The memory business has been notorious over the years for boom and bust cycles.
Whoever said, "The cure for high prices is high prices" or "The cure for low prices is low prices" had to have the memory business in mind.
In the past the shortages and resultant price booms have followed periods of low prices caused by over-production. The low price cause new investment in capacity to dry up. As the general electronics industry continues to grow, the demand for memory outstrips the available capacity, prices increase, allocation is imposed and money starts to flow into new capacity, which is overbuilt and prices decline again. Rinse and repeat. This has been the story of the memory business for the past 40 years.
Does this sound like the oil and gas industry? How about the airline industry?
The memory business was so attractive at times that nation-states wanted their countries to be involved (oil and airlines again) and would provide the money for the required capacity. Japan subsidized its memory business which destroyed most of the American memory manufacturers. South Korea and Taiwan took their turns at national funding of memory businesses. Taiwan failed. South Korea in the form of Samsung (OTC:SSNLF) and Hynix (OTC:HXSCL) (the former Hyundai semiconductor) survived the exit of government subsidies (Hynix, just barely) and are now members of the remaining four memory manufacturers that make up what has condensed down to an oligopoly of four players in the memory business.
We are now in another of the up cycles in the memory business and people like me have been saying that this time is different.
Some of the reasons I believe this time is different are:
Nation-States are now out of the memory subsidy game.
The ability to double capacity with a one-third linear shrink of the lithography to make memory chips is at or near its limit.
The biggest historical driver of the memory business, the PC, is now less than 50% of the DRAM business.
NAND flash memory has grown, largely unnoticed, to be equal to the DRAM business and will continue to expand to multiple times the size of the DRAM business.
The memory manufacturers are all building solid state drives, thus using a significant percentage of their own output and using the highest level parts (128Gb) to make these drives. DRAMeXchange doesn't even list a 128Gb chip, so they are no help with the market price of these parts that are rapidly coming to make up the majority of NAND bits.
Making SSD sub systems allows for product differentiation on quality, endurance and performance.
All of this is happening while an unprecedented business opportunity is staring the industry in the face in the form of solid state drives.
There are about 550 million HDD sold per year between Seagate (NASDAQ:STX), Western Digital (WDC) and Toshiba (OTCPK:TOSBF), for revenue of about $40 billion, give or take. About half of those HDD units "want" to become SSDs at an average density of 192GB (50% 128GB and 50% 256GB) over the next 2-3 years. I figure the price necessary to get that done will be about $.65 per GB, or about $125 for the 192GB average drive. So the business opportunity to convert mechanical stuff (HDDs) to semiconductor stuff is 275 million X $125 or ~$35 billion. That's nearly as big at the memory business is today and about 10% of present total semiconductor sales. That opportunity is so big and so immediate that, I believe, it has to happen. How it happens and who leads it and where the capacity comes from are some big questions. The scale of that question involves the higher level question of whether the technology to do it will even be NAND.
I'm betting that it will be NAND and further betting that it will be 2D NAND based on the Intel (NASDAQ:INTC)/Micron (NASDAQ:MU) high K planar cell (rather than the wrap around cell used by other NAND manufacturers) because it has better performance and is significantly simpler to fab (simpler if you know how to use the high K materials.)
I think 3D NAND will come on in years four and beyond and will lead the price to $.50 per GB and below. That price will sop up another 150 million units, leaving the HDD business at about 100 million units of very high capacity drives.
To get this done will take over nine million more wafers of capacity than exists today… or a different technology. No new technology has the scale or current cost to replace NAND in the short term, so I guess it will be more fabs.
With a new fab taking three years from dirt to wafers, at good yield, how does this happen in that compressed 2-3 year time frame?
I don't know exactly how it plays out, but there will be constant multi-year pressure on the memory industry to make this transition. We already see DRAM capacity being converted to NAND production. If DRAM can't support $3000+ per wafer, it will become NAND production, thus supporting DRAM price in the long term.
I see about 2 million wafers of capacity that could come on in a short time frame. Micron Singapore has about 500,000 wafers per year to go to reach full design capacity. Micron's Rexchip facilities has another 500,000 wafer per year potential expansion to reach full capacity. This potential for low cost and short-term expansion represents a $3 billion sales upside for Micron. Other bits and pieces among the other players could be as much as a million wafers. Samsung will bring on another million wafers in China next year. Ongoing shrinks and mix changes will equate to another million wafers.
That still leaves the industry five million wafers short.
Other than Samsung, there are no facilities being built or planned. The new memory oligopoly seems to be working as oligopolies tend to work; control output and maximize price and return on investment.
The HDD industry is a good example of how an oligopoly maximizes profit. Look at the increase in share price for STX and WDC to get a taste for this. Unfortunately for the HDD oligopoly is that the SSD Oligopoly will eat them.
The biggest pure play here is Micron. With the acquisition of Elpida, Micron has suddenly become the "Hammer" in the memory business. Samsung produces more wafers, but just barely. About 25% of Samsung's capacity is consumed internally, which makes Micron the largest merchant market memory supplier in the business. By using the term "hammer" I am implying that Micron will have at least some control over memory pricing.
With the Micron build-outs mentioned above, their total wafer capacity will become about 8 million per year. At price equilibrium these wafers should produce average revenue of $3,000 each. Micron revenue will peak at $24 billion at 50% gross margins and 25% net margins. At equilibrium Micron should earn $6 per share.
Seth Klaman thinks enough of Micron to commit fully 32% of his Baupost Group to the name and David Einhorn's Greenlight Capital bought 23 million shares of micron in the third quarter.