Description

Introduction

This writeup on Western Digital is slightly unusual as I think the main drivers for the company are not necessarily next quarter’s earnings but rather whether there is a key structural change in the market and second whether a potential transaction with Toshiba could create a special situation.

The additional way this writeup is unusual is that the consequences of the above will have an impact on Seagate, Micron and Toshiba itself.

In addition I am seeking advice from fellow VIC members as to whether what seems to me the ‘obvious endgame’ is logical.

I also anticipate significant debate from other VIC members on whether WDC is actually a value trap or an industry in terminal decline.

Valuation

1. Based on consensus numbers WDC is currently on a P/E of 10.6x 2017 and 8 x 2018 (note that the fiscal year end is June).

2. On EV/EBITDA the company is on 5.8x 2017 and 5.04x 2018.

3. For reasons I discuss below (see appendix) whether the valuation is cheap or expensive depends on whether the areas the company operates in are:

1. in the case of flash in secular growth or cyclical

2. in the case of HDDs if the secular decline is less severe than the market thinks

4. It would also be amiss of me to not highlight that the company has significant debt ($13.088Bn as of March 2017 – equivalent to 4x total debt / EBITDA or 2.37 x net debt / EBITDA – Bloomberg numbers). So getting the industry direction right is essential to ensure the company does not get into a debt death spiral.

The key hypothesis

Western Digital has two divisions  - a hard disk drive division and a flash memory division. Historically both these industries have been used by business schools as poster boys of highly cyclical commodity type industries. In the appendix I will argue that the impending death of the hard disk drive is overstated due to demand for high capacity drives from hyperscale companies (though I agree with the long term destination) and that both the hard disk drive and the memory markets maybe less cyclical going forward. Some of the appendix is an industry primer so readers familiar with the industry may want to fast forward some sections.

Western Digital has guided to an eps of $12 for 2018. The real question for investors is whether anything like that level is sustainable beyond 2018. Ultimately this goes to demand for flash memory and for HDDs. Hence the majority of the appendices discuss the market dynamics. However the short term catalyst is what happens at Toshiba so I discuss that in the main body.

We could argue that for a cyclical stock one should look at the cyclically adjusted earnings but part of my thought process for WDC is that actually it comprises of two businesses neither of which is quite as cyclical as the market considers – in one case in slow shrinkage but offset by market consolidation (ie HDDs) and in the other case in secular growth for several years to come (flash).

The trades

To make the following clear the possible trades I am proposing are:

• Long Western Digital on the basis that Toshiba selling its memory division will be a direct read across to Western Digital (this is my preferred trade)

• Alternatively synthesise a ‘synthetic’ Sandisk by going long Western Digital and short Seagate (given the sharecount of STX is 296M (mkt cap $12.7bn) and WDC is 291M (mkt cap $26bn) and the share prices of $42.87 vs $89.90 – a lazy setup would be short one share of STX to long one of WDC; happy to discuss more precise ratios or adjustments for enterprise value as opposed to market cap if any interest)

• Long Micron on the basis that if Toshiba sells to private equity or to Western Digital the stretched balance sheets enhance the longer term position of Micron in relation to WDC (this is a more tenuous trade)

• Long Toshiba on the basis that the current EV of $16.6bn (and the cost of writeoffs of Westinghouse) are outweighed by the value the Toshiba memory division and hence a delisting is unlikely (I am still working on a full Toshiba analysis)

Toshiba

Western Digital (via its ownership of SanDisk) and Toshiba own 50% each of a joint venture (‘Flash Partners’) that produces roughly 1/3 of the entire global supply of flash memory. Because of the issues at Westinghouse, Toshiba has had to writedown the value of its nuclear division which has led to it having negative book value at the parent level. Under Japanese listing rules Toshiba faces delisting unless it can recapitalise its balance sheet (or receive a special dispensation which appears unlikely). Reluctantly it has apparently put part (some articles suggest 20%) or all of its memory division for sale. The rumoured valuation ranges from $18 – 29bn depending on which analyst or press report one considers. There have been multiple rumours of potential bidders including KKR, Hon Hai, Broadcom / Silverlake, Foxconn (with or without Apple) and SK Hynix.

[In comparison Western Digital acquired SanDisk for $14.291bn – deal announced 21 Oct 2015.]

