I don't think they should have too much to worry about if they can bring out a transformative new model in the next year, but if GPT-5 is only slightly better t

I don't think they should have too much to worry about if they can bring out a transformative new model in the next year, but if GPT-5 is only slightly better than GPT-4 I would start to worry if I worked there - LLaMa, Gemini, Claude, Mistral are all about as good or a bit better than GPT. You can download LLaMa or Mistral for free, why pay OpenAI? GPT3/4 were like pulling a rabbit out of a hat, it seemed at the time that OpenAI was guaranteed to be a huge player, Google or Apple-level. But now there are these competitors, where does the money come from? You can't burn tons of cash on running GPUs forever, and if they switch to a paid model for ChatGPT I think that 95% of people will switch to Claude or something that's free - because ChatGPT isn't that much better. People were predicting that LLMs were going to be like a commodity and I think they've been proven right. A lot of companies can make them, and it's just GPU spend to keep them running. You need something to differentiate to make real profits. Plus, I think the future of AI is going to be on-device to solve the issue of GPU spend. When manufacturers start making AI-specific hardware and every major OS has its own local LLM, where does that put OpenAI? How do they make money out of that?