Description

• SK Hynix is the #2 out of 3 DRAM manufacturers. It also produces a substantial amount of NAND Flash, a business the market values negatively and will not heavily be concentrated on in this writeup.

• Ongoing improvement in capital allocation

  • Starting last year, SK group companies, including Hynix, have included stock Px appreciation and mkt cap appreciation into their list of main KPIs for senior exec promotions and bonuses.

    • In addition, traditionally-stingy stock-option grants to senior execs have recently become much larger, causing senior execs to care deeply about the stock Px for the 1^st time.

  • SK Hynix has just pre-announced a buyback of 3.1% of shares out during Q3. This is the first material buyback in three years.

  • Currently the dividend payout ratio is quite low – just 7% of trailing pro-forma NI this year, 18% last year, and 9% the year before. This could see a big jump if/when SK Telecom and/or SK Holdings – the largest shareholder and the ultimate parent – decide they need to tap SK Hynix’ cashflow to use for SK group’s myriad growth initiatives.

• DRAM – “it really is different this time”

  • The 3 main DRAM manufacturers now produce 95%+ (probably 96-97%) of global DRAM bits. The junior Taiwanese players, chiefly Nanya, continue to shrink as % of output. While Nanya has an outsized impact on the quoted DRAM spot market Px on InSpectrum or DRAMExchange, market participants on the buy, sell, and broker/consultant sides all say that spot Px now has very little to no impact on contract-Px negotiations.

  • DRAM bit supply growth is relatively constant at low 20%s YoY, despite big increase in CapEx $ recently. This has led to a stable ASP environment, heralding the potential end to boom/bust cycles of the past, where wild swings in ASP – similar to the NAND market today – were driven by volatility in supply growth rate.

Tokyo Electron total industry CapEx (WFE) estimates by vertical. Note that DRAM CapEx has significantly blown out since 2016, yet the supply growth rate has barely accelerated and the DRAM pricing environment has remained benign (SK Hynix received price/bit +4% sequentially in Q2 and expected to increase again in Q3).

Also note that Tokyo Electron, one of the key DRAM equipment suppliers and thus possessing likely the highest internal-projection quality in the sector, thinks that DRAM equipment spend in 2019 will be way down vs. 2018, despite SK Hynix itself significantly increasing DRAM equipment spend due to production-line deployment in the under-construction Wuxi DRAM fab.

This is due to expected equipment CapEx cuts YoY by both Samsung Electronics and Micron, as both those companies will be running low on DRAM-eligible cleanroom space, want to increase their own FCF (for buyback/dividend/cash build), and want to increase CapEx in other spaces (Micron – NAND and potentially Xpoint; Samsung – NAND and potentially Logic/OLED).

• Robust demand environment expected for DRAM over the near, mid, and long terms

  • Currently, server DRAM – mostly destined for the public cloud – is growing bit consumption by well over 40%/year. This will not slow down and many demand checks reveal that public-cloud DRAM demand may potentially even accelerate, given that Google’s TPUs are much more DRAM-hungry than the Nvidia GPUs that they supplant.

  • Assuming that Facebook, Apple, likely MSFT, and potentially AMZN all progress with their own TPU programs 2-3 years behind GOOG, this sets up a multiyear “supercycle” for DRAM demand as both public-cloud capacity as well as DRAM penetration within public-cloud spend both may surge upward.

  • SK Hynix said on the Q2 call that U.S. public-cloud players are both upsizing orders as well as trying to pull Q4 orders forward to Q3, and Q1 orders forward to Q4.

    • This makes sense in the context of:

      • GOOG Q2 core CapEx up teens % vs. Q1 (ex Q1 skyscraper buy) and doubled YoY

        • Signaled further YoY CapEx growth in 2H ‘18

      • FB Q2 CapEx up by 25% sequentially and well over double YoY

        • Notably, FB guided for the next two Qs of CapEx to be ~25% above even Q2 levels…and then 2019 CapEx going up even more.