Seagate is a pure play hard disk drive player with a market cap of $12.9bn and an enterprise value of $15.1bn. The Western Digital HDD business is actually bigger on some measures than Seagate – eg about 16% bigger in sales (FY 2016 to 1 July 2016) but has historically been less profitable so I think using Seagate on a one for one basis is roughly right (arguably I am being conservative; also the Western Digital business still has cost savings from its acquisition of Hitachi Global Storage to come through ). Therefore on an market cap basis ‘SanDisk’ (ie the flash division of WDC) has an implied valuation of $13.2bn – and hence it is possible for investors to create a synthetic ‘SanDisk’ (ie long WDC, short STX) – and the Toshiba deal ought to lead to some ‘uplift’ for the synthetic SanDisk. (On an EV basis a synthetic SanDisk has an implied valuation of $18.6bn – still at the lower end of the Toshiba memory valuation range). This may interest some VIC members, but as I discuss below, I think there could be much more ‘interesting’ scenarios.

Western Digital has expressed an interest in either acquiring some or all of Toshiba or of ‘protecting its interests’ in the joint venture.

Searching through EDGAR I came across the SanDisk / Toshiba joint venture agreement here - https://www.sec.gov/Archives/edgar/data/1000180/000100018010000105/ex10_1.htm . The joint venture is organised as a ‘godo kaisha’. I am not an expert on Japanese law but I note that Wikipedia states that ‘All members must consent to any transfer of ownership, unless the articles of incorporation provide otherwise.’ (see https://en.wikipedia.org/wiki/G%C5%8Dd%C5%8D_gaisha ).

Clearly I am not privy to all the agreements and legal documentation between Toshiba and SanDisk / WDC however this might fit in with some market suggestions that WDC has some blocking covenants in the j.v. In particular there have been some news items stating that WDC believes that Toshiba might be in ‘default’ of the j.v. agrements by putting its stake up for sale – I think the term ‘default’ may be particularly important as the master agreement (referenced in the previous paragraph) refers to the term ‘default’.

If Western Digital does indeed have blocking covenants then there is a non-zero chance that Western Digital could end up with an improved economic exposure to the joint venture (or a payment) as a result of any transaction – I do not think this is being properly reflected by the market. (It is worth reading the following - https://www.theregister.co.uk/2017/04/13/wd_tosh_memory_biz_sale/  )

It is worth noting that as I complete this writeup the Financial Times is reporting that Toshiba is locking out Western Digital staff from its sites and that WDC is claiming it does have blocking rights – see https://www.ft.com/content/1a91be22-346f-11e7-bce4-9023f8c0fd2e

The Endgame - Seeking advice from VIC Members

Some news reports have suggested that Western Digital is seeking funding from a private equity partner to fund an acquisition of the Toshiba memory division. I appreciate that Western Digital is probably trying to avoid dilution of its shareholders or to increase its leverage; however it seems to me that the company is at risk of effectively paying an ‘agency fee’ to private equity which will harm its own shareholders longer term.

W.r.t Toshiba, I think there was an interesting comment by Toshiba management during its Q3 2016 Earnings Call (Bloomberg Transcript 11 April 2017):

<A - Ryoji Sato>: One thing I'd like to comment is that the ¥2 trillion that you mentioned in relation to the Memory Business, it is an amount that is being floated in relation to possible sale of the business. And if you exclude the possibility of sale, the value to be booked in the balance sheet would be entirely different. We're using something entirely different. So, it's different from accusing us of using different accounting standards.

And:

<A - Masayoshi Hirata>: This is Hirata speaking. I believe that the – under the current accounting standards, the NAND portion will be so called self-created goodwill, and the current accounting standards do not allow that. So in that sense, I think that the accounting based on the U.S. accounting standards are the non-disputable number.

I believe that the company (Toshiba) is highlighting that under current accounting standards, despite their recognition that the memory business is worth substantially more than book value, it cannot restate its book value without a transaction. (My understanding is that a certain minimum – say 20% of the stake at least – has to be sold to allow Toshiba to revalue the rest and stay listed).

