The first one might not make sense but it gets them the pipeline for when it does make sense in the future. Plus for a lot of things using LLMs they'd be fine with older tech, especially if running it was even faster and cheaper.