May you live in interesting times." - old Chinese curse
About the time of the Western Digital (NYSE:WDC) acquisition of SanDisk in May of 2015, the NAND industry as a whole joined the 3D process competition with Samsung (OTC:SSNLF). Since then, all the companies have been deep in the throes of this most challenging conversion from planar to 3D die production. As we have documented in numerous other articles, this effort has been marked by significant difficulties that, all combined, have reduced yields to such an extent that the conversion to date has resulted in higher cost per bit than the planar process that was replaced. For all the entities producing NAND - save one - only now, with the advent of the 64-layer die is the density advantage of 3D showing strong promise of overwhelming the relatively poorer yields of the new process.
If these were normal times, this density achievement - 64 layers - could be counted on to greatly increase the supply of bits coming to market by next year at the latest. Once again, with the exception of one entity, customers are being disappointed. Industry productivity is still low due to a condition that could be called planar overhang - that being the amount of planar capacity that must be converted as fast as possible to 3D, so the company can take advantage of the denser 3D process. Unfortunately, this conversion process from planar to 3D is basically like buying a house that has to be completely renovated and then finding out that load-bearing walls are involved - and the foundation has to be reinforced. The net of this conversion is a loss of bit output for up to a year as planar production is torn down before the new etch-intensive 3D process can be begun. How big is the planar overhang? According to Ernie Maddock of Micron (NASDAQ:MU), almost 70% of the wafer output of the industry is still planar. (Bit output of 3D by the industry as a whole should reach crossover with planar by the end of this year.)
So here we are, just ending the first calendar quarter of 2017, and bit output growth from the industry is estimated to be at historic lows - at 40% CAGR and maybe below. (Indeed, Samsung is projecting only 30% bit growth this calendar year.) Demand, on the other hand, is as strong as it has been in years at 45% (plus or minus a few points depending on who you talk to) and is showing no signs of abating. Inventories across the industry are low, prices are high - up 6-10% - and customers are scrambling to find product.
The result of all this commotion is that customers are crying, but the industry oligarchs are a happy lot despite their productivity difficulties. This is especially true for the IM partners, Micron and Intel (NASDAQ:INTC), the one shining exception to this otherwise dismal picture of industry productivity. Micron's projection is that its YoY NAND bit increase will top 60%, and Intel's projected bit increase is even stronger at an estimated 100%. Planar overhang? Not for the IM partners. They are practically done with the conversions. Yield problems? Nope. They are getting "mature" yields equivalent to their planar process - something not even the mighty Samsung boasts of. The net result of all this is a large bit share increase for Intel and especially for Micron as it claws its way back from years of underperformance and share losses in NAND. Even better for the IM duo, they will have a roughly 12% density advantage at the 64L node, touting the industry's smallest 3D die at 59mm2.
The net of all this is what we might call the IM trifecta. To wit:
- The "architectural" advantage in die density.
- The best and most productive fab process with the highest industry yields.
- The least - by a considerable margin - planar overhang.
Boil all this down and we are facing the opportunity for a truly tectonic shift in the industry. Interesting times, no? And we haven't even touched on the most interesting aspect of the current period, and that is the prospect of even stronger NAND demand over the balance of the decade - a topic I addressed in my last article. Focusing solely on Micron now, there's only one little problem with all of this wonderful news, and that is that it doesn't have the money to truly take advantage of its competitive strength. Micron needs the ability to fund more capacity - more wafer growth. If it can't build more fabs, the company will find that its current bit share gains dissipate in 2019 and 2020. Here's a graphic description of that phenomenon showing a scenario where the industry only adds the capacity that is currently planned (5 new fabs, counting Fab 10N and Intel's Dalian Fab) and finishes converting all its planar capacity to 3D by 2019.
And here is the total industry production and demand scenario in tabular form:
There are several important takeaways from the graphic above. Here are the most important of them. The first is that the industry has the capacity to meet the needs of customers through 2020 if demand increases at Micron's forecast 45% CAGR rate and the industry maintains capacity discipline by building out only the five new fabs announced or in progress. (Please note that the driving assumption behind this growth scenario is that the industry solves the 64-layer limit they face through either string stacking or improved etching tools by 2019.) The second major finding is that Toshiba's (OTCPK:TOSBF) and WDC's strong share gains are a direct function of the building of a gigantic new fab coming online in mid-to-late 2019, allowing them to match Samsung's new wafer additions during this period.
