LLMs as we know them have no real moat, but few people genuinely believe that LLMs are sufficient as a platform for AGI. Whatever it takes to add object permanence and long-term memory assimilation to LLMs may not be so easy to run on your 4090 at home.