It seems to me that the obvious solution is for Western Digital to offer a share swap – so it offers shares in itself for a stake in Toshiba’s memory division. By doing this, WDC would avoid the cost and value leakage of private equity, and Toshiba would be able to maintain roughly the same economic interest in flash memory and the transaction would create the goodwill necessary to return its balance sheet to positive book value. One can argue whether Toshiba ought to swap only part of its holding in its flash division or all of it, and whether it ought to own shares in Western Digital or only in the SanDisk division but the underlying point is that rather than involve private equity it seems to me that Western Digital can structure a win-win deal.

Additionally I believe that there could be significant cost savings if the flash joint venture between Toshiba and SanDisk were collapsed into one entity. Technically the joint venture sells flash to both Toshiba and SanDisk both of whom then do downstream processing and packaging on their own; and then market through their own channels. It would not be inconceivable to assume that removal of this duplication could save several hundred million but it is hard to pin down exact numbers.

However I think the most interesting issue for any deal with Toshiba is tax. The following article might suggest that a share swap might allow Toshiba a capital gains tax deferral – I am not a tax lawyer and would love any feedback from VIC members who have insight into this - https://www.pillsburylaw.com/images/content/1/5/v2/1579/AD74895623D9DEA8CFCE15A2836992E0.pdf (Alternatively I am wondering if the tax -qualified merger rules in Japan are applicable – see https://www2.deloitte.com/jp/en/pages/tax/articles/rt/tax-qualified-merger.html ). I appreciate that due to the Westinghouse situation Toshiba may anyway have significant tax losses it can use to offset any capital gains however it will depend on the tax domicile of those losses.

There is some discussion whether Toshiba needs any immediate cash injection (remember that Westinghouse has been put into chapter 11 but Toshiba has indicated it will stand by any (or some) commitments it has made). In case immediate cash is needed I see no issue with WDC selling say $1 – 3 bn of shares given the proposed new scale – and I believe that investors would react positively to the consolidated entity.

Obviously, there are various ‘spins’ that could be put on the above structure – eg that Toshiba receive convertible bonds etc but the bottom line is that it seems to me that the above proposal offers an obvious solution that is attractive to both Toshiba and WDC shareholders. Furthermore I believe it will be attractive to regulatory and national security authorities in both Japan and the USA.

(The one wrinkle in the above is whether it would create an FTC issue given Toshiba’s exposure to HDDs; given that Toshiba is a lagging player and the industry is in decline I think this is less likely. Nevertheless the solution to this would be for Toshiba to put its HDD division up for sale or for the deal to only involve the SanDisk part of Western Digital. I think that there is less likely to be concern on the flash side as the joint venture is really the relevant entity and its ownership would not change).

I appreciate that there will be many investment bankers involved in advising the various parties but it seems to me that sometimes there is a lot of hubris around deals and simple solutions are missed, perhaps for those that will generate more investing banking commission. In addition, I suspect that WDC and Toshiba managements might not appreciate how supportive shareholders are for a deal along the above structure.

The other reason for Western Digital to intervene is because I believe it cannot countenance an Apple / Foxconn joint acquisition of the Toshiba memory business as this will substantially alter the market dynamics and impact its long term business interests. (Obviously Samsung is already a major player in both memory and phones).

Should WDC require a cash element for any deal with Toshiba it seems to me that the obvious sources of funding (other than the stock market) are the hyperscale players (ie Google, Amazon and Microsoft) [and possibly Apple as part of a consortium]. This would not mean that WDC would guarantee any supply to any specific player (at that would lead to FTC issues) but rather than ownership of a stake in WDC would act as a hedge to the hyperscale players w.r.t rising pricing for flash memory.

From the Toshiba board point of view the clear issue is going to be whether Western Digital can offer enough (in paper) to match other buyers. As the above mentioned FT article suggests WDC may be financially stretched and of course it is arguable whether WDC could really offer much more than $18bn given that is the implied EV of its own SanDisk division. I believe given the above mentioned (i) strategic (avoid Apple or a financial buyer) (ii) synergy (iii) tax items (both for Toshiba and WDC) I believe that WDC could justify a price much closer to the high end of the rumoured range. Furthermore the deal could be structured with a series of converts and puts (eg Toshiba holds back some of its stake but has a right to ‘put’ it to WDC in 1,2 or 3 years at a higher price) plus CVRs to ensure that an independent banking opinion would confirm that WDC is the most attractive deal for Toshiba.