There is one other vital piece of information to keep in mind. This new Toshiba/WDC fab has a proposed budget of $7.8B. According to the Anandtech article:
At the moment, Toshiba and Western Digital (which acquired Toshiba's partner SanDisk) produce more NAND flash at the Yokkaichi Operations memory production complex in Mie prefecture than any other manufacturer and Toshiba would certainly like to keep it that way."
This picture, from a WDC presentation at its recent Investors Day, will give you a notion of the scale, estimated to be 500k sq. ft.
The strategic question that leaps from this information is this: would Toshiba/WDC commit to such a large investment if A: they didn't believe that demand would stay at least at the 45% CAGR level, and B: they didn't believe that they could fix the BiCS density and yield issues? So now to the most interesting question of all.
Will this fab get built?
That is the question that arises out of the recent news that Toshiba has decided to sell its 50% stake in the Flash Ventures JV with Western Digital.
The facts as we know are the following:
- After initially saying it wanted to sell only a portion of its interest, it has concluded that it may have to sell 100% of its JV share.
- It wants bidders to respond by March 30th and it expects to close the sale by the end of June.
- It is looking for at least $14B.
From news reports including Barons Asia, and (thanks to H. Bruce Campbell) ITPro from Nikkei Business Publishing, we know that there are numerous bidders, including WDC, Micron, SK Hynix (OTC:HXSCF), and Hon Hai Precision Manufacturing (OTCPK:HNHAF) (Foxconn (OTC:FXCOF)), Bain Capital, Canon (NYSE:CAJ), and many others that we probably haven't heard about.
Here's a news item from the NY Times, dated Jan. 27th.
One public declaration of interest in Toshiba has come from Canon, the Japanese camera company, which uses Toshiba's chips in its products and bought a medical device producer Toshiba spun off in 2015. Canon's chairman, Fujio Mitarai, is a former head of Keidanren, the lobbying group representing Japan's largest corporations, including Toshiba. He said last week that the semiconductor business was a valuable asset for Japan and "must be protected" and that Canon "would positively consider" investing. […] Other potential buyers include Western Digital, the American semiconductor company, which works with Toshiba in some areas; Tokyo Electron, a Japanese company that produces equipment for semiconductor factories; and Foxconn of Taiwan, the contract manufacturer that recently took over Sharp, another ailing Japanese technology brand."
Sorting through the clutter, what we're seeing in the bidders' list is the following:
- Japan, Inc.
- Private Capital
- Memory Industry Competitors
- Non-memory Industry Companies who are memory consumers
Unmentioned but (thanks again to HBC!) on everyone's lips is Apple (NASDAQ:AAPL). Mentioned but not in serious contention due to serious national security issues is China, Inc. Who is going to win this deal? What follows is an assessment of the factors from each entity's point of view with particular focus on Micron (and by association IM).
Let's consider the current operating environment because this will have a big impact on the eventual choice of a winner. Yokkaichi is a joint venture of WDC and Toshiba. The partnership, originally between SanDisk and Toshiba, was initially formed in 2000. Last year, WDC and Toshiba formed Flash Ventures, a vehicle that rolls up other JV agreements signed over the years to finance and run the four fabs that have been built so far. The two partners share R&D, CapEx, and operating expenses of the 4 fabs equally. The 3D process that they have chosen, BiCS, is a charge-trap architecture. In this regard, it is similar though not identical with Samsung's V-NAND and almost identical to SK Hynix's 3D NAND product (primarily because SK Hynix stole the IP from Toshiba).
Although we do not know for sure (the Partners haven't published their JV agreement), it is quite likely that the Partners have some say in who acquires the other partner's share and they almost certainly have a right of first refusal. I am speculating here but I would be amazed if either party would countenance an agreement that did not include this. If this is true, WDC, whether or not it wins the bid, will have considerable power to shape the final deal.
