- The bad news: Memory prices will decline.
- The good news: Micron's margins will increase.
- The better news: It really is different this time.
If you are a Micron (NASDAQ:MU) investor, get ready for some share price volatility. This baby will do a lot of ziggin' and zaggin' from here on its way to $40-50, or maybe even much higher.
Also get ready for the inevitable claims from pundits that we are heading into another old memory price decline spiral.
For a beginning point, you get my nutshell opinion; I will send the invoices out later.
Micron is a $16 billion+ memory supplier. For simplicity, we will consider the major businesses as DRAM 65%, NAND 25%, other 10%.
In my always humble opinion, the Micron DRAM business is technologically the weakest segment, NAND is the technological leader of the business, and the other 10% is just a cash generator. It seems unfortunate that the largest segment of the company is also the weakest. Now remember, this is just my opinion.
I believe it is weak because virtually the whole DRAM segment of Micron is at a 30nm node, while much of the competition is at or moving to the 20nm node. Micron would have moved there as well, except the Elpida acquisition has been "something" of a distraction; a very nice distraction.
The good news is that Micron has started the conversion to 20 nm DRAM. The conversion will take a year of ramping and qualification at major customers.
Here's what happens:
Let's assume (I'm going to leave out a lot of arcane details here) that the 30nm 4Gb DRAM is 70 sq. mm and produce about 745 yielded chips per wafer, at $3.50 ASP (Average Selling Price) would produce about $2600 revenue per wafer. So, the 30-20 nm shrink would cut the die size in half… in the old days. With the super small nodes, we might only get a size reduction to 50 sq mm. That would get 1044 yielded chips per wafer. Customers would expect some sharing of the cost savings, so expect the ASP to decline at least 10%, to $3.15. $3.15 X 1044 gets about $3300 revenue per wafer. The cost is about the same, so Micron would make $700 more per wafer even at the lower price. Micron would, at the end of the ramp, produce 40% more DRAM chips per wafer, however, since this is something of a linear ramp, the total bit growth might only be 20%. Can the market absorb 20% more DRAM bits? Forecasters expect bit demand growth of 40% in 2014. Micron is a little more conservative at 30% bit growth, so the 20% output growth seems like no problem at all.
Pre-shrink annual revenue would be about 4,200,000 wafers times $2600, or $10.9 billion for DRAM. Post-shrink revenue (during the ramp) would be the 4,200,000 wafers times about $2950, or $12.4 billion. The extra $350 (one half the end point of $700 more per wafer) is all gross profit, so $350/2600 is 13.5% MORE gross margin even at the lower price per unit.
Nevertheless, expect a lot of noise when the DRAMeXchange spot price starts heading toward $3.15. This is the semiconductor business in a nutshell with a bright light shone on the memory segment of that business.
These new parts will require qualification by major customers. In the old days (I've been there), the memory supplier would trot into the customer with qual samples and data package in hand and beg for a timely qualification. At this point, the customer would say something like, "I have this new memory supplier from Mars who has quoted a price of $2 on that part. We might qualify him instead, unless you want to meet the price."
Well, today there are no "new suppliers from Mars," there is only Samsung (OTC:SSNLF) and Hynix (OTC:HXSCF), and they have largely moved to 2X nodes for DRAM already and there might not be any further shrinks.
Now Micron can trot into the customer with qual parts and data package in hand and suggest that they hurry up and qualify the new parts to avoid a supply interruption. The memory world is truly different with an oligopoly.
In the world of NAND, Micron is the technology leader with the 16 nm HKMG part (again, my opinion). It is the smallest MLC part in production and will be cheaper to build than any competitive 2D or 3D part for a long, long time.
NAND is a completely different story than DRAM.