      • MSFT Jun Q CapEx up high teens sequentially and 20-25% YoY

        • Signaled further CapEx growth in FY Jun ‘19

      • (AMZN Q2 CapEx did print anemically, barely up YoY, but the CFO framed on the call as lumpy spend with a high 1H ’17 base that may well re-accelerate, given the high-40% revenue growth.)

    • Compared to on-premise servers, the public-cloud players use significantly more DRAM per box given their desire to improve performance and reduce customer latency. The hyperscale players have the technical ability to do aggressive caching and even some in-memory processing (separate from Google’s unique TPU DRAM needs).

  • As of today, DRAM consumption is very roughly a 3-way split between PCs, servers, and smartphones/tablets. There are some very small specialty spaces like auto and embedded/industrial that are growing rapidly, but from a very low base.

    • Thus, driverless cars will eventually be a material consumer of DRAM but not for another 5 years or so. By our calcs, they will be a much more important driver for NAND demand given the enormous amount of uHD onboard map storage and video storage.

    • Server – addressed above, given that most incremental server installs are by one of the U.S. or Chinese hyperscale players. It is highly doubtful that these hyperscale players see their end-consumption growth decelerate meaningfully, nor that the hyperscale players lower their server-refresh rate or DRAM specs.

    • Smartphones/tablets – in September, AAPL will bump iPhone RAM upwards by 1GB per unit. This is necessary for graphics performance as well as onboard AI/ML and improved caching capabilities.

      • In China, DRAM has become a key headline spec for the Android OEMs. The Chinese market is substantially more-competitive at the mid and high ends compared to Europe or the U.S. Thus, major OEMs are spec’ing even upcoming midrange devices with 10GB of RAM (from max 6GB today), in order to offer a superior multitasking and graphics experience.

      • Among major global flagships, Samsung will spec its expected 2019 foldable phone very heavily with RAM, given the large screen and ultra-high resolution.

    • PC – YoY PC shipments were positive in Q2, for the first time in several years. PCs are slowly up-speccing on DRAM – too slowly, as those of us who use Chrome browser on our laptops would attest.

    • Networking – this is a fourth, smaller major bucket of DRAM consumption today. However, Networking DRAM is growing rapidly due to the explosion in demand for low latency and high speeds by public cloud players, their end customers, and other services such as Netflix and YouTube.

      • We believe there will be a few important incremental accelerants to Networking DRAM demand.

        • 5G mobile rollout (and to a far lesser extent mmWave 5G) – millions of 5G small cells will be rolled out globally from 2020-2026. They will attempt to bring seamless connectivity to driverless cars at sub-10ms latency (possibly lower) and 100 Mbps+ end-speeds (possibly higher). In order to handle high loads of data at ultra-low latencies, each of the millions of 5G small cells will need an astoundingly high amount of DRAM to facilitate the rapid data up/down links. Routers and switches on the back end of the 5G networks, many located small data centers on the Edge of networks, will also require substantial DRAM.

        • Prior to ubiquitous 5G coverage, it is likely that V2V (vehicle-to-vehicle) communication protocol will be built into driverless cars. V2V will allow cars to rapidly communicate safety issues, map changes, distress signals, etc. due to the at-times high latency and low speed of LTE (average TMUS speed today is ~10 Mbps). This will drive additional DRAM demand in the mid term, albeit less than the global 5G-mobility rollout. 

Overall, I believe underlying DRAM demand can continue growing at today’s rate for several more years. At the same time, the bit-output uplift of node shrinks keeps declining (unlike for NAND) while incremental cleanroom builds are relatively limited, at seemingly a single-digit annual % of the overall base. The overall base of cleanroom space is very large, permits and land plots are difficult to acquire, and incremental build costs are expensive.

For 2019, we’re hearing that forward pricing on server DRAM is strong, while smartphone DRAM indications are slightly negative given potential lack of AAPL RAM content bump-up next year.

• On net, we believe the server DRAM pricing strength – driven by price-inelastic public-cloud players - will likely offset weakness in smartphone or PC, especially considering that node shrinks still push manufacturing cost per bit (including depreciation) down gradually over time.

