September 14, 2018
The first 12 days of September witnessed a bear run on Micron and the Semicap Equipment companies due to a confluence of: 1) a number sell-side firms reducing target prices or raising concerns about Q4 being the peak for DRAM ASPs 2) a few choice articles in things like the Digitmes and DrameXchange/Trendforce that have exacerbated uncertainty and 3) the daily whipsaw of where the US stands in trade negotiations/tensions with China.
This update is going to cover three things:
1. A characterization of the three buckets that analysts on Micron have been falling into in September and thoughts on some of the research
2. How this time is different (assuming we are indeed going into a softening of DRAM ASPs)
3. Impetus for strategic action on the IMFT JV appears to be building
False summit or likely steep decline from the peak? There is a range of points of view on DRAM pricing.
Before breaking down recent analyst actions into three buckets, some foundational elements need to be appreciated by a Micron investor when reading any research on the Company:
– >70% of revenues and >80% of gross margins are generated by DRAM with the balance, for all intents and purposes, coming from NAND. Make no mistake, when making a play on Micron one is making a play on DRAM. NAND is arguably a nice (negatively valued many would argue) component of their business, but simultaneously minor driver of its cash flow and bottom line. Buying or selling Micron based on NAND is like buying or selling Apple based on a combination of its Mac and iPad product lines and disregarding the rest of Apple’s business.
– Micron has only just reached 1X node crossover in DRAM (which they have indicated leads to a 20% reduction in costs per bit vs. the prior generation 20nm node) and 64L 3D NAND (which is expected to lead to a 30%+ reduction in costs per bit).
– Medium-term expectations for bit demand growth in DRAM and NAND are 20%+ and 40+%, respectively.
– While there are effectively three manufacturers that make up over 95% of the DRAM market (Samsung, SK Hynix and Micron) there are six manufacturers that make up the bulk of the NAND market (Samsung, Toshiba, Western Digital, SK Hynix, Micron and Intel). DRAM has become very oligopolistic. NAND – not so much.
With those four critical points in mind, there have been a slew of analyst reports written on Micron since the start of September. The latest reports largely fall into three buckets:
Bullish – Either no or only modest ASP declines expected
BofAML, CS, DB, JPM
Bullish – But reducing TP, next year’s EPS estimates and expectations on DRAM and NAND pricing
Baird, RBC, Macquarie, MKM, Susquehanna
Neutral/Hold – Reducing TP, next few years’ EPS and expectations on DRAM and NAND pricing
Citi & GS
Here’s how EPS Estimates, Target Prices (TPs) and Ratings have changed in September.
A few observations before focusing on what’s been driving all of this selling pressure on Micron shares.
1) While the median 2019 earnings estimate of those firms that have published on Micron in September has not changed much (-6%), the range of estimates has widened considerably now going from $7.68-$12.73 (previously $9.83-$12.73). This speaks to how little visibility the Street has into medium term memory pricing and increased concerns over how 2019 DRAM ASPs will play out. Recall if one goes back two years ago, none of the sell-side imagined how profitable Micron would be today. Similarly there is an increasing debate about how profitable Micron will be two years from now. It is no coincidence that most of the sell-side firms shown don’t have forecasts out in FY 2020. Those should come within the next two weeks when F4 2018 has been reported and the bulk of the sell-side feel compelled to issue updated models. But big picture, I submit that when the Street is this divided on the EPS outlook it not only indicates uncertainty; it represents an opportunity for the focused buy-side analyst.
2) The median TP has come in 11% in September, but the stock is down >15%. While the median 2020 EPS estimate has come in 12%, that’s due to a sample set of only four and is catalyzed by the significant drop in Susquehanna’s estimate. Of note, GS’s action on the 12th(reduction of its MU TP from $63 to $50) was catalyzed by a mere 3% reduction in its 2019 EPS estimate. GS was already very lukewarm in its previous “Buy” rating of Micron with one of the lowest TPs and sets of EPS estimates on the Street.
3) Overwhelmingly the Street has “Buys” on Micron with a median target price amongst the firms listed in the prior table of $72.50, which is 63% above MU’s current stock price.
