Only the paranoid survive." - Andy Grove, Intel
There are old pilots, and there are bold pilots, but there are no old, bold pilots. - anon.
The one thing you can count on when you talk to a long time semiconductor memory industry investor is a certain attitude of sang-froid in the face of the unending ambiguities that haunt this industry. There is always something that is threatening to ruin the profit rationale - some factor that could overwhelm the otherwise sunny logic of the investing moment.
Exhibit one today is the NAND industry that is entering one of the truly transformational periods in its 20-odd year history. 3D NAND is maturing, bringing with that maturation unprecedented higher densities and lower costs leading to new opportunities in the data center and in mobile devices. Driven by those new use cases, TAM is rising. Supported by robust demand and some capacity constraints, profits are rising as well. Not every company is benefiting equally. Samsung (OTC:SSNLF) is dominating with a 40% share of revenues and probably a greater share of profits. Below Samsung, there is some turmoil involving the next largest producer - Toshiba (OTCPK:TOSBF) - and as we have discussed at some length here and here, with that turmoil and a decided IM technology advantages comes the golden chance for Micron (NASDAQ:MU) and Intel (NASDAQ:INTC) to take bit, revenue, and profit share in the industry.
Oh happy, sunny days, right? Well yes, if we consider only the next couple of years, but what about 2020 and beyond? In this business, the PE multiple is always held hostage to the Janus-faced character of supply and demand. With the rise of IoT and the new opportunities in the data center, demand seems relatively certain. But what about supply? The potential of greater densities with ongoing technology advance alone scares some folks, but what is often forgotten is that the initial conversion from planar to 3D imposes a 40% wafer tax and each process advance thereafter subtracts another 10-15% from wafer supply. Taken altogether, when the industry finally converts the last planar wafer to Gen 3 (and in some cases Gen 4) in 2019, the estimated 1.3 million wpm of NAND capacity in 2016 will have shrunk to only 500k wpm.
The industry is adding 3D capacity of course - Micron and Intel have each built new 3D fabs and the other suppliers have added as well. The point is, though, that industry discipline has held very well. Alongside of that, we have the disruption of the planar conversion and problems with 3D yields, all of these factors combining to hold NAND bit growth in 2017 below 40% CAGR.
Depending on your crystal ball, leaving demand out of the equation for now (i.e. demand trends at 40-45% CAGR), supply dynamics in the industry going forward are a function of the following factors:
- The fab equipment vendors bringing etch capabilities to the table that enable wafer layers to exceed 64 in the 2018 time frame.
- String stacking working for stacks of three or four layers.
- 3D yields improving to equal planar yields (90% +) - something only IM brags about at this point.
- The Toshiba acquisition seriously delaying Flash Ventures' Fab 6 - or not.
If all of those factors are satisfactorily resolved, then there is a possibility that the industry could be in an oversupply situation in 2019 or 2020. If none of them happen, then 45% demand has a good chance to overwhelm supply by an amount that might exceed 15-20%. Reality should fall somewhere in the middle, so on balance I like the odds for the IM partners. Samsung will take its 40-share, come what may, but Micron and Intel can take a lot of share from Toshiba, Western Digital (NYSE:WDC), and SK Hynix (OTC:HXSCF) assuming that nothing else changes that affects the overall supply equation. Personally, (as I have written) I feel that demand will exceed the current estimates, but even if my sunny forecast doesn't come true, the net of the situation is that the health of the industry is firmly in the hands of the suppliers. Even if all the technology enablers turn out to come true, just a little restraint on the part of the suppliers will keep the industry nicely profitable.
It is against this generally positive backdrop that the recent news of Yangtze River Storage Technology's (YMTC) mega fab coming online in 2019 arrived. The most recent report from DIGITIMES, dated March 15, was basically a PR piece for YMTC originating in the office of its President and CEO, Simon Yang. The key paragraph:
YMTC is engaged in the development of 32-layer 3D NAND flash chips, which will be in full production in 2019, according to Yang. The company aims to catch up with the world's leading memory vendors, in terms of technology, by 2020, Yang noted.
