Magical thinking is a form of reasoning that learns causative relationships through correlation alone. Science and the scientific method are designed to elucidate causal relationships through careful controlled experiments; magical thinking, given a correlation with an observed effect, pulls a causation out of thin air.
We look at the NAND business as a business that can have significant growth, but also has some interesting demand elasticity properties. - Mark Durcan - Credit Suisse Technology Conference Dec. 3, 2014
... we think with 3D it will be the tipping point that will eventually allow the PC space to convert 100% to SSDs. - SA Transcripts, Micron's Ernie Maddock - Raymond James Technology Conference 2015
In the last two articles, we have taken a hard look at the NAND business in 2016 and beyond. We have concluded that Micron (NASDAQ:MU) will see only a little benefit from the business in 2016, but can potentially harvest revenue and profit gains in 2017 and beyond if demand growth exceeds 35%. Intel (NASDAQ:INTC) is in much better shape because of its dominance of the data center. Because of that dominant position, Intel will likely see strong revenue and profit gains in 2016 and beyond into 2017 and 2018, assuming adequate demand growth. Depending on the overall market conditions, Intel is in a strong position to be the primary beneficiary of the innovative and cost-leading 3D NAND technology that the IM partners are now bringing to market.
But what will those market conditions be? Questions regarding the economic viability of this new generation of 3D NAND product inevitably devolve to the question of demand. It would seem obvious that for a market aspiring to overall revenue and profit growth, demand growth must exceed the rise in productivity. In this case, if the capacity increases in the new 3D technology fall in the 35-40% per year range, overall market demand must exceed that level for there to be gross margin expansion.
As we have seen, however, market growth forecasts from Samsung (OTC:SSNLF), Micron and Intel diverge sharply, with only Intel projecting explosive growth - specifically, a tripling of the 2015 TAM ($31B) over the next three years. How can Samsung, Micron and Intel look at the same world and Samsung project just 25% YOY growth through the end of the decade, Micron 35-40% and Intel project a tripling of the TAM to $78B in 2018? Think about that for the moment. It's not as if these forecasts are coming from some pie-in-the-sky think tank guru (though those forecasts differ greatly as well). These forecasts are coming from the industry itself - from the people who one would have to think are closest to the ground as it were, who have the most to gain and the most to lose. These guys have a major skin in the game.
Intel's forecast has the virtue of being congruent with its view that NAND and 3D XPoint are paradigm-shifting technologies whose demand profiles vary on a predictable basis as expressed in Geoffrey Moore's "Crossing the Chasm" formulation. Ergo:
In Intel's formulation, 2017 marks the point of inflection when demand ramps sharply with the "early majority" buyers adopting the technology.
One thing seems to be true - both Samsung and Intel are acting as if they believe their projections. Samsung has delayed the second phase of the Xian plant expansion and has doubled down on a high-end SSD client strategy with its 950 series announcement centered on its NVMe controllers. Intel's Dalian announcement is perfectly congruent with its bullish strategy and is politically convenient to boot.
What are we to make of this? Micron and Intel are now united in their confident view of their emerging product advantage, but part ways regarding overall market growth projections. Western Digital (NYSE:WDC) must think that 3D NAND has potential - surely it is not buying into Samsung's market forecast in making what is perceived to be a rich offer for SanDisk (SNDK)? Another bullish indicator - the latest market data from Seagate (NASDAQ:STX) seems to support the idea that the market is finally reaching an inflection point with SSD sales negatively impacting the sale of high-end HDDs.
If you take the time to read the various projections of NAND demand growth, you will find many different opinions and shades of emphasis on the subject, but there is one thing that almost everybody (ex Intel) seems to agree on: NAND demand is "elastic." What is meant by this term is that demand will vary with price. In one way or another, every prognosticator will link various levels of cost declines to increased NAND adoption (Confession - I have been guilty of this as well over the years). The holy grail of this analysis is "cross-over" - that time when the cost per bit of raw NAND becomes less than an HDD bit. Until that happy day arrives, it is universally agreed that NAND demand will be driven by displacing HDD in high-end applications (the 15K and 10K drives) where HDD bits are relatively more dear due the derating that occurs to improve the performance of the drives. Even in this area, the justification is wrapped in a Total Cost of Ownership (TCO) analysis that implies that SSDs are worth the investment only on a dollar-for-dollar cost replacement perspective. Here's a typical graphic seen in many of these projections from Gartner:
Micron is particularly fond of the "NAND is elastic" thesis. In one way or another, we hear the "elastic" argument being made in every major presentation. Unfortunately, for us, investors, who are trying to fathom the opportunities and risks in Micron's business, MU has never been good at explaining exactly how its business will benefit from future NAND cost reductions. This is particularly annoying these days because it is trumpeting its supposed 3D NAND cost advantage over its competitors.
