Depends on what a person or company is using it for. Some can and do spend a million dollars a day or more - but they are using advanced models with a lot of calculations. Molecular biology, coding, real time monitoring of a decentralized environment, movie making etc will burn a lot of tokens because it is running a lot of heavy, or constant, calculations. The cost for the biggest “power” users has only been going up this last year. For us regular folk the free or the $20 a month plans work just fine. But those are not the best models and the tasks are simple.
I find it helpful for several things in my line so I pay $20 just so it doesn’t time me out. And it remembers an entire conversation so I have ongoing threads where I ask for feedback, take it to the real world, then tell it the results the next day, and what I need, and it gives me relevant feedback. I’ve got several weeks-long conversations going back and forth until I get it right.
Next I will start tinkering with images and video and see what agents may be useful to me. It’s much more complicated and complex and may need to switch to a different AI for those. I’m not much into video or imagery but do need it so I’ll dabble and see what it can do. Oh I also had it redesign some web pages for me. I pasted in the code and told it what I wanted changed and it gave me new code. Worked perfectly. So it can be useful depending what your needs are. These are simple tasks. I’m sure large enterprises are doing way more difficult tasks AND/OR have many thousands of employees using it thus running up a massive bill.
https://x.com/i/grok/share/4e908a1a705f4fe5969de0a934b7db52
I use local AI with Continue on VS Code, and it works great.
I am working my way through learning the PyTorch stack.
https://www.youtube.com/watch?v=23VkGD-4uwk&t=325s
PS - My system cost less than $1500.