The applications for NAND are memory boards for digital cameras, thumb drives of all sizes, MP3 players, etc. All of these applications could use big parts like the 128Gb part, but they usually use 64, 32, and even 16Gb parts. The Apple iPod was the biggest boon for NAND memory as it replaced the tiny HDD (Hard Disk Drive) used in the iPod Classic (is this sounding familiar). Mobile phones were the next big user of NAND memory, and finally SSDs will consume an enormous amount of NAND while replacing HDDs. The introduction of the 128Gb part made MP3, smartphone, and SSD (Solid State Drive) storage much easier and, in some cases, even possible. For example, the iPhone with 16GB of storage could use two 64Gb parts stacked or one 128Gb part. The 64GB iPhone could use four 128Gb parts or eight 64Gb parts. Eight of these 64Gb chips is really starting to pack two pounds of stuff in a one pound bag. The 128Gb chip make SSDs much more compact.
When compared to all NAND chips below 128Gb in size, the 128Gb chip is now the lion's share of the NAND market dollars.
A quick look at the PC potential for SSDs made of NAND flash is mind boggling. Fully developed, that market alone is 300 million SSDs with about an average of 12 128Gb chips per drive (192GB average drive size.) That's 3.6 billion of these chips at about $8 per chip. That's a $30 billion market, virtually all of it made up of 128Gb chips. A billion smartphones using an average of 32GB of storage is another 2 billion of these big NAND chips, and another $16 billion market. You might add another $15-20 billion in data center storage. All in the form of 128Gb chips.
The other interesting development in the NAND business is the tendency for the chip manufacturers to make the SSDs in-house. The third party SSD suppliers are dropping like flies, and very soon all SSDs will carry the trademarks of Samsung, Micron, Hynix, Toshiba (OTCPK:TOSYY)/SanDisk (NASDAQ:SNDK).
There are a couple of points to be made here. The first is that the overwhelming percentage of dollars in the NAND business will be in the form of the 128Gb chip (or, eventually a 256Gb chip), so looking at DRAMeXchange for NAND spot price is the ultimate fool's errand since DRAMeXchange doesn't even post a spot price for the 128Gb chip. I guess the site could tell you something about pricing on the lower density parts that, as a percentage of the total market, is only about 10%, going to 5%. Even contract pricing is of limited value since much of the high density output is kept by the NAND manufacturer itself for use in SSDs.
All the above really doesn't matter because you, as a Micron investor, are going to be told every day that NAND pricing is going in the sewer.
A far better endeavor would be to track SSD pricing. Since about 90% of the value of a SSD is in NAND chips, you could divide the SSD capacity in GB by 16 to get the number of the all-important 128Gb chips in the drive. For example, a 128GB SSD would have 128/16 or 8 chips in it. If it sold for $100, 90% of the price or $90 would be in those eight NAND chips, which give a value for the chip of $11.25. Given that the cost of that chip is under $5, NAND margins (60%) in the form of SSDs is a very good business.
Given that we have been told about collapsing NAND price all of the past quarter, I find it reassuring that the most recent SanDisk quarter result (the only pure play NAND company) showed 50% gross margin.
Don't let the talking heads scare you about "declining NAND prices."
The other point to make is the sheer enormity of the NAND market. Between client PC, smartphones, and the data center market, a $60 billion total market is developing at a rather high compound growth rate. The total NAND business is expected to grow 13% to $28 billion in 2014, but SSDs will nearly double consumption of NAND to 25% of the total in 2014.
In order for this market to achieve its potential, another 9.3 million NAND wafers per year will have to be manufactured. Since a very large NAND fab would have capacity of about a million wafers per year, something like nine new giant fabs will have to be built. Based on that little tidbit of information, the fairy tale, in defense of the idea that "this time is different," that no new fabs will be built is obviously flawed.
New fabs WILL be built AND this time is truly different; any surpluses will be predictable and short lived. New fabs will have to be capacity flexible in order to ramp a very short step after demand increases.
I can see a first phase of this necessary expansion of output of NAND memory as an expansion of the Micron/Intel (NASDAQ:INTC) joint venture that's already in place. This early growth phase could utilize some of the fully depreciated excess Intel fab capacity around the world.
In the next few years, NAND at Micron will add $10 billion in revenue to the company's total. Expect a $27-30 billion Micron within five years. The company will earn 25% of that spread over 1 billion shares. Hang on, and you will see earnings of $7.50 per share for Micron.
Memory prices will decline, however, this time costs will decline faster, so gross margin will grow to 50% or more. More fabs will be built, but this time is truly different; there will be no pricing death spiral.