So why has Micron’s stock dropped >15% since the start of September?
Bears on Micron and traders steeped in DRAM pricing history are jumping on comments out of those like GS, Susquehanna, Citi and others that DRAM pricing is likely to peak in Q4 and that 2019 could see declines in DRAM pricing of anywhere between 5-20%. Some of those more negative voices on DRAM pricing on the sell-side throw in there that DRAM price declines are likely to be more 1H 2019 loaded (Susquehanna). The conventional wisdom in DRAM is that when prices peak, there is inevitably a steep decline in prices over the next six quarters (the average period of decline in DRAM pricing according to GS). One of the things that often exacerbates the reversal in prices is that instead of middlemen and end customers “over-ordering” in an increasing price environment, which leads to larger than normal inventories, those same customers typically under-order when prices start dropping.
So if you break it down in the three aforementioned buckets into the plain and simple:
1) You have the CS, JPM, BoAML and DB’s of this world that think something between nothing significant or just modest price declines in DRAM are going to happen and Micron is dirt cheap. In fact if you read BoAML and JPM based on their checks they think there’s little to worry about. Demand is robust and likely to stay that way. TPs for DB, CS and BoAML are $80, $90 and $100, respectively.
2) In the 2ndcamp you have the still bullish types like RBC, Susquehanna, Baird, Macquarie and MKM that basically arrive at “yes estimates need to come down some, but trough earnings in this cycle are going to be so high (somewhere between $7-$8) that if you put any reasonable multiple on those trough earnings Micron is a Buy and worth anywhere between $63-$75.”
3) In the final camp you have those like Goldman and Citi that even though they have price targets of $50 and $62, respectively, basically think that you can’t own Micron while DRAM price declines are in the offing.
Having characterized the debate and the rationale for the recent weakness in Micron’s shares, here are some thoughts:
1) Referring back to those “foundational elements” enumerated earlier, I believe that there’s a fundamental modeling error that’s not flowing through in the models of those sell-side firms that believe a 15% decline in DRAM ASPs will lead to dramatically lower EPS. If manufacturing costs are dropping by approximately 20% per DRAM bit per annum (due to node migrations) and DRAM demand/supply bit growth is approximately 20% per annum, even if ASPs decline at a 15% CAGR there should still be modestly increasing GM. I’ve taken a high level approach to a model and will include it after these bullets.
2) While NAND is clearly suffering greater ASP declines, I believe a) it very likely that costs are also coming in a lot lower (going from 32L TLC to 64L QLC is a 133% pick-up in memory per chip – I believe it likely that costs per bit drop dramatically more than 30% when making this leap) b) Micron is still in the early innings of converting its trade NAND sales to higher margin solution sales (SSDs, Managed NAND, Multi-Chip Packages) which should help combat greater price declines in trade NAND and c) Micron is very likely to be re-pricing its low margin, cost plus contract on NAND with Intel in the coming months – this should have a meaningful impact on those sales (over $500mm per year).
3) 5G is just hitting the market in the here and now with Verizon this week announcing its launch in four cities on October 1 and AT&T announcing that it will be launching its 5G service in 11 cities sometime in 2018. Waymo is anticipated to be launching its self-driving car service this year. Self-driving cars have been described as “servers on wheels” given how much DRAM and NAND they require. The point being this, while the biggest buyers of memory semis (mobile device manufacturers materially increasing memory content, enterprise computing growing very rapidly and PCs/notebooks experiencing stabilized unit growth and significant memory semi content growth) are all evidencing robust demand for memory semis, there are some new, significant, high potential demand for memory semis areas (namely 5G related services and smarter cars/autonomous vehicles) that are just emerging. It is not clear to me that this has been factored into demand forecasts.
4) Those analysts with the most pessimistic EPS forecasts basically have as their thesis “we are about to embark on a typical high ASP decline six quarter rout in memory prices because when these declines come, that’s how they usually play out.” They could be right. But a) as shown in my model below if declines are only as bad as the pessimists are forecasting in 2019 I think Micron should be in great shape and b) I’m going to highlight a few additional reasons that the pessimists could just as easily be very wrong in their outlook for Micron as they could be right.