Here are the assets in play:
- Fab 2 (new) - mid-2016 initial production, volume 2017 - 290k ft2 - $7B
- Fab 3 - 2004 - $7B - 300k ft2 (est.), 5 stories,1+ m ft2 floor area
- Fab 4 - 2007 - $7B - 382k ft2, 5 stories with a floor area of 1.2m ft2
- Fab 5 - phase 1 2014/ phase 2 2016 - 400k ft2 (total cost unknown)
- Fab 6 - 2019 - $7.8B - 400k ft2 (est. - construction just begun)
The total wafers of capacity in these fabs in the period before the 3D conversions began was nearly 500k wpm, with Toshiba claiming rights to roughly 30k wpm more than WDC because of the sale by SanDisk to Toshiba of 30% of the capacity of Fab 3 in 2008, leaving Toshiba with about 270k wpm capacity to WDC's 230k wpm.
Now that we know what is in play, let's handicap this deal from each bidder's perspective, starting with Western Digital. For WDC, the options are stark, with the positives being extremely compelling while the negatives of another supplier winning the deal are quite punitive. Consider for a moment an example of the negatives. Let's say Micron (or another competitor) bids and wins. Micron obviously is not interested in the BiCS process, so the company would convert Toshiba's share of the site to IM's 3D NAND. Doing this deprives WDC of its R&D partner, and at a stroke doubles WDC's NAND R&D expense. Beyond that, WDC suddenly has a competitor that, having acquired Toshiba's staff as well as its equipment, knows exactly how productive the BiCS process is. Micron would know Toshiba's yields and thus know its costs. In the notoriously secretive memory industry, this information is jealously guarded because of the competitive advantage to be gained. This would be intolerable for WDC, unless of course IM agreed to share and license its 3D NAND IP to WDC. Why would the IM partners do that?
Beyond that, why would Micron agree to fund its presumptive half of the new mega-fab? Micron would not want to share the new fab for all the reasons listed above. From WDC's perspective, it is now left needing to plan a new, presumably smaller fab, and because of the delay caused by the acquisition, this new WDC-only fab could be considerably delayed. In addition to the delay, WDC's risk in bearing the CapEx cost alone would grow considerably. This in itself is not a deal killer issue, but in combination with all the other negatives we can easily see why WDC would never willingly countenance a Micron winning bid.
On the other hand, if WDC wins the bid, beyond the formidable financial commitment (presumably only done with Japan, Inc. partners) it now can move forward with the same staff it has been partnering with (it would simply transfer the Toshiba personnel) and it has now more than doubled its NAND output. How would the deal be structured? Could it be managed as an all-debt transaction? Remember that the bid price - something above $14B, would need to generate an additional $8B in free cash flow to fund the new mega-fab, if WDC decided to move forward with it. And don't forget that WDC is already bearing $13B in debt from the SanDisk acquisition.
Considering another competitor who would love to win this deal, an S.K. Hynix winning bid might be more palatable to WDC. Compared to Micron, one thing that is notably different about the Hynix bid would be what we might call process affinity. Given the identical antecedents of the two companies 3D process, might it be possible that Hynix would simply adopt the Toshiba BiCS process company-wide? On the positive side, alone or in combination with Foxconn, an S.K. Hynix win would provide WDC with a deep-pocketed partner who would have little issue with funding Fab 6.
Looking at the negatives from a WDC (and Japanese) perspective, however, one inevitably bumps up against the question of enabling a competitor, especially a Korean competitor. Why would WDC want that? Beyond that question, there is the very real issue of geo-political risk. If anything happens on the Korean peninsula, WDC could be left with a crippled partner potentially unable to fund its half of the JV. Considering for a moment the Foxconn combination with Hynix, I find it hard to believe that Japan, Inc. would sanction a deal with a bidder with such close ties to China. Whatever assurances Terry Gou might give, the risk of a potential IP transfer to China, Inc. would be intolerable.
Bottom line, I find it difficult to believe that a bid from the Korean company would be contemplated. A Foxconn win is even less likely. Let's move on.
Let's consider the customer/vendor category. There are any number of Japan, Inc. potentials including Canon and Tokyo Electron (OTCPK:TOELF). Would they be attractive to WDC? Probably, because almost any customer that bought the Toshiba half would not want to operate the fab. Like the IM partners, where Micron runs the fab using jointly developed IP, presumably WDC would operate the Fab and provide the "silent" partner rights to half of the die coming out of the Fab. In Canon's case, would it even do that? Possibly it would want to simply take a split off the gross profit of the product sold by WDC sales and marketing resources? Quite possibly, using a template employed many times before, Japan, Inc. would limit its involvement to sponsoring the financing of WDC buyout at very attractive terms. Could one possibility be some Japan, Inc. funding entity (consortium, state funds, etc.) that could provide attractive financing to WDC of some share of Toshiba's interest while also funding Toshiba sufficiently to allow the company to retain some amount of the business?
