Just to point out some actual information for people on why it could be 4GB only. HBM, it comes in 1GB stacks that provide 128GB/s bandwidth each at standard clocks of 1Ghz. So 4 stacks would give you 4GB and 512GB/s of bandwidth along with it. Can you do 6 stacks and get 6GB, sure, 8 stacks, sure.
The main situation is that in sticking chips together on an interposer (effectively a pcb on silicon scale) the more chips you stick together the worse the yields get. Stick 4 memory stacks and a gpu on an interposer, if one connection fails they all go in the bin basically. So 8 stacks ends up having a significantly higher failure rate than 4 stacks, there is going to be a financially sensible limit to how many stacks you can put on a cheapish consumer card.
It's all new tech and we don't really know what yields they've hit. Maybe it will be a 6GB card because 6 stacks is feasible maybe not. Professional cards could probably absorb the costs of adding more memory due to the selling price but in the £300-450 range it would really hurt prices.
First cards could be higher capacity or use more stacks. Same way DDR4 was first specced lower than when it eventually came out, people are taking the lowest spec for HBM as absolute certainty of the first chips we'll see, which is frankly nonsense, and HBM2.0 as the only possible time we could get a higher capacity, again also nonsense.
Personally I'm fine with 4GB for a couple years, get me a nice 120-144hz 1440p 16:9 monitor for now. 2-3 years from now with affordable and decent 120hz 4k screens we'll be another generation on and have 8-16gb cards by then.