Here’s the Model:
What’s different now vs. prior points at which DRAM hit peak ASPs?
For the 1sttime in memory (no pun intended), Micron has a strong balance sheet (bankruptcy risk is remote).
As of FQ3 it is expected that Micron will be in a net cash position of approximately $2bn, a gross cash position of $7+bn and have an untapped $2bn revolving line of credit. It’s fair to say that Micron has never had this much net liquidity on its balance sheet. In past cycles, not only was Micron facing rapidly declining profitability, it often had significant net debt positions thereby compounding business risk with financial risk (back during the last downturn in 2015 Micron had $4bn of net debt and negligible operating income). Under current leadership, those days seem firmly behind the Company.
Micron has a $10bn stock repurchase program representing almost 20% of the Company’s market capitalization with which to buy-in shares.
As of September 1, Micron is live with the program and it is believed that it may have already started buying in shares pre-the earnings black out. While Micron has committed to spend at least 50% of free cash flow, I see no real reason (absent Micron’s being as pessimistic on the DRAM pricing outlook as GS) for not buying in $2bn per quarter for at least the next few quarters. Not only is $2bn a quarter approximately what Micron has been spending the last few quarters on debt reduction, I calculate $2bn per quarter as representing approximately 75% of FCF over the next few quarters.
Diversity of demand for memory is much greater than in past downturns
There have been many that have highlighted how different and robust the diversity of demand drivers is now vs. 3, 5 or 10 years ago. PCs represent less than 20% of memory semi demand now – it used to be greater than 40% 10 years ago. Mobile is now close to 40% and enterprise is 30%+ and the fastest growing major sector of demand. But there are a multiple of emerging areas not least of which is the increasingly smart automobile that is growing faster in terms of memory content than any of the top three demand areas currently. Not only is the diversity of demand more robust than ever, so is the absolute level of demand. Samsung Securities published a bullish note on SK Hynix on September 5thhighlighting that that company’s (and the security firm’s) optimism was driven by data points like: 1) only 70-80% of DRAM demand for servers is currently being met 2) that same figure for mobile DRAM is “on par or slightly lower” and 3) the chipmaker’s monthly operating results have been improving every month of the third quarter. SK Hynix believes that DRAM pricing will stay strong in 2019 and that any price declines will come in markedly less than cost per bit declines.
The DRAM supplier base has never been this concentrated and they appear to be acting more rational than at past junctures going into a near term ASP peak.
During the last downturn Micron ingested Inotera further consolidating the DRAM industry. A recent move by Samsung to pushout some DRAM Capex before ASPs had even peaked is unprecedented according to industry watchers. There are rumors that SK Hynix may be doing the same thing. Samsung has two very interesting pressures on it that are arguably keeping it as disciplined in memory semis as its ever been: 1) Samsung has Elliott Management hovering over its shoulder and memory semis are now the, far and away, biggest profit driver of Samsung’s business and 2) it is understood by those close to Samsung that the Company is sensitive to becoming more than 50% of the DRAM market – which could garner unwanted scrutiny. Micron has made it clear that up until now their technology investments have been focused on node migrations and not on producing incremental wafer throughput. Market share does not seem to be the focus.
Micron is not the same degree of laggard on the technology front as they were in the last (and past) cycles
At the outset of the last cycle, Micron was a tech laggard and by their own admission were two years behind Samsung (the tech leader in leading edge memory). That’s no longer the case. Micron is arguably only 6-9 months behind Samsung in some areas and in other areas (like QLC 3D NAND) Micron is ahead of Samsung. Further, Micron has developed some form factor advantages for its chips that make them especially desirable for mobile applications. Micron has also developed a strong reputation in the automotive industry in terms of reliability and is the leader in that market for its memory products; the growth of which is one of the reasons for the $3bn expansion of its Manassas, VA production facility.
Even the most pessimistic sell-side forecasts project earnings that are higher at the trough of this potential cycle than Micron has achieved at the peak of any previous cycle.
