To clarify: I'm a retired industrial Design Engineer (from Engineer to Project Manager to Director of R&D). I worked closely with many manufacturers over decades. Including TI, Motorola, IBM, Intel, Toshiba, NEC & many others. I was involved in helping with prototypes of new design's for some of these companies.
Some of you may not be able to read the full article on Seeking Alpha (a free account is required). So I'll post the relevant part here:
According to multiple reports, Samsung will be able to build DRAM, NAND and logic chips at this new fab. Many reports have also stated that the main function of this new Samsung fab is to build DRAM. This would mean Samsung is massively increasing worldwide supply of DRAM when its supply is already currently too high. To make things significantly worse, Micron's 3D XPoint will likely begin to reduce DRAM demand as soon as mass production begins. With this set of information, there are a few plausible conclusions.
The first possible conclusion is that the people running Samsung hate money. Putting that much new capacity into an already oversupplied marketplace will crush the average selling price per GB of DRAM and cause Micron, SK Hynix and itself to bleed money. Electronics manufacturers who need DRAM as a component of their products benefit enormously at the expense of the three DRAM manufacturers. I think this scenario is highly unlikely.
The second possible conclusion is that Samsung believes DRAM demand will grow tremendously over the next several years to the point that current worldwide wafer capacity is nowhere near sufficient. Unlike most industries, it isn't as simple as spending extra dollars to produce higher volumes of a product immediately. It takes at least 2 years, in the most optimistic case, to build a modern fab for memory. Realistically, a number closer to three years is more likely. In order to meet demand 3-5 years from now, a memory manufacturer needs to start building a fab today.
PC demand is in decline, tablet and smartphone growth is not as robust as it once was and the Internet of Things does not seem to be picking up steam as fast as some had hoped. While the growth rate of the overall units of electronics using DRAM isn't as strong as it was a few years ago, the amount of DRAM going into each device is still growing rapidly. I believe the market could experience an era of slower declines in average selling price per GB over the next 3 years as DRAM manufacturers start shifting capacity away from DRAM production as XPoint production begins to ramp and SSDs (Solid State Drives) start consuming more NAND. As a result, this scenario of projecting a decade-long massive DRAM demand expansion in the future seems unlikely to me.
The third possible conclusion is Samsung is not planning on using the majority of this new fab to build DRAM. I believe this is the best conclusion to draw from the evidence. It is more likely that the company is building this capacity for logic chips to compete with Taiwan Semiconductor and Intel and NAND capacity to compete against Micron, SK Hynix and SanDisk. While some small amount of this new capacity could be used for DRAM, a larger portion should be dedicated to logic chips. The vast majority of these up to 6 million wafers per year should be dedicated to NAND for SSDs. This is the beginning of the end for the HDD (Hard Disk Drive).
I believe there are four main criteria to analyze when projecting how quickly a new semiconductor technology will obsolete an old one. These four criteria are: 1) cost; 2) speed; 3) energy efficiency; 4) compatibility and other engineering issues. In the case of SSDs, speed and energy efficiency strongly favor SSDs over HDDs. Compatibility and other engineering issues for SSDs compared to HDDs are relatively minor. The only thing that has stopped SSDs from completely obsoleting HDDs quickly is the high cost differential between SSDs and HDDs when measured per GB.
Intel and many others may argue that SSDs have a TCO (Total Cost of Ownership) that is lower than that of HDDs when measuring in dollars per IOPS, or the TCO differential is not nearly as great when factoring in the cost of electricity and longer lifespan of an SSD. However, it's only a part of the story. If all IT guys analyzed SSD cost in this manner, Western Digital (NYSE:WDC) and Seagate (NASDAQ:STX) would already be out of business. Clearly, that is not the case.