• Game theory dictates that Samsung will not ruin the DRAM market this time around

  • As is well known in the industry, Samsung sparked the last 3 DRAM ASP routs (2015, 2012, 2009) by investing heavily in CapEx for 1-2 years and flooding the market with supply.

    • Samsung did this because it was afraid that junior Taiwanese and Japanese fabs were getting too profitable and could potentially afford to make big CapEx and R&D pushes. However, these players have now exited the market (Nanya is the only one left, and it’s almost out of cleanroom space with little probability of a greenfield).

  • DRAM was not a major piece of their overall profit pie for Samsung, historically. However, DRAM now represents the majority of Samsung earnings. Samsung would be foolish to spark a 1+ year bear market in DRAM ASP, as this would severely hurt the Samsung stock Px, further emboldening the current activist investors in the name (e.g. Elliott).

    • Currently, the Korean stock market is worried about pressure from the Korean gov’t on Samsung to invest CapEx domestically. But we believe it’s likely that Samsung elects to instead deploy that CapEx spend into OLED production – where Chinese competition is way further along than in DRAM – Logic (where Samsung has once again fallen behind TSMC), and NAND (where industry bit growth is well over 40%, the Chinese are slightly more-threatening than DRAM, and there will be huge future demand from driverless cars and public cloud transitioning from HDD to SSD).

  • Samsung has historically been somewhat sensitive to its market share in DRAM. We have heard several times over the years that, internally, they don’t want their DRAM market share (unclear if they mean bits, revenue, or wafers) to exceed 50% for an extended period of time. This would qualify Samsung for “market dominance” by some definitions, making Samsung a more-likely target for the ire of foreign politicians, antitrust regulators, and major customers.

    • Currently, we believe Samsung is already half the market by revenue, roughly half by bits, and approaching half the market in terms of wafers. Thus, we think Samsung is unlikely to sustain high DRAM equipment spend into 2019.

  • Micron has signaled they will be cautious in terms of DRAM equipment spend, particularly on new wafer capacity, over the next 1-2 years. They are also running very short on cleanroom space, and no large new DRAM shells seem to be in the offing. Instead, the new MU CEO is signaling a renewed focus on increasing buybacks and the dividend.

• Weak KRW and RMB help the I/S incrementally

  • Contract pricing for DRAM and NAND is typically in USD. A high amount (though not all) of the OpEx, COGS, and CapEx are in KRW, with some in RMB.

  • KRW averaged 1080 in Q2, vs. 1118 today. RMB averaged 6.38 in Q2, vs. 6.81 today.

     

• Extremely low valuation multiples given good DRAM outlook, even giving the NAND segment negative value

  • At 58t KRW TEV today (adj for the 15% stake in Toshiba memory biz at its 4t KRW cost), SK Hynix trades at 3.6x actual run-rate Q2 earnings, ~3.4x seasonally-adj run-rate, and an even lower multiple of ~3x seasonally-adj 2H earnings, given the strong Q3 guidance.

  • While we believe the D&A on the I/S is understated vs. steady-state CapEx, clearly the enormous 8.8t KRW in 1H ’18 CapEx is also inflated vs. a more-normalized pro-forma CapEx assumption. E.g. CapEx was 9.6t KRW in all of 2017.

    • From mid-2015 to mid-2017, SK Hynix was not spending CapEx on any cleanroom shells. Then, starting in mid-2017, SK Hynix commenced construction on not one but two greenfield cleanrooms (one NAND, one DRAM). CapEx has blown out due to simultaneous buildouts of these cleanrooms, including very rapid progress on the shell to be used for NAND. Unlike equipment CapEx, shell CapEx does not yield revenue or EBITDA uplift until equipment is deployed ~2 years after the buildout begins.

  • While arguments over “relative competitiveness maintenance” or “EBITDA maint”-level CapEx can become deep and philosophically-charged, we estimate that a robust steady-state CapEx est is somewhere in the range of 11.5t-13.5t KRW for 2018.