Trough earnings in this potential cycle as forecast, by the most pessimistic sell-side firms, come in at $7-$8 per share. Micron has no history of delivering more than $3 per share (per annum) at the peak of any previous cycle. Most current Micron investors would be thrilled with a 10x valuation on those trough earnings.
Micron in the next quarter, if it isn’t far along already, is expected to be renegotiating the pricing it has historically provided Intel, as part of its IMFT agreement, on NAND.
The IMFT JV, for all intents and purposes, is quickly shrinking down to only entailing its huge 3D Xpoint manufacturing facility in Lehi, UT. And that, I believe, Micron is likely to exercise its option on in the coming months. No more joint development work on future generations of 3D Xpoint or 3D NAND. Seemingly part of this arrangement was a cost plus pricing agreement on a certain amount of trade NAND per year. That should be going away which should help counter declines in NAND ASPs. Micron and Intel have had this JV for more than a decade through at least three major downturns in NAND pricing. Having more control of pricing its NAND to Intel should be beneficial in this environment.
Micron has never been as focused on value added solutions, which should also provide a counterbalance to declining NAND ASPs.
As highlighted if FQ2 results, NAND margins actually improved (despite sell-side expectations to the contrary) due to a better mix of higher margin memory solutions (including managed NAND, SSDs, Multichip Packages). This has been a concerted effort by Micron’s new CEO and should meaningfully benefit its NAND business over the short and medium term.
Micron’s stock under-appreciatedrelative to the remarkable operating performance it has delivered over the last two years. If Micron’s stock dramatically under-appreciated, it only makes sense that if we experience anything but a dramatic decline in product ASPs, that its stock should under-depreciateas well.
At its peak in fiscal 2015 (the last peak year of its last up-cycle) Micron traded at $36 per share. That year it delivered $2.71 in EPS. This year, which will wrap up in November, Micron is likely to deliver $11.72 per share and the stock only trades <$45. Despite EPS that will be +330% greater than back in 2015, a balance sheet that’s night and day in comparison and a huge stock repurchase program just being rolled out, the stock trades only 25% higher than its fiscal 2015 peak!
Quick Update on IMFT JV: Impetus for Strategic Action Appears to be Building
Two highly relevant articles have come out of Asia over the past couple weeks that if nothing else provide a stalking horse situation around the IMFT JV. On September 6, the Nikkei Asian Review published an article entitled “Deal-shy chipmaker TSMC says it open to memory acquisition.” In that article it discusses how Taiwan based Nanya would be a logical target as would a potential JV with Micron. Even more relevant was the article published in the EE Times on September 12 entitled “TSMC: Chip Scaling Could Accelerate” referencing the same presentation TSMC made at Semicon Taiwan. Why the EE Times article is more relevant is it drives home the strategic rationale for closer collaboration/integration between logic/foundry and memory semis. Back at the 2018 Lam Research investor day, Lam’s CEO forecast that in coming years the market will no longer be delineating between logic, foundry and memory semis. The technological roadmap for the greatest increase in performance is for logic and memory to be built on top of one another; which could lead to a 100-fold increase in performance.
Now consider that endgame in the context of the current state of the market and some potential catalysts that are emerging in the coming months. Intel could very well try to go it alone and marry 3D Xpoint (inferior, but high performing alternative to leading edge node DRAM) with its CPUs. That said, as alluded to previously Intel’s memory semi manufacturing capacity is not nearly of the scale of Micron and it is bereft of DRAM. Not only is Intel, arguably under-scaled in memory semi manufacturing, Intel runs the risk that any day now it could be informed by Micron that Micron plans to exercise the option to rest 100% control of the Lehi, UT plant on January 1, 2019.
“Only the Paranoid Survive” Intel must also have observed, with some trepidation, as Micron was rolling out videos at its Investor Day with Nvidia’s CEO Jensen Huang extolling the virtues of working with Micron and how important integrated memory is to the future of AI. BTW for those not familiar, TSMC is the manufacturer of Nvidia’s AI chips.
So as we march toward the end of 2018, there are reasonably four obvious parties that should want to either cement a new JV or outright buy Micron: TSMC, Intel, Nvidia and. . . the dark horse . . .Apple.
