- How these three companies saved Elpida.
- Who won and who lost.
- Immediate alpha opportunities.
I started out writing an article I intended to title Micron (NASDAQ:MU): from zero to hero in 18 months. As I developed the timeline and players, I realized that this is a longer story, involving more than just Micron.
The starting point for the story is January 2006 when the Intel (NASDAQ:INTC) Micron Flash Technologies (IMFT) was formed for the purpose of producing NAND memory. The output of the JV was to be split between the two owners. The ownership was 49% Intel and 51% Micron. In this arrangement Intel would not report IMFT results on their top line where it would destroy the high Intel gross margins.
The two companies had different motives for creating the JV; Micron simply wanted a larger footprint in the NAND market and could use Intel's help. Intel needed a process development fab where processes could be developed and moved the microprocessor business when appropriate and they could also see the potential for SSDs (Solid State Drives) made from NAND memory. Remember, this is back in 2006 and Intel wanted to be involved in SSDs. I didn't know what a SSD was back then. Those early SSDs must have been small, cumbersome and expensive, since the largest NAND chip was 16Gb at the time.
The original fab was in Lehi, UT and that fab is still the primary fab for developing the new NAND processes for the JV.
Around 2008 the JV built a 100,000 wafer start per month fab in Singapore, which promptly went on hold for two years while the Great Recession played out.
The Great Recession began a slow recovery and IMFT - Singapore began to ramp in early 2011.
The Singapore fab is a real-time example of why the memory business "is different this time." The Singapore fab was brought on line in about 20,000 Wafer Starts Per Month increments and is still only about 60% full. That kind of supply control will be the rule, not the exception in the future of memory business.
I wonder if it is generally understood that the Intel/Micron JV has created arguably the best and lowest cost NAND process in the 16nm node.
During the Great Recession something else was going on, Elpida Memories, a combination of three excellent Japanese memory companies, NEC, Hitachi, and Mitsubishi, was being ground toward bankruptcy by the twin forces of a high Yen and insanely low DRAM prices.
In Apr of 2011 the drama that is the point of this article began in earnest with Apple (NASDAQ:AAPL) suing Samsung (OTC:SSNLF) for...everything they could think of. At this time, Samsung was supplying something like 80% of the value of the iPhone, everything from memory to application processors to displays and batteries. Obviously, Apple had to stop feeding their worst nightmare as a competitor. Apple began to do just that; batteries and displays went away, memory was multi-sourced and the three year rumor mill began to churn on about who would build the Apple A chips. Changing suppliers on the A chips might not have been the super urgent problem that it would seem, since Apple had invested in the Samsung Austin fab that made the A chips, so Samsung would find it legally "unpleasant" to cut Apple off from the supply of A chips.
By mid-2011 it was obvious that Elpida would go bankrupt, it was just a matter of time. Now, if you are Apple, the prospect of Elpida going away was as scary as a bad horror movie. Micron supplied practically no mobile DRAM, leaving Samsung by far the largest supplier of this critical component. Hynix also built mDRAM, but was having a near death experience of its own at the time.
Elpida had to be saved for the very survival of Apple.
Now for my speculation:
Apple had to reach out for help on mDRAM. I think there is no question that they contacted Micron and Micron probably contacted their JV partner, Intel.
The three way discussion probably went like:
Apple: We're in trouble, we want to ditch Samsung on mDRAM, but Hynix and Elpida are on death's door. We can buy them, but we don't want to be in the memory business, and they would extract a gigantic amount of money from us.
Micron: We could wait for the bankruptcy and try to buy them, but we aren't in wonderful shape ourselves and don't know whether our balance sheet would pass muster with the Japanese bankruptcy court and, guys, the DRAM business really sucks right now.
Apple: If you could gain position on buying Elpida, we could load the company with orders that would, at least, stop the bleeding.
Micron: Great, now what do we do about the balance sheet problem?
Intel: Well, we do have the assets in the IMFT JV. We have our piece on the books at $1.3 billion and our share would probably appraise at $2.5 billion. What if we did some flaky deal like sell back most of the assets for…say $600 million, and you keep $300 million of the cash, "just in case". We can make up some goofy story about an "expanded JV." We can bury the loss in our investment operation. Your balance sheet would pick up about $1.9 billion in phantom assets and make your bid on Elpida more believable.
A,M,I: OK guys, here's the deal: Intel will do what they just described, Micron will position to make a favorable bid on Elpida and when you are known to be the sole negotiator on the buyout, we will load Elpida with orders for mDRAM, so we can dump Samsung.
I believe that some form of the above meeting and agreement took place to save Elpida for Apple and keep Samsung from screwing the entire industry on memory.
The Elpida acquisition closed on July 31, 2013.
So, as Mark Adams mumbled in an analyst meeting, Micron does a step function in sales from $8 billion and losing money to $16 billion and hopefully makes a profit.
Who could have known that the price of DRAM made a bottom on December 2, 2012 of $.84 for the 2 Gb chip and began an increase that peaked (so Far) at about $2.40? Of course Apple made the Elpida mDRAM instantly profitable. Oh, there is the little matter of the Hynix fire. And all this time DRAM cost are coming down and the transition to 4Gb becomes near complete, with the associated increase in gross margin that density transitions always bring.
Investor relations presentations mentioned a low on PC DRAM of -50% margins, so that would make the cost of a 2Gb PC DRAM chip at that time of about $1.10. That would make today's pricing for that part a 40% margin deal, but, of course, the cost has continued down so the margin on a 2Gb chip is probably positive 50% instead of negative 50%. When we consider the cost of the now mainstream 4Gb part to be substantially less than 2X the cost of the 2Gb, I'll guess at $1.80, the contract resale of $3.50 generates a gross margin on 4Gb chips of 48%, that's for PC DRAM chips; other markets are at least as good.
So, in summary, Micron has really gone from zero to hero in about 18 months of intensive activity, Tim Cook at Apple has to be sleeping better with life breathed back into Elpida, the workers at Elpida should be happy, the Japanese government should be happy, Japanese companies will be buying Elpida/Micron parts to the exclusion of anyone else. The SSD opportunity is now obvious and will provide a growth path for Micron (and Intel) well into the future. I suppose Samsung is a little miffed, but who cares.
The only loose string to this whole deal is with Intel. Intel participated in the "Saving of Elpida" with the crazy JV deal (crazy like a fox), thus removing Apple's behind from a very hot fire. What do they get?
The Apple "A" chip fab business? That would be a win for Intel and a big win, again, for Apple. And apple could keep the business in the US, where they could suck all those cool American fab guys out of the Samsung Austin fab. Naww, everyone knows that TSMC (NYSE:TSM) has the Apple business with their wonderful 20nm planar process, right? Of course, TSMC was a big help in the Apple memory problem and should be paid off for that...yeah, right.
I just love it when a plan comes together.
Buy Micron real hard and soon, Intel pretty hard for Jan. 2015, and Apple, why not? Sell the HDD duopoly of Seagate (NASDAQ:STX) and Western Digital (NYSE:WD) on the first indication of decreasing unit volume. Samsung is too diverse to play.