"You don't need a weatherman to know which way the wind blows" - Bob Dylan - Subterranean Homesick Blues
With the dismal residue of our current political season settling in like a suffocating black fog, it was so good to get a breath of bracing fresh air in the news that Bob Dylan had been recognized by the Nobel Committee for his contributions to poetic literature. It's therefore only fitting that we use one of his famous mots to kick off a discussion of Micron's (NASDAQ:MU) NAND business in FY 2017.
I've been noodling an industry NAND capacity model for some time, and as I have gotten buried in the details of it, I've been stricken lately with the conviction that, as devilishly complex as this infernal business is, the big picture with the business is as clear as day. There is a big wind blowing in this business and no, we really don't need a weatherman to figure out that it is really filling the sails of the IM partners Micron and Intel (NASDAQ:INTC). I'm only going to talk about Micron in this article, but the dynamics of the business are mostly similar for Intel, so take that as you may.
All this couldn't have come at a better time for Micron because boy has it been becalmed in the NAND business. I have written about this on several occasions so I'm not going to dwell on how awful the whole segment has been for Micron. The facts are stark and incontrovertible. While it is undeniably true that Micron needs DRAM contributions, 3D NAND is the product that Micron must successfully build and sell at historically high gross margins in order to generate enough cash flow to fund all that will need to be done with XP and its successors. Right now, Micron planar NAND is sold at low margins on the spot market or in short-term contracts that leave negative operating margins draining the company's bottom line. Can that change? It had better if the company is going to survive, much less prosper.
The good news is that Micron's opportunity to fix the NAND business will happen in the midst of two historic transitions, both of which provide opportunity for Micron. The first of these transitions is from planar to 3D technology - the second, the seismic jump in market demand for NAND storage that will be needed meet the needs of the "digital enterprise."
Addressing the last point, here's recent data from Dan Maslowski, Citi's Global Chief of the Storage and Engineered Systems Group:
Everybody knows […] that the cost of storage, the amount of storage, is growing as a percentage of the costs of the data center - and some estimates are 30% a year, some are 70% a year […]. I was talking to (the head of a large storage organization in the financial industry) and he was saying that if we don't do something the cost of storage will eat up the entire IT budget […]. The data storage explosion is real and it's happening right now."
Regarding NAND demand (versus HDD), here's Maslowski again:
"The financial industry has models that show that the total cost of ownership of flash storage is less than that of spinning disks." […] The (financial) industry is making a huge bet on flash. I, personally, am making a huge bet on flash. […] We expect flash to be the default answer for the backing store going forward." - YouTube video of Storage Developer Conference Tuesday Morning Keynotes.
Later in that same conference (1:03:23 mark), Mark Carlson of Toshiba (OTCPK:TOSBF) - the SNIA Technical Council Chairman, referenced an Allied Research Survey which projected a 20.7% CAGR for Storage Spending (Servers and Storage Combined) in the years through 2022.
One final data point. Here's Samsung (OTC:SSNLF) from its Q2 concall:
"Demand for mobile storage such as eMCP and eMMC is surging, due to expanded adoption of high-density products over 32 gigabytes, especially from Chinese smartphone makers. Overall demand is expected to increase, thanks to increased demand for high-density solution products in major SSD and mobile applications, supply and demand is expected to become much tighter in the second half due to supply restrictions caused by 3D NAND ramp-up delays in the market. - SA Transcripts, Samsung Q2 Earnings Concall.
I have discussed this topic in recent articles, but it bears repeating that recent signs point to the idea that the inflection point in NAND demand is at hand. Let's review why:
- The Hyperscale Data Centers, both public cloud like AWS and private cloud entities like Citi and the big money center banks are driving demand for storage. On the same slide (16), Mark Carlson references an Allied Research Survey that already over 50% of all storage spending is driven by these hyperscale customers.
- Per Maslowski's comments above, this spending by the hyperscalers is accelerating at an alarming rate (from their perspective at any rate).
- This spending on storage is overwhelmingly weighted towards NAND.
- Content in the mobile device market is surging.