What about non-Japanese customers? Let's consider the case of Apple, which presumably wouldn't want to bid on the full Toshiba share because it wouldn't want to pollute its industry-leading gross margins with the lower margins that "Apple Memories" would be generating. On the other hand, if Apple participated with other entities in a consortium where it provided less than 20% of the equity, it could keep the memory division's results off its books. If this is the approach that Apple takes, who would be likely partners? In my view, this is where Japan, Inc. would provide the most value. Imagine a WDC, Canon, Apple, and Japan Inc. loan consortium splitting the deal. Apple and Canon get access to product, and WDC gets to operate the fab independently. Would this combination pass anti-trust scrutiny? I don't know the answer to that, but taking some arbitrarily large portion of the new JV's production out of the merchant market seems to me to raise the question, especially if say, Apple's deal is at more favorable terms than those provided to other customers.
If anti-trust concerns can be accommodated, such a combination makes a lot of sense to me because it solves so many problems. First, it allows WDC to keep competitors at arm's length. Second, it provides the financing that enables the silent partner assured access to memory product. Finally, it poses no risks to the IP being lost or stolen. All the partners' interests are aligned.
There, of course, is one gigantic assumption that underlays Apple's involvement and that is that Apple is convinced that the BiCS technology is competitive with V-Nand and IM's 3D NAND processes.
Is this the case? Whatever the density disadvantages of BiCS compared to IM's 3D CuA architecture, (about 12%), the real issue for BiCS is yield. How soon can they fix the yield problem? We do not have enough information to make a definitive call on this. Samsung uses a different charge trap die architecture and, according to sources I trust, is finally achieving competitive yields with its V-NAND product. So why couldn't WDC and its new partner accomplish the same? One thing for sure - the prospective purchaser of Toshiba's share will want to make a determination of this issue prior to closing the deal.
Let's close this. My analysis leads me to handicap this deal as follows:
There are only three possible winning bidders, with Japan, Inc. being the most likely, in some combination with WDC and a US or Japan-based customer (like Apple, Canon, etc). Toshiba may well be bailed out and allowed to retain some portion of the winning consortium's interest in the deal, most likely as an operator who might be offered terms that would eventually allow it to reclaim much of its former share.
No competitor is going to win the deal unless there is no option left. Less likely but very tantalizing is the prospect that private capital - say Bain or the like - combines with WDC to enable WDC to buy out Toshiba's share. There is only one way this happens, and that is if the private capital bidder really buys into my exploding NAND demand hypothesis. If Bain really believes that NAND demand is going to outstrip the capability of the industry to build capacity, why wouldn't it fund the buyout of Toshiba? I think that it is quite likely that a private capital provider will make an aggressive bid, but in the end Japan, Inc. is not going to let this technology jewel get away from them.
Finally, for those of you out there thinking Micron might somehow have a chance, there is only one circumstance that could enable such an outcome. Is it possible that WDC and Toshiba believe that their BiCS process is fatally flawed? If they do, then all bets are off, and they will do whatever they have to invite either of the IM partners to the party. Problem is, I don't believe they do. One has only to listen to Dr. Siva Sivaram from WDC at its recent Investors Day presentation. They may have a density and yield disadvantage now but Sivaram is clearly a man who thinks that BiCS process will be competitive over time.
Clearly the industry is on the cusp of what is potentially a tectonic-scale change to the industry structure that could pose major challenges or offer significant benefits to incumbent firms depending on how it turns out. The IM partners, Micron and Intel, will probably not be a part of the eventual solution that emerges in Yokkaichi, but they will be watching the result with great interest. It is entirely possible that Fab 6 could be delayed while the new partner(s) and WDC establish a working relationship and come to agreement on their CapEx plans. Given the industry's need for capacity going forward if there is any material issue with the funding for Fab 6 that will be a bright green light for a least one of the other players to get started on a new Fab. Micron doesn't have the money to build it. Would Intel?
Interesting times indeed!
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.
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.