You’ll have costs to store the data and constantly training the models. Chatgpt costs millions. Also LLMs and NN in general have an issue with catastrophic forgetting so if you try to train on new data it will overwrite old data it’s learned on, meaning you have to train it on all the data at once, which means bigger and bigger training sets.

I don’t work on these myself but this my high level understanding.

Still, depends on what you want service you want to offer (a chatgpt clone might be too costly but maybe another service could be cheaper)