On top of all of that. It's that AMD thought they could get a profit out of it. It's not that AMD can't sell this cheaper it exists because they could sell it for this amount.
AMD planned to only have this be the MI50 and MI60. Because Nvidia took what was normally a $500 dollar product range product and priced it at $700, AMD having a similarly performing card could sell it at the same price. If the 2080 was $500 or lower the margins would have been tight and AMD probably would have been better off tossing the chips or creating another pro bin for the RVII quality chips for it. AMD has proven on several occasions that they will attempt to undercut Nvidia and boost Value when they have a product that succeeds at being cheap enough to manufacturer and off a chance at great volume. The RVII is not that. It's not as tight some have suggested as apparently it will have enough supply for AIB options later. But it exists because Nvidia left the door open to sell a little more of their low volume Pro part. If Navi comes in offering great performance or even decent performance. That is the product that AMD will price for Value.
The genius thing AMD did this launch on top of this rumor, was under supply the chip to partners and have the most readily available card on their website. This means AMD is not only eating the partner margin, they are eating the retailer margin. Add the 1 year warranty(low risk long term future cost), and AMD is laughing their way to the bank on this card. When you consider how little AMD has spent on R and D compared to Nvidia, the margins on this card are pretty good.
I don't see where people come up with these ideas. All the HBM2 cost estimates I've seen are still very high.
Just look at the memory cost currently for DDR4/SSD. There is a vast over supply in the market.
When HBM2 was first priced out for Vega, it was in a market where DDR4 was 200+ for 16gb(is is now half or less now) and the price of SSD has fallen vastly.
DDR4 and HBM2 are made on the same 20nm process. How this effect HBM2 and what made it so pricey on top of the manufacturing complexity, the opportunity cost of making HBM2 over DDR4. Since DDR4 could be sold for so much, it effected HBM2 pricing directly.
In addition production of HBM2 has ramped up substantially. In Samsungs case, double. So combine bad time for new discrete sales(over supply in market), death of mining(over supply of used market + less discrete sales), lower demand in the data center market, lower electronic sales for cell phone market(big impact on memory prices) and there is no reason for HBM2 to not be cheaper, especially when you consider the selling price of Vega back then compared to now.
If AMD was taking a loss initially on Vega 64 at 499 and are now able to sell for 150 dollars or 399, how is this type of pricing possible? Is AMD losing 150-200 dollars per card at $399. No, the answer is pricing for components have fallen and considering the high cost of HBM2 initially and how much it makes up the total bill of materials, that's is like the biggest source of why Vega 64 can be sold for 400 and Vega 56 can be sold for 350/330 even with a big game bundle.
Analysis of Radeon VII cost have mostly been done by Fudzilla whom are very unreliable along with wccftech which is mostly the same. Gamersnexus made a broad statement saying he see no reason why HBM2 is cheaper now compared to the launch of Vega shows he has little understanding of the financial market and failure to see the link between Vega pricing and HBM2 pricing.