Can decentralized GPU networks still contribute to AI with hyperscale centers dominating training but real potential in inference?

The majority of AI training is currently conducted in hyperscale data centers, with a few large data centers processing the bulk of the workload. However, as inference and daily tasks become more prevalent, there is a growing opportunity for decentralized GPU networks to play a significant role. This shift indicates a potential opening for smaller scale GPU networks to handle a portion of the processing load, diversifying the landscape of AI computation.