Regarding the first point, the transition from planar to 3D, here's a pertinent recent comment from Mark Durcan, CEO of Micron:
We want to get the planar NAND capacity converted to 3D as a top priority […] So my sense is that as we get more and more 3D NAND bits out into the marketplace, you will start to see that play out in terms of many applications demanding, yes, I want that 3D NAND bit. You can sell that planar bit to somebody else. Which I think eventually will lead to a situation where you see potentially oversupply of planar bits relative to 3D NAND and a divergence in the pricing of those bits. - SA Transcripts, JP Morgan Technology, Media, and Telecom Conference, May 25, 2016 (emphasis by the author)
Translated - planar NAND's long tail is about to get amputated. Why? Because the planar die is not as fast, not as durable, not as power-efficient, and (at Micron and Intel at any rate) its 3D cousin is considerably cheaper to boot. If Durcan is right, and I believe he is, the current planar NAND capacity, approximately 1.2 million wafers per month (wpm) of industry capacity, is a boat anchor that must be cut loose as soon as possible to allow the industry producers any chance at profit.
The other aspect of the shift to 3D that we must keep in mind is that IM partners feel they have a real and sustainable cost and quality advantage. Ultimately, competitive advantage in the semiconductor memory business is a function of die capability and cost. In the capability dimension, the IM partners assert a charge durability advantage due to the inherent characteristics of charge isolation in relatively large geometry configurations. This means that the cell can be relied to not leak charge and thus degrade over time. The charge trap implementations of the three other competitors cannot say this.
In the cost dimension, there are two elements that combine to determine ultimate cost per bit. The first of these is die density. In this area, the evidence available points to a layer generation (i.e. 32L, 48L, 64L, etc.) density advantage over at least three of the leading competitors, WD (SanDisk), Toshiba and Samsung. Here's a chart from TechInsights published in EE Times on Oct. 6 of this year.
Table 1. A comparison of die floor plan for SanDisk/Toshiba 15 nm 2D, 48L 3D and Samsung 48L V-NAND (Source: TechInsights).
For comparison, IM's 32L TLC 384Gb die density is 2.57Gb/mm2, meaning that IM packs as many bits into its 32L die as Samsung does at 48L. Assuming that IM fulfills its commitment to deliver the rumored 64L Gen device on schedule - the announcement being immanent and with volume production projected by mid-calendar 2017 - the two partners will be producing the densest die on the market at that layer count. Why is this important? Because at the end of the day, the supplier with the bit density advantage has the opportunity to generate more revenue from a given wafer at a lower overall cost, if its yields are equal to the other competitors.
Note that we said "opportunity." Here's where yield enters the picture. Whatever your die density, that die is not worth a jot to your customer if it's defective, right?
Here again, the IM partners assert an advantage, claiming that their yields are higher than the other suppliers. How do they know this process yield data with that information being the most closely held secret of these fiercely competitive semiconductors manufacturers? Well, unless IM has spies at work they don't know for sure but they infer that their costs are lower because they are the only 3D NAND producing entity that has claimed a cost-per-bit advantage of their 3D process over their current generation planar die cost. In IM's case, that advantage is 25% with the 32L TLC die, progressing another 30% with the Gen 2 die (rumored to be a 768Gb 64L TLC product). Let's compare their comments to Samsung's from their Q2, 2016 earnings call regarding their 48L TLC die:
"...it's difficult to compare the productivity apple to apple between V-NAND and planar NAND because of various factors…" - SA Transcripts, Samsung Q2 Concall
This is a remarkably weak claim from a vendor that is delivering its third generation of its V-NAND product and has never been shy about claiming process advantage for its various chip technologies. In fact, it is not difficult at all to compare the productivity of one process over another. Either you are getting more bits at less overall cost or you are not.
Where is WD/SanDisk in this competition? In a recent article, I quoted statements from WD's Q2 conference call that claimed the achievement of cost-crossover with its planar process would not occur until late in calendar 2017 and even at that did not project a specific cost advantage as IM has asserted with its claims.
The bottom line is that the data point to Micron having a strong and possibly durable competitive advantage with its 3D process. Not only does it have a density and quality advantage, it has a production efficiency advantage in terms of yield and thus cost. How might this play out?
The four production entities that comprise the NAND business are all looking at the same strategic game board. To review, here are the factors driving the grand game:
- Demand is exploding, especially among the private and public hyperscalers - and in the mobile space as well.
- Once 3D NAND is qualified and cost cross over has been reached, planar NAND is no longer tenable and must be converted to 3D as fast as possible.
- Samsung is the hands-down market leader with possibly half a generation process cadence lead over the other competitors. Despite that lead, it is still struggling to reach cost crossover with 16 nm planar NAND.