So who is YMTC, you might ask? A recent Forbes article sums up the pertinent facts very well:
In 2014, the Chinese government announced that it would invest $150 billion to expand locally made integrated circuits in the domestic market from 9% to 70% by 2025. And it created the country's largest chip maker, Yangtze River Storage Technology (YMTC), in July from state-owned companies Tsinghua Unigroup and XMC.
All considered, this is obviously a portentous report - one that is made all the more ominous by this news related to the fab's scale from Simmtester.com:
Yangtze River Storage recently broke ground on a new memory semiconductor fabrication plant located in the Donghu New Technology Development Zone, Wuhan (Hebei province, China). Three 3D NAND flash production lines will be set up at the plant with volume production slated to kick off in 2018. The plant is looking to produce 300,000 12-inch wafers monthly by 2020.
So if you take what Yang is saying at face value, not only will the industry be facing a new competitor in 2020 that will be operating at a scale that is larger than any of the current players except Samsung, but also has commensurate technology and has the backing of unlimited State capital to boot. How can this be possible?
Well, while it is true that the CPC is deadly intent on throwing tons of cash at this effort, the first two points above are a fantasy. The press release, produced presumably by Yang to please the gullible CPC masters who are funding YMTC, is total nonsense. Let's consider the implications and do some math to get a better idea of what is actually happening here.
First off, keep in mind that while YMTC has partnered with Spansion (now Cypress Semiconductor (NASDAQ:CY)) to create 3D NAND product, its only experience with fabricating Flash memories has been through XMC acting as a foundry for Spansion's NOR products. Nothing in YMTC's antecedents is anything close to the massive scale and complexity of today's state-of-the-art 3D NAND fabs.
Granted, Yang is a credentialed semiconductor technology veteran and we can certainly assume that he has recruited an experienced staff, some of whom will most probably have been recruited from the various memory suppliers. The point is though that his team (a staff that will eventually count in the thousands) has zero experience working together to accomplish the myriad of complex steps necessary to produce NAND wafers at scale. Let's give them the benefit of the doubt in that the equipment suppliers can certainly help educate YMTC in the basics of 3D memory technology, but this is what you might call a steep learning curve indeed!
Mark Lepedus in an article in Semiconductor Engineering dated January 17 of this year put it this way:
But 3D NAND is a difficult to make. YRST [sic] is working on its own 32-layer 3D NAND device, according to sources, but the yields are low. It's unclear if it is still developing the 3D NAND technology from Cypress/Spansion. Regardless, China will struggle in 3D NAND. "Looking at the struggles that the current NAND players have had in developing their 3D NAND, it may take longer than expected (for YRST to make 3D NAND)," Forward Insights' Wong said.
Second, even assuming that it manages to produce a serviceable 32-layer product in 2019, it will face the long process of customer certification that can take six months or more. So let's make a very big assumption and agree that the Chinese will be as good as their competitors in ramping the first fab line, producing certifiable dice, and surmounting the learning curve to achieve 80-90% yields on their 32-Layer product by early 2020. What will the market look like in that timeframe? Let's make a further assumption that either the etching and deposition technology has surmounted the pitch obstacles that created the current 64-layer barrier or the string stacking issues limiting the number of stacks have been resolved, thus removing any real barriers to cost-effective scaling of 3D NAND. Given these assumptions, YMTC's competitors will likely be at or close to production yields on their Gen 4/5 128L processes. Die density, which is a surrogate for cost, will be on the order of 9 Gb/mm2 compared to a little more than 2.25 Gb/mm2 for YMTC's Gen 1 32L product.
The bottom line is that the industry incumbents will be selling a die that has considerably less than half the cost of the YMTC product. But more than cost is at issue. Because of the smaller and denser die, the incumbents will be able to package the die into smaller form factors that are still considerably denser than the YMTC product. This provides the OEM with much more value, allowing the customer to create innovative end products that would not be possible with the larger, less dense dies. Think, for instance, of the Chinese smartphone OEMs like Xiaomi (Private:XI). How will YMTC overcome these enormous disadvantages, even if it can manage to get good product out the door, which is by no means assured in the 2020 timeframe?
There will be two principal ways that YMTC will attempt to overcome these crushing competitive circumstances. First, it will (literally!) give its product away. It will have to - why else would a customer use it? Remember, YMTC is not a business in any normally accepted sense of the word. The shareholder Yang answers to is the Communist Party of China (CPC). This "company" is a creature of the State, and is not concerned in the least with "profits" or cash flow in the normal sense of the word. The only bottom line for YMTC is CPC's goal of taking a 30% share of the global memory business (in bits) by 2030.
