Data Processing is Becoming a GPU Workload

The article argues that data processing is shifting from tabular data handled with SQL on CPU clusters toward unstructured multimodal data (video, audio, images, sensor data) that requires model inference on GPUs. It frames three concurrent changes - tabular to multimodal, SQL to inference, and CPU to GPU - and notes that existing big data systems built for homogeneous CPU clusters struggle with heterogeneous AI workloads needing specialized compute and streaming execution rather than batch processing.