I think it's less about LLMs suddenly hitting a theoretical ceiling or politics getting a grip on the industry. But there is a very real threat from actual open

I think it's less about LLMs suddenly hitting a theoretical ceiling or politics getting a grip on the industry. But there is a very real threat from actual open AI research. Free models have advanced remarkably and if you choose to, you can run something between GPT3.5 and GPT4 all by yourself right now. With the spread of cloud GPU providers, this could become cheaper than anything OpenAI offers very soon. I fully expect a GPT-4 (or better) level open weights model before this year is over. Right now OpenAI is desperately trying to differentiate itself from other chat model providers by offering tons of side features on the backend that have little to do with the LLM itself. That might lock in some of the bigger customers for a while, but this strategy isn't sustainable for long in the current climate.