Micron has company. No one, it seems, has any idea of how much demand will vary with price. There is a very good reason for this.
It doesn't. Welcome to the world of "magical thinking."
On the contrary, NAND bit demand increases are a dizzyingly complex function of multiple factors encompassing flash form, function and density advances, industry competition dynamics and business model, technology packaging innovation, application software function and age, company IT culture, the pace of industry standards, the pace of price declines in hard drives, and finally cost/price advances in NAND itself. Let me be clear - I'm not saying that price is not a factor - it certainly is. Price reductions of NAND flash appliances certainly are a good thing. What I'm saying is there are so many other factors which are more important than acquisition price that price per se is a very poor indicator. Bottom line, demand is not so much "elastic" but "eclectic," and cost/price declines are simply one factor of many that determine overall demand patterns and growth. Price alone is not a terribly good metric for modeling demand changes.
No better example of this magical thinking is Mr. Maddock's comment (quoted above) that 3D cost improvements will mean that the days of HDD in PCs and other client devices are numbered. Why should this be so when hard drive innovations are maintaining the cost ratio of HDDs to SSDs at a factor centering on 1 to 10 on a cost/bit basis? Do you follow Western Digital or Seagate? If so, you have probably seen this slide a great many times:
Lately, there have been some detractors (most notably Toshiba (OTCPK:TOSYY)) that doubt the ability of the HDD industry to continue to innovate at the 40% CAGR rate shown above, but this doesn't really change the overall strategic cost differential situation through the balance of the decade. HDD $/Gb has, relative to NAND flash, remained stubbornly consistent over the years at roughly 1 to 10 and that situation is changing very slowly. So how is it again that price elasticity is driving NAND demand growth?
Now, let me be the first to say that the fact that Maddock is using magical thinking doesn't necessarily mean he's wrong. Despite the ongoing price differential, I believe Maddock is (mostly) right. My MacBook is configured with SSD, and there is no way I would ever go back to HDD. But my decision has nothing to do with price. PCs will eventually ditch spinning rust, but the reason that will happen has much more to do with IT model changes (the cloud requiring less local storage) and device maker design and packaging innovation (lighter and thinner and more durable matters more than capacity) than cost.
Once we move from client devices to the data center, we enter a whole new realm of adoption rationale. Here again acquisition price is simply a smoke screen. In this realm TCO (Total Cost of Ownership) rules the acquisition process. And here again, we are entirely dependent on the specific use case in a specific industry within an individual corporate IT culture. Bear with me while I use an outrageous analogy that hopefully will bring this discussion into sharp focus.
Let's assume for a moment that companies are in the freight hauling business and need to make a capital acquisition to haul the freight. Companies find that they are being offered two fundamental choices. They can buy a dray wagon pulled by a six-horse team (an HDD) or they can invest in a diesel-powered tractor-trailer (the SSD). We will stipulate that the new diesel rig is 10x the investment cost of the alternative.
Our companies face a decision: they can stick with the dray wagon approach or they can go with a new state-of-the-art diesel-powered tractor-trailer. Now, from a pure technology perspective, this analogy is unfair to NAND, but be that as it is let's follow through with it. On one level the decision is easy. The dray wagon/horse combo is 10X less expensive from a cost of acquisition standpoint. So it's clear, right? Our metaphorical companies will always buy the dray wagon.
There are in fact some use cases we can imagine where buying the dray wagon is a rational choice. Maybe I'm only hauling the freight 10 miles and my customers don't particularly care when it arrives. In the IT realm archiving is an analogous activity. Up-front cost rules the acquisition logic. Hard drives win. $/GB stored is the ruling metric. Let's say, however, that I want to change or augment my business model and offer freight service to more distant customers who care greatly about speed and flexibility of service. In this realm, the rules of the game have changed, and all of a sudden, there are many more factors to consider than the simple cost of acquisition. This is the world where our metaphorical diesel conveyance shows its value.
