Sure they are. Surf around on HuggingFace and you'll see dozens of open-weight models up to 120B parameters in size from for-profit US companies, freely downloadable (if not freely runnable, alas). Probably hundreds of them, at this point.

And a 3T parameter model is scheduled to be dropped by the Chinese on Monday.

They should certainly publish more, and if somebody were to argue that model weights trained by scraping copyrighted data should inherently be accessible to everyone, I'd be 100% in favor of that.