Once computer constraints ease up, you will see much larger models. The reason LLM seems to have stalled a bit is because there just not enough compute.
You have more people using AI which requires more compute, and you want to build larger models which requires more compute and you have limited compute. What do you do?