Taking this one step further - in the case where our adventurous customer has decided to expand his business to points distant from his current market - let us now imagine the acquisition competition between the dray wagon and tractor-trailer salesmen. Currently, the price between the two is $15,000 compared to $150,000. The tractor-trailer sales rep, hearing from noted industry consultants that tractor-trailer demand is "elastic," goes to his customer and says proudly that he has dropped his price to $130,000. If we are to believe the industry gurus who ceaselessly plot the cost decline curves in storage, this price drop is determinative. The tractor-trailer sales guy will win the deal based on his new lower prices.
It is clear how ridiculous this is? Hopefully, I have not offended you with this simple-minded analogy, but I think this brings the demand question into sharp perspective. In the example above, the customer could care less about the price differential between the two choices. His decision is driven solely by his business-computing model and his desire to change it in order to take advantage of the opportunities that are afforded by the change. The second and third order effects of the decision relate more to revenue and profit than cost. On the one hand, the customer can stick with his old business model and buy the less expensive alternative or he can take advantage of a highly value added transformation in his business that is directly facilitated by his choice of the tractor-trailer. "Price" has nothing to do with it. This is exactly the position that NAND is in today's business and IT environment. Indeed I argue that by the time the second generation of 3D NAND arrives in volume in 2017, this computing model transition will happen regardless of the ongoing price decreases (Which in any case are being matched by HDD).
So does acquisition price never matter? Of course not. I'm not privy to studies that show conclusively what applications NAND has been used for, but we have broad anecdotal evidence that the fertile ground for NAND has been "hot data," meaning those applications (generally IOPS intensive) that are very sensitive to latency. Because $/IOPS is a more useful metric, these applications are highly value added, and HDD companies have bent over backward to accommodate the liabilities of hard drives by short-stroking them and offering 15k RPM drives that can offer approximately 4.3 ms of latency best case in service to the application. In our cartoonish example above, imagine a new class of customers - a very small class - in our town 10 miles away and the dray wagon manufacturers responding with a new type of smaller, lighter wagons. They can't carry as much freight, but our horse teams can pull them fast enough to meet a few customer requirements. Enter our SSD sales rep who now responds with a tractor-trailer that is slower and less reliable than the previous model (think TLC) but can still carry more freight on faster round trips. It is here that price matters. The customer is still in the same business - no paradigms have been changed. He is still in the same 10-mile HDD box.
Before we leave this simple-minded analogy, let me acknowledge that there are several other good reasons to buy a dray wagon over the tractor-trailer. One of the most important of these has to do with state of the road infrastructure. Let's imagine the decision is being made in an environment where there are no roads beyond our 10-mile box. Our NAND tractor-trailer doesn't do us much good there, does it? Taking this one step further, let's imagine a customer who wants to enter the overnight package delivery business and equate 3D XPoint to a 747 freighter. Let's now stipulate that in our mythical world there are few airports with runways strong enough to handle the plane. This is the situation for prospective buyers of 3D XPoint within the current software model. And here again the decision the buyer makes has very little to do with price.
The central factor delaying the shift from HDD to NAND (and beyond, to other types of non-volatile memory (NVM) like 3D XPoint) is the 60-year-old von Neumann computing model, a corollary of which is that "memory is fast - storage is slow." Not only have three generations of IT folks grown up in this paradigm, but also almost all of the existing software (system and application) has been written by developers who have that tenet tattooed on their brain. The new generation of storage class memory like 3D XPoint is the metaphorical equivalent of our freight-hauling business hearing that a whole new world has been discovered beyond that town 10 miles away. Suddenly an uncharted universe of opportunity has been revealed. All they have to do is abandon the dray wagon and horses and invest in these new-fangled tractor-trailers and jet freighters. It is literally that extreme a change. And yet that is what is happening. And yeah, you guessed it. Price has nothing to do with it.
I hope that makes sense. Just as it does in so many aspects of life, leaving magical thinking behind exposes us to the reality of our situation with all its risks and opportunities. I believe this ongoing disjoint between magical thinking and current reality explains the chaos evident in the forecasts. Reality is clearly so much more complicated than single-factor "price is determinative" magical thinking. To the extent that there is rough consensus at 35-40% CAGR, (Gartner, IDC, Micron, SanDisk), it is as if our forecasters are blindfolded and, given the job of identifying the elephant, have groped aimlessly and inconclusively until congregating at the elephant's trunk where they find they are in general agreement. We are relieved to hear that the bulk of them are unanimous in their opinion that it is a snake (One has to wonder if the NAND demand forecast they have agreed upon is any closer to the truth).