Benefits for Toshiba for doing a deal with WDC are (i) likely to be faster and more certain to implement than deals with alternative bidder; (ii) maintains exposure to flash which it has indicated it wants (iii) if the whole deal is done in one block Toshiba would own more than 25% of WDC which may have consolidated reporting benefits (I cannot quite work out the threshold at which Japanese companies can consolidate and whether a board majority is required) (iv) if Toshiba does intend to retain some stake in the business post transaction then a stake in WDC is more liquid than a stake in a private equity led private entity and furthermore a private equity led deal risks underinvesting at a critical time so ultimately a WDC deal is more attractive

I would appreciate input from other VIC members whether the above concept of a share swap sounds sensible and secondly whether dialogue with the boards of Western Digital and indeed Toshiba might help focus on this. Admittedly I run a very small fund (smaller than probably many people’s drinks budget!) so it might be more effective if someone else were to take up the baton on this?

Below in the appendices I give some additional thoughts on WDC.

Appendices.

Appendix 1: The Toshiba / SanDisk jvs:

Technically (i) there are actually three joint ventures (Flash Partners, Flash Alliance and Flash Forward) between Toshiba and Sandisk (ii) secondly Toshiba owns 50.1% of each jv (iii) some of the facilities for the joint venture are leased from Toshiba. On the other hand part of the SanDisk agreements involve SanDisk business units in Ireland or Holland and so I suspect that SanDisk has a lower tax base than Toshiba – so I think it is reasonable to assume a 50:50 valuation is roughly right.

The Y2 Facility agreement is at https://www.sec.gov/Archives/edgar/data/1000180/000100018016000068/sndkex-1037xnewy2facilitya.htm .

The operating agreement for Flash Forward is at http://corporate.findlaw.com/contracts/operations/joint-operating-agreement-for-flash-forward-ltd-sandisk-corp.html// ).

2. Hard Disk Drives (HDD)

2.1 Industry concentration

Historically there were numerous manufacturers of HDDs - for instance Wikipedia suggests at least 221 (see https://en.wikipedia.org/wiki/List_of_defunct_hard_disk_manufacturers  - in particular reference 1) – in comparison there are only 3 major players now – Seagate, Western Digital and Toshiba.

According to Storagenewsletter.com market shares are Western Digital at 41.6%, Seagate at 35% and Toshiba at 23.4% (see http://marketrealist.com/2016/10/hdd-shipments-expected-fall-seagate-western-digital-fiscal-1q17/ ). Anandtech has slightly different numbers (http://www.anandtech.com/show/10315/market-views-hdd-shipments-down-q1-2016/2 )

I am pre-empting my writeup – but I would suggest that the ‘leading edge’ for HDDs is 10 – 12TBs drives (in demand by hyperscale companies) – and the only games in town here are Seagate and Western Digital ie Toshiba appears to be lagging the competition. Thus effectively the market is becoming a duopoly – especially at the leading edge.

2.2 Customer base and usage

The customer base for HDDs has altered – at the turn of the century there were five major PC makers (Dell, HP, Compaq, Gateway, IBM) that dominated the PC market so we had a relationship of numerous HDDs to 5 PC makers. Nowadays, though the number of tier 1 PC makers have shrunk, HDDs are used by a more diversified customer base – including settop makers, gaming consoles, some auto makers and hyperscale infrastructure players.

It is worth a brief comment on what we mean by hyperscale infrastructure players – this refers to companies offering cloud based services (including compute platforms) such as Amazon (AWS), Microsoft (both Office, OneDrive and Azure) and Google (with its multiple services – including importantly Gmail and YouTube). Note that the hyperscale players often specify their own server designs or sometimes even have them made exclusively for themselves (so in many ways are a distinct and different customer base from PC makers).

There are a number of additional drivers for storage that are often ignored – in the consumer market particularly photographs and videos; and in the enterprise market the increasing regulatory demands requiring (say) 7 years of archiving. An additional driver is security and in particular digital CCTV – more and more companies and cities are installing digital CCTV cameras and storing 7 – 30 days or more of digital video for security purposes.

