[D] How did JAX fare in the post transformer world?

A few years ago, there was a lot of buzz around JAX, with some enthusiasts going as far as saying it would disrupt PyTorch. Every now and then, some big AI lab would release stuff in JAX or a PyTorch dev would write a post about it, and some insightful and inspired discourse would ensue with big prospects. However, chatter and development have considerably quieted down since transformers, large multimodal models, and the ongoing LLM fever. Is it still promising?

Or at least, this is my impression, which I concede might be myopic due to my research and industry needs.