While talking with a group of people about my current project, I mentioned how much money I am spending on my RenderStreet subscription, and they suggested that my own hardware should be sufficient. I have already done some simple things to speed up render time, such as light portals and "tricky" window glass, and I did not think that my scene is so computationally trivial. I think what I really need is your subjective judgment of the computational complexity of my scene.
I am working on representing my apartment. I started on January 19, 2020, and I hope to have it looking fairly reasonable by January 24, 2021. I work on it every weekend and then upload to RenderStreet and let it render during the week. Here is my most recent video:
https://drive.google.com/open?id=1fUG9iF3sWxBNAoxtW-D4VLmnjAZl65q2
That video has 369,271 triangles, and the number of triangles increases with my work each week. I am using a 16k environment map, but I must admit that is not strictly necessary. I am planning to try microdisplacements, maybe about two months from now.
Consider the indirect lighting involved, especially on the back of the toilet where you may notice that for now, I am not even using adequate samples. I imagine I could get good performance with things like light baking and Eevee, but I do not think such things make sense when I am in the middle of working on the modeling and I just want to render a video that shows my progress.
Based on your experience, would you expect that what I am doing should be feasible just by rendering on my own computer, or does it make sense that I need RenderStreet?
Do some calculations; you now have 1021 frames, but the video is way too fast, so you might end up with a lot more frames...
If one frame would take 3 minutes to render on your computer, that would mean about 48 hours of total render time with the current number of frames.
The 'Classroom' scene took an average of about 8 minutes to render (Blender Open Data), to give you an idea. (but that is on good computers, on my pc it takes about 90 minutes at 300 Samples.)
Denoising doesn't perform optimal in 'animations'.
Eevee will of course be a lot faster, but you lose some realism and it doesn't support Displacement.
Personally I'd just render out a few shots each week and only make the whole video when finished, but that's up to you.