- The IM partners are the quality and cost leaders, with the cost advantage being as much as 20% by the end of 2017.
- WD/SanDisk will be a 3D laggard in terms of production and cost during the calendar 2017 time frame with their cost crossover point on the 48L TLC 256Gb die not arriving until late in the year.
- Toshiba is in exactly the same position as its production partner WD. It is delivering product (e.g. as Apple 7+ partner) but likely at no better than breakeven margins.
- SK Hynix (OTC:HXSCF) is behind, perhaps even more than WD. It is currently in production with a 36L product and have announced a 48L but when it will have volume production and cost crossover is unknown.
"The early movers, such as Samsung and Micron, are ramping up 3D NAND quickly, while SK Hynix and SanDisk/Toshiba are lagging," said Greg Wong, an analyst with Forward Insights. "The technology and yield learning is taking longer."
With most forecasters projecting 40%-plus demand growth and slightly less production growth for the 2017 calendar year, one might think that the prospect for overall industry revenue growth from what is likely to be a roughly in balance supply/demand environment in 2017 is probably on the low side. And you would be right. Initial forecasts in early 2016 were for basically no revenue growth in 2016 to very low growth in 2017 despite a possible shortage. While not addressing the issue of NAND industry revenue growth, Micron is on record in its Q4 concall projecting that overall NAND bit output in calendar 2017 will lag demand by as much as 10% - a bullish forecast that hints at much better outcomes for the industry. There are more explicit forecasts. One is from KDB Daewoo Securities that expects a NAND supply shortfall of 5 EB in 2017. If we connect the strategic dots, I think that there is an even more positive potential outcome in 2017.
Let's ground this discussion with two quotes, the first from Jim Handy of Objective Research from earlier this year:
"NAND flash vendors have converted some of their planar NAND fab capacity over to 3D NAND. If demand picks up, and vendors can't ramp up 3D NAND fast enough, OEMs could face product shortages in the second half of 2016 and perhaps beyond. "OUR OUTLOOK IS FOR A SHORTAGE IN NAND IN GENERAL," said Jim Handy, an analyst with Objective Analysis. "It all depends on whether or not people are able to overcome the technical barriers to get out 3D NAND. And it's really hard to predict when that will happen."
Here's Ernie Maddock of Micron spelling it out for us:
"I think it speaks to a really important reason why we don't believe the industry will be dramatically over supplied here as you go through this transition 3-D which is the first thing you have to do (to) transition to 3-D is take planar capacity off-line and when you take planar capacity off line to do the conversion there is not a - there is a period of time which is not inconsequential where you're actually reducing your bit output and then you begin rebuilding that as you implement that 3-D capacity and that comes online and begins to yield. […] So unless you contemplate an environment with their massive amounts of Greenfield3-D NAND capacity being built before the planar transition starts you know you're going to have a more orderly transition and what Micron experienced over the course of last few quarters relative to bit growth is not unusual, that's what happens when you make a technology transition." - SA Transcripts, (emphasis added by author)
So Maddock the practitioner and Handy the analyst are seeing two sides of the transition conundrum the industry faces. The first is overall process yield ramp difficulties, and the second, perhaps a more subtle but even more consequential point, is that while building greenfield plants would be easier, converting planar is the single highest priority facing the NAND producers.
Please bear with me while I get granular by working through the likely emerging scenario. Here is my estimate of industry wafer capacity at the beginning of this year:
Here is how I expect the industry will look at the end of 2016 in terms of thousands of wafers per month capacity as almost 25% of planar capacity is converted:
As cost cross over is finally reached in 2017, I expect another 40% of the existing planar capacity will be converted. The leader in the planar conversion wars? It's IM, hands down. Micron got an early start this year by converting what I believe is likely to be 60k wpm of the total 80k wpm of Fab 10N's 16nm planar capacity to 3D. The remainder of Fab 10N in Singapore will be converted in early 2017 as Lehi IM fab finishes its conversion to 3D XPoint.
Per Ernie's formulation, what would the result of the massive conversion effort? Here's what my model shows:
The table above is the result of applying a 50% penalty to the planar capacity migrated to 3D, meaning that the assumption is that a supplier converting 60kwpm of planar to 3D would be able to wring out half of the planar production they would have had if they had left the line in place all year. Arguably that factor could be even higher. For example, if we assume that the entire planar line must be torn down before 3D conversion can start then another 10 EB of capacity would be subtracted from 2017 production. Whatever the factor, the point is this: there is no way that industry production in 2017 will meet demand unless massive amounts of greenfield capacity are being built that the suppliers have not announced.
