There are open models, eg. Bloom. Same with Stable Diffusion (open) and Dall-E. Creating these isn’t impossibly hard. Some ML ops company showed how you could train a GPT-like model for like $250k or something.