things are changing too quickly for it to be worth it yet. eventually LLMs won't really increase in capability or resources anymore, and at that point, if the hardware itself isn't becoming more optimized for LLM workloads, you'd see people do this.