Boxes, detecting smoke and flame within 1 km. To eliminate false alarms by 90%, an on-device VLM (Llama 3.2-3B) semantically validates visual evidence and generates natural-language incident reports. Upon ignition, a spatial-temporal AI engine (ConvLSTM + GNN) fuses video feeds, terrain elevation, and weather vector fields (wind speed/direction) to predict spread routes. Long-endurance hybrid EO/IR drones (100+ min flight time) with onboard NPUs fly over predicted fire perimeters, streaming live aerial video to update spread models dynamically. Synthesized intelligence is visualized on a GIS command dashboard and broadcast via municipal emergency networks for immediate evacuation.