Analysts also often overlook that there is still considerable amounts of active tape and optical storage / archiving that HDDs are cannibalising (anyone really interested in this ought to look into the discussions on whether Amazon Glacier uses HDDs, tape or optical storage – https://news.ycombinator.com/item?id=13404840 and on Microsoft Pelican - https://www.microsoft.com/en-us/research/wp-content/uploads/2016/09/pelican-hotstorage2016.pdf and https://www.microsoft.com/en-us/research/publication/pelican-a-building-block-for-exascale-cold-data-storage/ ).

2.3 Criteria for storage medium selection

The general pushback on being excited by HDDs is that it is a declining industry and therefore the terminal value of the suppliers is zero. The assertion is basically that ‘as everything is going to the cloud, there will be no need for HDDs’ (this assertion has, for instance, been told to me by an analyst – when I asked him what he thought the ‘cloud’ saved data on he was rather flummoxed).

I believe that this is incorrect and reflects a basic misunderstanding of how the cloud operates. The cloud is not a magical place – and data must be stored on some device – whether it be a hard disk drive or flash storage. There are various trade-offs that hyperscale infrastructure players have to consider including:

• The cost per byte stored (and the total lifetime cost of ownership)

• The resilience of the storage medium (surprisingly if you really want to store something for 50 + years acid-free paper might still be the best)

• The ‘footprint’ or physical density of the storage medium (data centres are expensive on a per sqft basis)

• The electricity consumption of the storage medium (it takes power to keep a hard disk drive spinning; less so for flash memory) – this also depends on whether the storage is ‘hot’ or ‘cold’ – see https://www.backblaze.com/blog/whats-the-diff-hot-and-cold-data-storage/

• The failure rates of the medium (hard disk drives are subject to failure – the following blog is very insightful - https://www.backblaze.com/blog/hard-drive-benchmark-stats-2016/ )

• The read / write characteristics of the data flows – in particular there is a dirty secret with flash memory that a lot of people seem to be missing – it can wear out – see http://www.storage-switzerland.com/Articles/Entries/2012/3/6_Why_Flash_Wears_Out_and_How_to_Make_it_Last_Longer.html

There are in addition some subtleties regarding transferring data to the cloud:

• A typical home or corporate user might only use a fraction of the hard disk drive capacity in his computer – let’s say 50% (please bear with me on this – I accept that corporates tend to use network drives etc)

• When data is transferred to the cloud however typically 3 copies of each file are stored to allow for drive failures / errors and crashes; historically this was done with hardware RAID systems but often for large scale systems nowadays it is a mix of hardware (JBOD) and software. On the other hand, hard drives in storage arrays are run ‘hotter’ ie there might be 90% plus utilisation of the capacity rather than 50%

• Overall however ‘backing up’ or ‘storing’ data on the cloud does not lead to demand for storage vanishing – but rather there is demand for larger drives (it is more cost effective for a hyperscale provider to use 10TB drives than 1TB drives)

• Often people point out to me that by storing files on the cloud duplicate files can be deduped – this is true in theory. However it is not clear to me how much deduping really occurs as it requires large numbers of symbolic links, a lot of computational overhead and can make a large file system unstable. Additionally, if data is encrypted before transfer to a cloud service provider deduping becomes harder (and possibly impossible – there are ways to dedupe before uploading but that then risks security).

• (Technology aside – I have ignored that OS system files are typically not archived)

Currently hard disk drives are significantly cheaper than SSDs (ie flash drives) on a per GB basis – there has however been a lot of discussion whether the TCO (total cost of ownership) including electricity consumption etc means that actually SSDs might be more cost effective than HDDs. This does depend on whether the HDDs are being used for hot or cold storage and the cost of interface and a lot of other issues. However for the next 2 years our analysis suggests that there will be sufficient shortages of flash that this will not really be an issue.

Our view is that there are some applications where the data is of low value or is rarely accessed (eg photos of the family cat) where HDDs offer better value; however there are other applications (eg analysing high frequency market data) where SSDs offer a better solution.

