I use a MacBook Pro with 18GB of unified memory.
I chose that amount because it seemed sufficient for my development workloads at the time of purchase. I develop a Phoenix server and some Docker containers for testing, and I’ve run out of memory once or twice when memory-hungry applications consumed too much. 
Several reasons, but mostly the following:
I bought 32GB for this MacBook Pro because it was double what I had before in my Surface Book 2. For my next machine, I would probably get at least double again, maybe even 128 GB if I could afford that much at the time. But, other than running some local LLMs or VMs, I don’t really need much more than I currently have.
There are several rationales (rationalizations, perhaps) that have gotten me into larger than normal machines.
In my consulting work, I’ll often times be working multiple projects with multiple clients with differing technical demands (non-technical management to data heavy technical work). So one thing I like to do is set up virtual machines for each client where I can keep data, administrative work products, etc. reasonably segregated and secured since sometimes the data can be sensitive. There are times when I’m working with multiple clients at once so having multiple client VMs running and ready to switch to is helpful; and since often I want each environment to be robust in its own right, the bare metal has to be substantial enough to service those environments.
When I am doing technical work, I’m often working with largish data sets… at least large as compared to personal workstation standards. So for example, I’ll be called on to do ERP data migrations in support of new ERP implementations or corporate mergers where systems are merging. While the final data migrations are not typical done on my equipment, my data conversion activities in advance of cutover are unless the data is simply too large. In the last migration I did for example, I needed to create a combination data migration and bi-directional data integration layer between a legacy SQL Server version of their application and a new PostgreSQL server based version. All of the real development work happened on my workstation where I setup A SQL Server VM, a PostgreSQL server VM, and a VM to host the integration software I was using; a reasonable equivalence to the final cloud production environment, but where I had access to more nobs/dials/logs to really know the dynamics. The dataset (to that point) was 10-20GB. I was able to get a reasonable end-to-end picture and a reasonable modelling of whole system performance (which was important) prior to bringing it all to the cloud PaaS brain damage that would complicate things.
There are non-development reasons, too. All of my formal education and even work for a time when I was young (many years ago) is in music composition (specifically film music composition). So… I will sometimes use this workstation as a DAW and use various other composition tools like notation software… the various sample libraries I use can be very RAM hungry. Also, I like being able to have Blender available, though I’ve only really toyed with it. Some of this stuff is admittedly aspirational right now since I don’t have enough free time, but assuming a long enough dry-spell comes my way and I’ll have some major works of art (LOL) in progress.
Anyway, there’s also a strong case that I really, really don’t need all of this and I’m just satiating my “I never want to buy another computer again” urges.
I too have a MacBook Pro with an odd 18 GB of RAM, it was the minimum option when opting for the M3 Pro chip.
I updated from 64 to 128 because I test a lot of AI models locally.
Client data should never be exposed to cloud. So I need a good setup, just to run different AI pipelines.
Thanks for the responses everyone - very interesting!
Ah that’s awesome! I used to have a little home-studio when I lived with my parents and it’s something I’ve wanted to get back into for a while. Which DAW do you use?
Do you find much difference between models below/above 64GB?
When @TimButterfield and were testing DeepSeek models I felt that the 4GB model produced better results than the 8GB version. Curious whether you’ve noticed much difference between 64GB vs 128GB models?
For DAW, I use Reaper. Sometimes it feels like the musical equivalent of Emacs (for better and worse), but the set of features for the price make the value proposition very difficult to beat. For notation I use Dorico… I was a long time Finale user (like from version 1 in the late 80’s), but that product got discontinued several months ago.
Pretty much just because it’s possible. I went from an ancient ThinkPad with one DDR3 slot to a Framework, which has two DDR5 slots and my boss just thought it would be funny to get two 48GB sticks. Haven’t really found a use for all that memory.
Late to the party, but my answer is “whatever the max amount of RAM I could put on that MacBook when I bought it”.
Weirdly, it’s mostly an amortisation technique to lengthen the useful life of the machine and spread the cost out more.
With MacBooks , you pay a heck of a lot for the core components like the screen, casing, keyboard, trackpad etc. I don’t resent it, because they’re all of the highest quality and will last for many years.
But then, if you skimp on the memory, you artificially limit the useful life of the system (e.g 18GB - 3 years tops, IMO), and are forced to replace everything, even though there’s still plenty of life in them.
In fact, anyone who bought an M1 with minimum RAM back in 2021, would be looking at an upgrade where they pay again for the exact same screen, keyboard, trackpad etc, even though theirs are probably still great, just to get more performance.
OTOH, having gone all in and maxed the RAM, I’ve never once thought “I wish I had less memory”. The systems are an absolute joy to use, and allow me a degree of optionality that love. The first few years even offer a sort of “if I can’t do it on my dev machine, I probably can’t do it (economically) anywhere else” sanity check.
Finally, I hate switching machines, so the less often I have to do it, the more time I can spend on other things 