It costs an HDD manufacturer less than 3 cents to produce a GB. Prices vary depending on whether the drive is intended for consumer PCs or data center storage or some other purpose. The cost per gigabyte to produce NAND is 5-10 times higher than the cost to produce a gigabyte on an HDD today. Because of the significant performance benefits of SSDs, I do not believe SSDs have to be cheaper than HDDs to take significant market share from the HDD market. The price only has to be "reasonably close". I am not sure what "reasonably close" means yet, but I suspect if SSD cost per GB is within 2-3 times the cost to produce a GB on an HDD, there will be a massive increase in demand for SSDs and a corresponding drop in demand for HDDs. If price per GB drops below the 2x threshold, it is hard for me to imagine a scenario where HDDs find a market if there is enough NAND to supply the world's storage needs. This fact alone should explain why the NAND manufacturers are so keen on driving the cost of SSDs down to effectively compete against HDDs. With HDDs gone, a memory oligopoly's ability to control prices would increase tremendously.
Since SanDisk and Toshiba do not make DRAM, it is pretty clear any new capacity they create will be used to expand NAND capacity. Micron's capacity expansion in Singapore is for the purpose of increasing 3D NAND capacity. SK Hynix's plans are vague, but I would be shocked if any future capacity expansion during the next few years is not dedicated mostly to the production of 3D NAND. Unsurprisingly, I also believe Samsung's capacity expansion will be dedicated to 3D NAND. Over the next 5 years, I believe the growth of HDD aerial density will be significantly less than the 40-50% annual increases of the bit density of 3D NAND manufacturers. If I had to guess, HDD aerial density increases will be in the 10-20% range annually over the next decade. That is significantly slower progression than what Seagate and Western Digital have been able to achieve recently. If SSD price progression relative to HDD price progression deviates significantly from these estimates (over an extended period of time), it's highly likely that my thesis of what Samsung will use its new fab for is wrong.
I believe this gigantic Samsung fab, along with other planned new fab facilities from its competitors are intended to dramatically increase the supply of 3D NAND, lower manufacturing costs per GB and speed up the process of reaching the "point of no return" for HDDs. The storage industry will go from six (the four NAND manufacturers plus Seagate and Western Digital) to four (just the four NAND manufacturers). The customer base for the NAND manufacturers is diversifying, and the number of bits demanded in aggregate will increase dramatically (especially if HDDs are killed off) over the next few years, while the number of suppliers of data storage will shrink. The overall impact on the memory industry of HDD death will be significantly larger than the impact of Micron consolidating the industry with its acquisition of Elpida and Rexchip.
I will leave you with a quick estimate of when this HDD death might occur. Let's make the simplifying assumption that an increase in bit density is perfectly inversely correlated with the cost for manufacturing a GB of storage. That is to say, if bit density were to double, the cost per GB would be halved. I am also assuming that if the point of no return becomes clearly identifiable, the NAND manufacturers will massively increase capacity to take away the entire market share of the HDD manufacturers.
Most pessimistic scenario: If SSD cost per GB is 10 times higher than HDD cost per GB in 2016, 3D NAND bit density increases 40% per year, HDD bit density increases by 20% annually and the magic price differential per GB required for the "point of no return" is 2x, it will take roughly 10.4 years from the end of 2016 for SSD cost per GB to reach a point that it can kill off HDDs.
Most optimistic scenario: If, instead, we assume the SSD cost per GB is 5 times higher than HDD cost per GB in 2016, 3D NAND bit density increases 50% per year, HDD bit density increases only 10% annually and the magic price differential per GB required for the "point of no return" is 3x, it will take less than 1.7 years from the end of 2016 for SSDs to reach the critical price point to crush HDDs.
This gives an estimated time frame for the death of HDDs of sometime between late 2018 and early 2027. If I had to guess, I believe NAND manufacturers will experience a massive increase in demand and HDD manufacturers will fall off of a cliff sometime in 2021 or 2022. As far as my bet on 3D NAND, I have my money on Micron. Micron will go from the NAND laggard to the leader in 2016 in terms of bit density (and cost per bit) once its 32-layer 384 Gb TLC NAND hits the market in large quantities. I believe the company's choice of 3D NAND architecture may allow it to maintain this lead for many years to come.