None, exactly zero, of the companies in AI, no matter who, created any of the training data themself. They harvested it from the internet. From D*scord, Reddit, Twitter, Youtube, from image sites, from fan-fiction sites, wikipedia, news, magazines and so on. Sure, they used money for the hardware and energy to train the models on, but a training can only be as good as the input and for that, their core business, the quality of the input, they paid literally nothing.

On top of that everything ran and runs on open source software.

Therefore they should be required to release the models and give everyone access to them in the same way they got access to the training data in the first place. They still can offer a service, after all running a model still needs skills: you need to finetune, use the right settings, provide the infrastructure and so on. That they can still sell if they want to, however harvesting the whole internet and then keeping the result private to make money off it is just theft.

Fight me.