Now let's be clear about one thing: I'm not saying that all industry forecasters are in the thrall of magical thinking (Indeed that is why I think we see such variability in many of the forecasts). I'm saying that up until recently Micron has been and it has encouraged us, investors, to be as well. Why is that? I believe it is because it has retained its engineering-driven business culture. I can hear the protests now from our friends in Boise who will point to numerous slides and commentary on how the company has evolved from being purely product driven to recognizing that the memory market has evolved to be more specialized and use case/market specific. And indeed Micron has tailored its DRAM business to be more market specific. NAND is a whole 'nother animal, however, and it is here that its lack of investment in marketing resources has caused it no end of grief (See exhibit 1 - the ongoing planar NAND TLC debacle). Its engineering-driven culture has yet to make the needed transition to being "marketing" driven and this has blinded it to market needs.
I have written about this topic before, and I encourage you to read it because it will provide much more context for this important subject. The key point to be made is that there is only one way for Micron to completely free itself from the perils of its NAND magical thinking and that is to invest in its customers and partners who use and make a business from its NAND product. It must become a marketing-driven company.
The good news here is that it has begun to make this vital transformation. The best evidence of that we have seen to date is the recent webinar given from its new Austin "Storage Innovation Center." While the production values of this event could have been much better, the content was spot on. Micron invested in this center to provide a working laboratory for understanding real world customer and partner opportunities in converting applications and operating environments from HDD to SSD. Its stated objective - "reducing risk and complexity while increasing predictability and performance" in the enterprise's transition from HDD to NAND flash and other forms of NVM. Here customers and partners can find a safe and convenient environment to explore the benefits of the new data-centric computing paradigm. This is exactly what is needed to reassure the slow-moving, hyper-conservative IT management teams whose first instinct on hearing of our new metaphorical world beyond the 10-mile box they live in is to cower in the corner and mumble darkly that "there be dragons" out there
Much more could be written about this presentation (The Next Platform's piece is excellent), but one of the important elements to come out of Micron's presentation is its discovery that paradigmatic change is rarely one-dimensional. Take for example Micron's VP of Storage Solutions Eric Endebrock's description of the "unexpected results that surprise you" when Micron collaborates with customers to convert significant customer applications and platforms from HDD to SSD. For example, a NAND flash hosted Hadoop works even better in a virtualized environment than on bare metal in a single server. Micron expected to show the customer better performance through the use of flash, but it did not expect the virtualization benefits that come from ease of administration. This is the "win-win" that comes from really knowing your customers and their needs. Only if Micron continues down this path, can it hope to build a profitable and growing business in NVM.
What does all this mean for the future? Should we be investing in Micron, Intel or any of the flash producers now?
Hopefully, Intel's forecast is closer to the truth than Samsung's, but let's be clear: if Samsung is right with its forecast of 25% CAGR in NAND, it will make little difference how sophisticated and differentiated any company's product strategy is: nobody wins - by 2017, products will sell at cost and sometimes below cost. If Intel is right everybody wins - they all will make money, whatever the competitive merits of their 3D product or NVMe controller IP. But some will make more - a lot more - and the biggest winner of all is likely to be Intel.
There is no bridging this gap. On this point, the world turns. I personally cannot believe that NAND flash adoption won't reach the inflection point and growth won't accelerate to levels well above 40% per year. But I also wouldn't recommend that any one invest in any of the NAND producers based on an assumption that it will happen soon. Doing so at this time requires a massive leap of faith that is simply not justified by the facts before us. I'm not going to do it, and I wouldn't recommend that you do either. Although we have some intriguing and hopeful evidence that the turn has begun, it is still too early to be placing large bets on the flash market. This is because 2017 will see a flood of new capacity coming to market that will drown the industry in excess capacity if demand does not ramp up significantly above 35% CAGR. This is the dreaded "nuclear winter" scenario that observers like Jim Handy have predicted for 2017 and 2018. It is to that topic that I will turn in my next article.
Demand must rise as we exit calendar 2016. One thing for sure. The move to accelerated NAND flash adoption will not be gated by acquisition price or ruled by $/GB metrics as the "NAND is elastic" catch phrase implies. The dragons we fear that lurk beyond our HDD box are not dragons at all - they are opportunities for increased revenues and profits that come from new business and operating models that are derived from the enhanced agility, capability and speed that NAND and new generations of NVM offer. Investing in customers is the only way that Micron and the other flash providers can get a firmer handle on the true dimensions of NAND demand. Micron has begun to do this. We can only hope that its efforts build rapidly on the Austin initiative. It will end up spending more on the SG&A expense line, but the return in materially higher margins will be worth it.
Disclosure: I am/we are long INTC.
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.