>annual global DRAM bit demand for 2018 will grow by a larger margin of 20.6%.
>the plans from the top three suppliers indicate that their respective wafer start volumes will expand by only 5% to 7% annually in 2018
Maybe it isn’t possible to built that much extra capacity. Maybe there isn’t the necessary budget/capital for that level of expansion? Maybe 5% is a sustainable rate of growth (and their supply chain), but 20% isn’t? There are lots of other factors at play here.
Contracting ‘relative to demand’ is a really disingenuous way to frame the argument.
Then there is the maths of 20% increase in "bits".
You see you don't need to build new fab to have increase in "bits" capacity. 3D NAND, 3D DRAM, EUV DRAM which Samsung is working on. Sub 10nm Node which Samsung / Micron are also working on and coming soon. All of these will increase their "bits" capacity without substantial investment. And as DDR5 standardise with larger maximum capacity per bank, better yield and leading node, and an expected increase of "bits" capacity by 50%+ in the next 3 years. Who will be there to say we will sell 50% more in 3 years time? Especially when Mobile sales are already slowing.
There is the China question. Before any media coverage started in late 2017 and 2018. China had always have plan to have their own DRAM and NAND Fab, with government money started pouring in 2015. That is, in the worst possible situation, assuming China somehow suddenly understand how to make DDR4, LPDDR4, what is there to stop them the race to bottom, and their massive influx of cash to build out new Fabs to bring huge capacity into the industry. Are you really willing to burn your own money, in the tens of billions against a nation?
Look at what happen to Steel, Solar Panel, etc The only difference is making DRAM is a lot harder then those two. I specifically leave out NAND here because 3D NAND are even harder for them to understand and grasp.
And I too, wanted cheaper DRAM and NAND. And I know in the old Cartel working, where they actually "reduce" capacity and hold price together via talks. But right now, I think the market dynamics is that they don't need to do any of these tricks to get what they wanted.
Unless you have evidence to support this, all the company above you listed has have expansion capacity planned and made during the last two years.
>That's exactly how it's supposed to work. Every manufacturer is individually incentivized to increase supply to increase their own relative share of the market.
And they are. Yes, there may be a cartel going on. But it is a simple fact demand is outstripping supply. And no memory maker are willing to invest money into building fabs. Because building a new Fabs along with leading node is increasingly expensive. Moore's law didn't just ended for CPU, it is for all semiconductor companies. All current Memory makers has roadmap that will add 30% plus bit capacity to NAND without even making a new Fab with 3D NAND. GDDR Memory is also in complete shortage, and was blamed multiple times where GPU maker cant ship as much as they could, not that Nvidia and AMD could built them fast enough with TSMC anyway. TSMC has stated they have a whole dedicated fab for customer making their own Bitcoin ASIC, ( And has since confirmed as BitMine ), no GPU maker can make a prediction of how and when the Crypto craze ends, how are memory makers suppose to plan for their GDDR production capacity other then being conservative about it. And Nvidia now are getting problem of having high volume of GPU sitting in channel after BitCoin and Crypto crash.
Then there is the machine learning, AI craze, Datacenter segment seems to be growing with no limit. As shown outpacing Intel's projection quarter after quarter.
With Fab investment continue to increase, and takes longer to recoup, do you expect any sane company to take risk into building new Fab? When you know somewhere in the back of your head these craziness may not be sustainable?