AI will need less HBM (high bandwidth memory) because flash memory unification is changing training and inference. This episode of the Tech Field Day podcast features Sebastien Jean from Phison, Max Mortillaro, Brian Martin, and Alastair Cooke. Training, fine-tuning, and inference with Large Language Models traditionally use GPUs with high bandwidth memory to hold entire data models and data sets. Phison’s aiDaptiv+ framework offers the ability to trade lower cost of infrastructure against training speed or allow larger data sets (context) for inference. This approach enables users to balance cost, compute, and memory needs, making larger models accessible without requiring top-of-the-line GPUs, and giving smaller companies more access to generative AI.