There is circumstantial evidence that, among others, NVIDIA and Together AI are behind it in one way or another.
Together AI hosts LLM inference on their own GPUs. Together AI subsidizes this inference from the profits of $PRL (Pearl), but miners with their GPUs in the Pearl network are not running actual inference. They are running useless PoW in an AI trench coat. "Useful" is snake oil, but more wasted compute, more greed, and higher prices are very real.