    • Even on an actual FCF basis – Q2+Q3E run-rate (based on Q3 guidance numbers from the call) ignoring working-capital - you’re paying 7.5x even giving the NAND biz well under zero value (NAND is barely above breakeven on an IFRS OP level, and is seemingly the majority of CapEx at the moment).

    • Using the high end of the proposed steady-state CapEx range, SK Hynix trades for 6x Q2, 5x Q3E, and 4x our NTM est of owner’s cash earnings. (Even assuming accelerating NAND price headwinds, given ongoing oversupply in NAND.)

  • Mgmt has guided for Q3 DRAM received price to increase, and Samsung mgmt as well our own channel checks show similar expectations.

    • Mgmt has also guided for above-seasonal Q3 DRAM volume growth for SK Hynix – the implication is that AAPL drew down its large Component inventory during CQ2, but will be accepting shipments again in CQ3 for the new iPhones.

  • Mgmt color augured toward strong Q4 DRAM ASP as well. Samsung color is for flattish ASP, while our channel checks indicate that Q4 DRAM received pricing will likely increase a bit vs. Q3.

     

• NAND as a free option

  • As stated above, NAND is bleeding huge amounts of cash, making the rough SK Hynix TEV to 2018 DRAM owner’s earnings multiple around 2-3x.

  • I and most other analysts are negative on NAND pricing over the next 1-2 years, due to the extremely high supply growth spurred by the very-efficient 64-layer production node.

  • However, given upcoming secular demand drivers such as autonomous cars spec’d with 500GB++ SSDs and public cloud players (mainly on HDDs today) going to SSD environments, I think there’s a reasonable chance that NAND demand grows enough on a 3-4 year timespan that NAND market-price declines moderate and the NAND segment returns to maint-cashflow breakeven or better.

  • This would catalyze a huge valuation stepup for Hynix, given that the implied NAND value today is in the negative tens of billions of USD. (Our rough estimate is that the market ascribes -15t to -20t KRW in value to the NAND unit, given negative FCF of many billions of dollars this year.)

  • Remember, WD bought SanDisk a short two years ago for $19b. And a consortium led by Bain just closed on Toshiba’s NAND unit for $18b last month (admittedly, market conditions have deteriorated rapidly since the price was agreed upon in Sept ’17).

  • If and when NAND gets back to $0 or greater value, SK Hynix’ market cap will jump significantly. Even getting back to NAND segment value-neutral would be ~30% accretive to the SK Hynix overall TEV.

    • At worst, given that aggressive industry DRAM CapEx will likely entail some pullback on NAND CapEx, the NAND segment’s value may serve as a counter-hedge to the DRAM segment’s value.

      • And I think that the NAND segment’s negative market valuation cannot go much lower, given that investors should have a limit to the negative value the ascribe to the unit (given theoretical option value in a rapid demand-growth environment).

• Target price

  • Overall, the target price is highly dependent on 1) the value you personally ascribe to the NAND segment, 2) the multiple that the DRAM segment deserves.

  • SK Hynix is roughly the size of MU in the NAND space, though growing faster. NAND is doing slightly over 9t KRW in run-rate revenue. Potentially this product could someday see owner’s FCF margins in the 10-15% range.

  • We ignore the SOTP analysis, admittedly valuing NAND highly negatively. We believe that owner’s cash earnings will continue to grow into 2019, to a 15t KRW seasonally-adj run-rate a year from now.

    • With a rapid increase in buybacks as well as potential dividend stepups on the horizon, we believe SK Hynix deserves a 7.5x multiple on our year-out est, potentially/arguably “peak,” 15t KRW owner’s cash earnings.

    • 7.5 * 15 à 112.5t KRW tgt TEV à 158k/existing share. Add in some cash build (at a discount), small dividend value, and sharecount shrink through buybacks - 3.1% in Q3 and likely more beyond that – and you get to 172k target Px per share in a year, a 1-yr double from the current share Px.

Catalyst

-Improving capital allocation

-Good Q3

-DRAM pricing