2.4 Compute platforms

Compute platforms such as AWS (Amazon), Azure (Microsoft) and Google Compute have attracted a lot of investor attention. The mix and demand for flash memory or HDDs for such platforms depends on the problem being solved – some computational problems require access to lots of data – in which case one might use a HDDs, other problems require more processor power but less data so one might use an SSD. In practise most large and complex problems use a mix of both HDDs and SSDs. This can be seen by reviewing the following page for AWS which lists various offerings and whether it includes SSD storage or Elastic Block Storage (see https://aws.amazon.com/ec2/instance-types/  and https://aws.amazon.com/ebs/details/#VolumeTypes )

2.5 Form factor

We do expect HDDs to be phased out of laptops and many desktops. For instance some forecasts suggest half of all laptops shipping by year end will have SSDs rather than HDDs.

2.6 Putting it all together

The number of units of hard disk drives shipped going forward will fall (roughly units have fallen 20% in the last year alone). However especially in the high capacity market there is an effective duopoly and so we anticipate that pricing will not be wholly irrational. In addition both Western Digital and Seagate have been on significant cost reduction programs and active in reducing the unit capacity of their drive businesses. In the case of Western Digital the company also has synergy savings coming through from its acquisition of Hitachi Global Storage Technology (and also Sandisk).

3 Flash Memory

Both DRAM and Flash memories have also been historically very cyclical industries. Again however there has been substantial industry consolidation such that the top five players in each segment control over 90% of the market.

In the case of flash the major players are Samsung (37.1%), Toshiba (18.3%), SanDisk (17.7%), Micron (10.6%), SK Hynix (9.6%), Intel (6.8%) (Q4 2016 – source https://www.statista.com/statistics/275886/market-share-held-by-leading-nand-flash-memory-manufacturers-worldwide/ ).[Note that SanDisk is owned by Western Digital].

However the above market shares hide the market concentration – specifically that Toshiba and SanDisk manufacture in a joint venture (Flash Partners). Furthermore, Intel and Micron are in a joint venture (Intel Micron Flash Technologies = IMFT).

3.1 Current shortages

There are currently shortages in flash memory – this has led to price increases. This is best reflected in the spot market – for instance the following chart shows NAND MLC 64Gb flash memory which has moved from approximately $2.10 in the summer of 2016 to $3.45 currently.

(I should highlight for non-tech specialists that there are lots of issues with how spot prices are calculated and how demand migrates between densities so the above chart is intended to illustrate a trend rather than be interpreted completely literally – and the majority of memory is sold via contract pricing which lags spot).

The significance of this price increase is best appreciated when one considers that memory companies generally anticipate a fall of 17-20% in price per bit per year and their internal roadmaps for cost reduction model this.

3.2 Why are there shortages – supply transition

The industry and analysts had been anticipating 35-45% bit growth in flash memory supply this year – in fact supply looks like it is going to come towards the lower end of this range.

Our work suggests that there are several reasons for this:

• The transition from 2-D to 3-D 64 layer flash has taken longer than expected to stabilising production yields

• Typically memory companies regularly introduce ‘shrink’ (ie more to smaller geometries) to drive down cost. However in order to move to 64 layer they have actually slowed shrink – or in some cases moved back one node (eg a company might have shrink from 18nm to 14nm to 11nm but then moves ‘back’ to 18nm to perfect 64 layers)

• Moving from 2D to 64 layer 3D does not automatically lead to 64 x the capacity – there is various overhead on a chip (eg interlayer connectors) which take space and some lines were already on 32 or 48 layers (admittedly much of this was for testing the process of layering)

• Not all fabs from all flash memory OEMs are migrating to 64 layer at the same time – so this transition will take into 2018 (Micron in particular has emphasised this point yet analysts seem to be ignoring it)

3.3 Why are there shortages – demand growth

We touched earlier on the demand drivers for HDDs. Many of the same drivers and considerations apply to the demand for flash memory with one additional consideration – mobile phones. The competitive arms race is driving increasing flash memory capacities in mobile phones. The average amount of flash in a mobile handset has been increasing in the last few years and we anticipate that to continue.

In the cloud compute space many high end Intel processors (eg the 22 core processors giving 44 virtual cores in AWS’s largest instances) frankly require the speed of flash memory to keep them filled. For certain applications, depending on electricity costs, and the cost per sqft to rent a data centre SSD costs are close to HDD costs. Though some analysts suggest that there has been a cross over I disagree as often hyperscale infrastructure companies get very attractive bulk pricing deals on HDDs and they can power down drives (most analysts assume all drives are rotating 24 x 7). Nonetheless cloud infrastructure is a big driver for flash demand.