The second strategy for selling its product also flows directly and organically out of the Yangtze River's status as a state-owned company. Customers will be compelled to buy its bits. Here's a portion of its PR release in the DIGITIMES article that hints at this strategy:
"China consumes as high as 55% of the total memory capacity. With the strong domestic demand, and financial support from China's central government, YMTC should be able to enhance its competitiveness against the current major memory players, Yang said."
The CPC will have absolutely no qualms about imposing domestic content laws on memory consuming companies, foreign and domestic. Other tactics will include banning outright the sale of foreign product from certain sectors and industries. These are just two among many tactics that China employs against foreign firms. The following is from the US Department of State's 2015 Investment Climate Statement: [emphasis by the author]
Meanwhile, China's revised Foreign Investment Catalogue, released in March 2015, fell short of meaningful progress toward broader market access for foreign investors. Poor enforcement of intellectual property rights (IPR), the forced transfer of technology, and systemic lack of rule of law are additional concerns. Over the past year, China released new policies to exclude foreign technology from various sectors, including: proposed banking sector guidelines that could disrupt supply chains and increase firms' operating costs, a draft counter-terrorism law which imposes onerous requirements on foreign technology firms, and an intrusive national security mechanism that could be used to hinder or even block foreign investment.
So, where does this all lead and what will be the impact of the China's memory initiative in the medium and long run? Will any of the Chinese policies really work? In my opinion, no. Domestic content will fail miserably because the competitive disadvantage of YMTC is just too great. Can we imagine Xiaomi, for example, de-rating its smartphone specifications to accommodate a 4X memory density disadvantage? Apple (NASDAQ:AAPL), Samsung, and the myriad (by that time) of Indian competitors will feast on their lost market share. Will Alibaba (NYSE:BABA), Tencent (OTCPK:TCEHY) and Baidu (NASDAQ:BIDU) adopt YMTC in their storage architecture? The thought is absurd. The fact is that Yangtze River, in the great majority of use cases, will not be able to give their product away. It will simply cost the customer too much to use it.
What about industry sector mandates - for example the banking industry? That is less certain in terms of product uptake. Since the banking industry is also state owned, the politically smart move may be to "buy" the Yangtze memory and use it for landfill (all the while sourcing the incumbents' memory through a use of any one of numerous political covers). Beyond customer workarounds like that there is the inevitable trade crisis that such nakedly mercantilist trade policies will provoke with the rest of the G-20 world. Whether it be a new administration or the beginning of the second Trump term, no American President is likely to ignore Chinese actions as extreme as these. In the end, the CPC will decide that it is simply not worth it.
In the end, the real reason China, Inc. will give up the NAND business is because they will never catch up to the incumbents. For each "new gen" of the YMTC memory, there will be a "gen next" for IM, and for Samsung, and Flash Ventures, and for SK Hynix.
So what of China? Will this effort devolve into a destructive trade war that balkanizes the world economy and, in so doing, deprive the incumbents of the 60% of global demand that is the China market? No, that's not going to happen, and there are two principal reasons for that. First, China can and will build a significant domestic technology vendor capability (centering on IoT devices) that will demand access to best-in-class memory. There is, after all, no substitute for it. But second, and most of all, because the memory supplier market is not now, and has not been restricted to non-Chinese firms. It is restricted to companies with the appropriate intellectual property (IP).
Once again with feeling, the issue with memory is IP. China is not going to get NAND IP - but they can create their own future with other types of memories. They can create their own IP. China needs to invest in next gen memory R&D and focus on commercializing that. Mr. Yang, if he is still piloting YMTC after the bloodbath in Chinese NAND, should be able to convince even his CPC masters of that simple truth.
Let's close. There are many reasons for this memory investor to be long Micron, and there are a few good reasons to be concerned. China isn't one of them. In the end, the memory industry oligarchs are paranoid enough to survive and flourish. Mr. Yang may be a bold pilot, but he will not live long enough in the memory business to be an old pilot.
R.I.P. and adieu.
Disclosure: I am/we are long MU, WDC.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.