The idea of Apple may sound like its coming from left field, but its worth ruminating upon. Apple is one of, if not the, largest fabless semiconductor manufacturers in the world by virtue of all of the logic chips it designs for its mobile devices. Apple is probably the second largest memory customer in the world only being behind Samsung – which is not only, in all likelihood, the largest memory semi customer in the world but also the largest producer. Might Apple be interested in negating one of the advantages that Samsung has in having its own captive memory supply (and also in turn be able to stop sending all those dollars to its largest competitor as Samsung happens to be Apple’s biggest vendor)?
Most have given up on big semi combinations being able to happen in the wake of the scuttled Broadcom/Qualcomm and Qualcomm/NXPI deals. But if China and US trade tensions ratchet down, the game for the right deal could well be back afoot. Further, new JV’s that could be very beneficial to Micron seem eminently possible. Watch this space.
Three end notes and then an addendum on the WFE stocks:
1) We haven’t heard from Morgan Stanley lately on Micron and it would be unsurprising if they didn’t land in “bucket 3” when they publish their next Micron update.
2) There have been rumors from multiple outlets that Intel is having trouble keeping up with demand for its processors for PCs and that might leave the PC industry unable to fulfill 5% of the demand that they’d like to have filled by Intel in Q4. Some analysts have then extrapolated that to saying that this will be bad for memory semis because it will lead to 5% less demand for PC memory in CQ4. Two rebuttals: (1) Intel isn’t the only logic company in the world and AMD is surging (partly due to these Intel execution issues) – so just because Intel supply might be constrained doesn’t mean that the PC makers can’t go elsewhere and (2) even if PC/notebook sales were 5% constrained in Q4 given that PC demand is only ~20% of memory semi demand that would thereby impact memory semi demand probably less than 1%.
3) Micron earnings come out next Thursday after the close (Sept 20). I expect and in-line quarter and that FQ4 guidance should also be in line. For a stock trading at less than 4x earnings – that should be more than enough – but any overly negative talk on ASPs is not likely to be well received.
WFE related Addendum
The WFE space deserves a similar write up and when I have time I plan to do something. But for the moment I’ll highlight these few items:
– Foundry spend is highly likely to be up in 2019 as indicated by previously reported pushouts into 2019 by industry leader TSMC.
– Intel’s problems in delivering supply to meet high PC CPU demand combined with difficulties migrating to the 10nm node would seem to have one obvious solution – put a lot more money behind the problem. It will not be surprising if we see healthy Capex demand growth for Intel in 2019 over 2018, which is already looking like a record Capex year for Intel.
– Due to throughput capacity loss at each node migration and expectation for DRAM growth of 20%+ in 2019, I’m seeing forecasts for the need for another 10% increase in DRAM wafer supply in 2019. Presumably there will also need to be more re-tooling to keep on converting prior nodes up to 1X and 1Y. None of these node migrations require EUV.
– If NAND pricing drops as quickly as the sell-side expects, demand should explode and very quickly (by mid-2019?) we could be seeing the need for significant new NAND supply. Not only do the major memory semis believe in the long-term outlook for NAND, I believe there’s a deeper game of chess going on here. While there will inevitably be short spurts of putting on the breaks, I believe that Samsung, SK Hynix and Micron are more than comfortable spending on NAND at a breakneck pace with the ambition of doing in the NAND industry what they effectively did in DRAM: reduce the players down to as tight an oligopoly as possible. So not only is node migration a competitive and economic imperative, it, in all likelihood, leads to a much more healthy long-term environment for the remaining NAND players (and has the benefit of limiting inroads by mainland Chinese upstarts). This dynamic should be beneficial for the WFE companies.
– Again this is not to say that there won’t be periods of slowdown in WFE spend, but this one appears to be a two to three quarter slow down to a jog before the next spurt. The best part of it is that those like Applied and Lam are taking advantage of stock weakness by doing record levels of stock repurchase. The secular trend, I submit, will be the long-term investor’s friend.
Disclaimers: All information gathered from sources believed to be accurate, but the accuracy of the information cannot be guaranteed. A number of points made in this update are a matter of opinion and opinion is subject to change without notice.