My CNC machine is an older version of this : http://www.axminster.co.uk/axminster-engineer-series-zx30m-mill-drill-505109
Nothing special, but good enough for my purposes. 50 micron repeat positioning may sound small, but I use a lot of dental burs down to 0.7mm diameter, and under a jewelers loupe you can definitely see if it's out by 10%, even 5% with a bit of practice and a good eye. So yes, 50 micron is totally doable on modest equipment. (And I wouldn't dismiss claims of 10 micron on a newer machine and well-setup backlash compensation.)
As for the moulds and mouldings, yeah, they're still homemade, but as a proof-of-concept they're better than a 3D printed alternative.
And you can churn out hundreds of them in an afternoon.
And test out different materials on the fly.
And work out moulding issues like where to feed and where to vent.
And learn what size machine and what settings you need for mass production.
And impress potential customers and investors.
And it's also quite satisfying!
Who cares about positioning? That means nothing. The mold you're cutting doesn't care if you got the endmill in the vicinity of the toolpath you were trying to cut. All that matters is the finished part off the machine.
I don't mean this in reply to you at all - so please don't take any offense. But I always laugh when I see these ridiculous claims (usually from hobby 3D printer makers/owners) about how their machines will hold 50, 20, 10, 5 microns or whatever. It used to be that they would perhaps take a stepper resolution and divide by a gear tooth count and quote some theoretical number, but now it seems they just pull numbers out of thin air, or perhaps they just use a 32 bit number instead of a 16 bit number and quote the theoretical accuracy based on how many bits they are sending to the motor driver? But I digress
For making a mold, all that matters is what gets cut. I've seen hobby machines make parts with tooling marks that are way, way over 50 microns. Holding 50 microns is pretty difficult, even for a 'real' CNC machine. It varies depending on what material you're cutting, what sort of cutting operation, how you're holding the material and all of that. I'd definitely agree that some hobby machines can hold 50 microns under some circumstances. But then again, I can throw a dart and hit a bullseye under some circumstances... but the only thing that counts if I wanted to be a professional would be my ability to do that reliably again and again. I'd say "some circumstances" would probably be some small percentage of the time when conditions are right and the operator knows the machine well and so on.
No offense to the hobby machines intended or their owners. But 50 microns most definitely isn't easy.
EDIT: I just checked out the machine you have. Looks neat! Very similar or maybe even the same as the Grizzly G1007 I started out with years ago. But if yours is similar to my Grizzly, it uses acme screws - and in that case, that is not machine that will hold 50 microns unless you go to lengths to work around its limitations - like putting measurement instruments on its axes. And there's enough backlash in the screws that cuts are generally a "set it up and make the cut, then stop it and set it up for another one" rather than "make a program, hit go and wait for your part to be done" affair. My G1007 had to be re-setup every time I raised and lowered the head too, due to the cylindrical column. And the spindle bearings and R8 collet setup were very limiting and contributed to enough runout that anything other than quite short tooling was already pushing the 50 micron boundaries just due to that issue alone. And I'd say your machine is most definitely on the very high end of hobby level... IIRC a lot of people CNC those things (or did back in the day). People were putting laser pointers with mirrors on the head to get the head re-aligned when they moved it. No knock on your or the machine - sounds like you are getting fantastic results but would you say you could make an injection mold in aluminum on that machine and be able to hold 50 microns on the finished part? Including curves and transitions and radii and such? I think that would be too tall of an order, IMO.