What will be the pricing dynamics of a supply situation like my model predicts developing in 2017? Very good to great - depending. Depending on many factors, among them being (in order of importance):
- Product qualification status
- Overall 3D bits available compared to planar
- "Trade" end-product mix (i.e. mobile vs. server, EMCP, EMC, DDR4 configurations)
- SSD product strength at the high end
- The ability of the overall market to moderate demand in lieu of bidding up NAND prices.
There is another factor and this is one that is not really getting the attention it should. Industry costs will not be declining for the half of the market supplied by SK Hynix, Toshiba and WD/SanDisk. This fact will tend to set a floor under price declines in the face of overall shortages. Why are costs going up? Because of the planar to 3D conversion imperative and because the "easy-money" of planar MLC to TLC bit cost reduction has been harvested. For Micron specifically, as we look at the four factors above I expect it to continue to lag during Q1 and Q2 fiscal 2017 because its 3D product is so relatively new, especially its mobile 3D NAND product and related eMCP packages. Early on I expect to see it sell a lot of "trade" product into the hyperscale cloud providers who will want to build out their "JBON" (just a bunch of NAND) storage infrastructure. This is where we will see its cost advantage really pay off. It should garner 40% gross margins, a far cry from its negligible margins received in the planar product arena. As the year goes on and its 3D offerings get qualified, we'll see Micron gain good share in both the data center and enterprise markets with higher margin SSD product. In the mobile arena, by Q2 we'll see design-in wins for the 3D product, a phenomena that will accelerate as the year progresses, once again garnering high margins. Bottom line, I expect Micron to double NAND revenues in FY 2017 over FY 2016 and deliver almost 40% gross margins.
My models show Micron, because it started the planar to 3D conversion earlier and bigger than every other supplier, will show capacity increases in the 50% plus area. Let's look at the last year of NAND sales and put these increases in context.
Here's how I expect to see Micron's business react to the overall shortage of NAND and the strong increases in qualified 3D bits as the year progresses. Why are gross margins increasing? Because prices remain flat and Micron's costs are steadily decreasing as the year advances.
The takeaway from the above is that Micron will be a big winner compared to previous results and compared to its other competitors, excluding only Intel. Why, you may ask, isn't the win a bigger one in terms of revenue and profit? After all, one might think that a shortage that big would result in actual prices increases for NAND. If the product were DRAM, this would be true. But not for NAND, whose demand curve is very elastic. In effect, I expect buyers to moderate their purchases and wait for the big production bump that will eventuate from the planar NAND conversions progressing and 3D generational density increasing.
The overall result of the win in NAND for Micron will be $1.50/share in NVM contributions to Micron's FY 2017 bottom line. Let's add in a roughly equivalent DRAM contribution and Micron should exit 2017 with roughly $17B in sales and $3B in net profits. (And yes, I am assuming DRAM capacity moderation on the part of the industry and in particular Samsung.)
Let's conclude. FY 2017 is going to be a great year of recovery for Micron's NAND business. To be very clear, the forecast of $1.50 of NAND profit puts me way out on a limb. Micron's own forecast of bit capacity growth and demand, both for the industry and for Micron itself, differs from the scenario I have projected. Current analyst consensus for FY 2017 for Micron centers on $1.70, way below my $3. Obviously, there are many elements that support my forecast that I have not discussed in detail. Clearly, I am assuming Micron executes well in ramping its 3D NAND Gen 1 and Gen 2 product and that the cost and quality advantage IM claims proves out. Inherent in this forecast are share gains for Micron that I have not explicitly addressed but will be substantial. My model projects a 4% bit share gain on the year with Intel and Samsung also gaining share. All this share gain will come at the expense of the "weak 3" - SK Hynix, Toshiba, and WD. Because I expect a two-tiered pricing market to develop with 3D product garnering price premiums, I expect Micron's share gain to be more substantial at 6%.
The biggest risk I see to this forecast lies in the market's reaction to flat NAND pricing in the face of slower capacity growth. Is the NAND market really as elastic as everyone seems to say it is? We're about to find out. My forecast of a $40B total market assumes some demand moderation and considerable planar NAND price degradation as the weak three try to maintain share with an inferior product.
What about 2018 and beyond? Next article.
Disclosure: I am/we are long MU.
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.
Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.