Zero to Mastery started as Andrei Neagoie's web development courses and has grown into a proper academy, and the part that matters for this site is its AI and machine learning track. Most of that track is taught by Daniel Bourke, and honestly he is the reason I rate this so highly. His Complete Machine Learning and Data Science Bootcamp, his PyTorch for Deep Learning course and the TensorFlow material all share the same quality, which is that he explains the why before the how and he writes code live so you see the mistakes and the fixes rather than a polished final answer. That sounds small but it is the difference between watching someone cook and actually learning to cook.
The whole thing is structured around building projects, so by the time you finish you have notebooks and models you can point at rather than a list of videos you sat through. The single best thing about ZTM commercially is the subscription itself. You pay one membership and the entire library opens up, so if you start in machine learning and realise you need to shore up your Python or your data analysis, you just go and do that course too at no extra cost. The Discord community is also more alive than most, and it is not unusual to get a real answer from an instructor.
The trade-offs are the obvious ones for this model. If you only want the ML content, you are still paying for everything else, and the annual price makes more sense than the monthly one only if you are going to keep at it for months. There is some overlap between the courses, the path is not perfectly tidy, and the certificates are not accredited, so nobody is hiring you because you have a ZTM badge. The real risk, as with anything self-paced, is you.
Nothing here chases you for missed deadlines, so the people who get the most out of it are the ones who can sit down every week without being told to. Used that way, it is one of the best value routes into practical machine learning I know of.