And as touched on earlier we anticipate laptops to be predominately driven by SSDs by next year.

3.4 Why have analysts missed this?

One of the reasons for my tardiness in writing this note is that we spent the autumn and winter of 2016 – 2017 confused as to why analysts did not seem to be upgrading their numbers for Micron and Western Digital.

I will use Micron to illustrate some of the issues (but exactly the same applies to Western Digital):

• At the 23 March 2017 results the market expectations for Q3 (to May 2017) were around 86 -90c; the company guided to $1.50 and emphasised that it foresaw demand / supply mismatch to continue into 2018

• Yet as recently as earlier this week one large bank analyst downgraded Micron and moved his price target to $30 (from $32) with an eps estimate for FY 2018 of $3.75 and $2.75 for FY 2019

• I would argue that analysts have been sufficiently burnt by memory companies that they refuse to accept that they are anything but cyclical

There are a number of issues that seemed to be missed:

• Memory fabs are truly expensive (new fabs cost over $10 billion) and the industry has consolidated so new capital is not deployed as fast as it used to be; analysts seem to be assuming that new capital can flow freely into the industry

• Some of the production was already at intermediate 3-D stages (ie 32 or 48 layer) – combining this with moving back a node (discussed above) led to less bit growth than anticipated

• The transition to 3D was slower than expected (if you get 1% error on a 2D wafer that means that 1% of the chips are bad; the same maths on a 64 layer wafer leads to a substantially higher rejection rate (ie 1- (1-0.01)^64  -> 47%).[The maths does not strictly work like this but is illustrative].

• The demand from the switch from HDD to SDD (especially in laptops and cloud) is faster than expected

3.5 Surely there is more supply on the horizon?

Over the winter when discussing the flash industry with analysts the response has been that there is bound to be more supply coming. However there are a number of issues that are different from previous memory cycles:

• Scale is essential to be competitive and the industry has consolidated

• There are high patent and knowhow barriers

• DRAM and Flash have different dynamics – and the lines are not easily fungible (Samsung did transfer one line at the end of last year from DRAM to flash but it did take production offline for a period of time)

• A new production line takes time to achieve stabilised yields

3.5.1 The Impending Silicon Wafer Shortages

The real killer issue however is that our work suggests that for at least the next 9 – 12 months there is global tightness in 300 mm silicon wafers (the market is concentrated with four major suppliers). We estimate that the supply / demand is within 2% and hence prices are rising.

Thus even if a new startup flash fab started today it would not have the raw material (ie silicon wafers) to produce its chips for over a year.

3.5.2 The Chinese are coming

There is considerable activity from the Chinese (specifically Tsinghua Unisplendour) but we anticipate that to be a 2018 – 2019 issue and for various reasons we think that the supply will be managed). In particular, the company has relationships with Intel and Micron and we believe will effectively be outsourced production for these companies (see eg https://www.bloomberg.com/news/articles/2015-10-12/tsinghua-s-latest-acquisition-is-the-chairman-of-micron-venture).

Flash memory has significant patent and knowhow protection. Interestingly some conversations suggest to us that as Chinese companies sell outside of China they become more concerned about the impact of patents and pay up license fees; whilst some more established Western firms are more ‘relaxed’ about this issue.

3.6 Laptops and DRAM

The next generation of laptop processors from Intel (Kaby Lake) might increase the maximum ceiling of memory a laptop can practically utilise (and move to a new low power technology LPDDR4) which might lead to some increase in DRAM demand – this may increase pressure on tool makers. (There has been much speculation on whether Apple’s MacBook Pro autumn 2016 refresh was tepid because allegedly CPUs capable of using LPDD4 were late).

3.7 3D Xpoint

Given the discussion on flash it would be irresponsible for me not to mention 3D Xpoint which has been jointly developed by Intel and Micron. However I do not see it disrupting the above market structure any time soon.

Catalyst

1. Sale of Toshiba memory business to a third party leading to read across for Western Digital or the acquisition of the Toshiba business by Western Digital itself.

2. Timescale - the form of a transaction should be clear within months

3. Recognition by analysts that the flash storages are going to continue at least into 2018; and recognition that wafer shortages prevent new entrants making significant impact until late 2018 at the earliest.