September 9, 2018 | Equity Research
Intel 10nm Ice Lake vs. AMD 7nm Rome: Semiconductors & Server CPU Share Shift Ahead Semiconductor Cap
Equipment $ Rating Price FY EPS Price Target 2018E 2019E Ticker Curr. Prior 09/07/18 Curr. Prior Curr. Prior To From Semiconductors & Semiconductor Cap Equipment, Market Weight AMAT 1 V NC 39.85 4.46 NC 4.22 NC 60.00 NC AMD 1 V NC 27.38 0.50 NC 0.67 NC 34.00 20.00 AVGO SR NC 232.58 NE NC NE NC NE NC INTC 1 NC 46.45 4.18 NC 4.29 NC 55.00 NC MU 1 V NC 44.86 11.69 NC 10.24 NC 63.00 NC NVDA 1 V NC 271.86 4.92 A NC A 7.96 NC 315.00 NC
Source: Company data and Wells Fargo Securities, LLC estimates
1= Outperform, 2 = Market Perform, 3 = Underperform, V = Volatile NA = Not Available, NC = No Change, NE = No Estimate, NM = Not Meaningful
What Matters: In this report we discuss: (1) Intel’s elongated 14nm process node (19+ quarters vs. 8 and 13 quarters with 32nm and 22nm, respectively), or more importantly, our thoughts on the competitive implications of Intel’s well-known delayed 10nm server CPU transition – Intel’s 14nm+ Cascade Lake and Cooper Lake CPUs + Optane DC Persistent Memory solutions vs. AMD’s 14nm Naples EPYC and forthcoming 7nm Zen 2 based Rome EPYC CPUs. In addition to AMD also outlining the 7nm+ Milan; as well as 5nm offerings looking into 2020 (see MTr/mm2 & Transistor Gate Pitch comparison on page 5). (2) Our thoughts / model scenario analysis based on AMD’s ability to gain server CPU share into 2019 and 2020. Our industry checks highlighting strong / accelerating interest in AMD’s 7nm Rome EPYC CPUs – leaving us to consider a 15%+ ship share with sustainable $600+ ASP ($/unit) into 2020. (3) Intel’s increasing capex and construction-in-progression balance and the implications on depreciation expense – our Intel capex-to-depreciation waterfall model implies a significant depreciation expense increase looking into 2019 and beyond (see pgs. 12-13) Our model illustrates a high historical correlation with reported depreciation expense (R2 = 0.994).
Well’s AMD Server Share Scenario Analysis: 35% Non-GAAP EBIT / EPS Upside (vs. Street); Focus on Negative Mid-Single CAGR Decline in Intel’s DCG Platform Revenue. On pgs. 9-10 we provide a 2020 server CPU share scenario analysis in which we conclude: (1) AMD’s ability to capture a mid-teens server CPU ship share, at ~$600 ASP estimate, equates to ~$2.5B/annum in revenue (vs. our C2018 at ~$420M). Applying this to our current (unchanged) estimates implies a 35%+ EBIT / EPS upside. With a positive view on AMD’s ability to drive +$2B/annum in server CPU revenue Aaron Rakers, CFA (coupled with forthcoming 7nm GPUs and a belief that investors Senior Analyst|314-875-2508 could start to consider a next-gen game console cycle as a [email protected] positive on AMD’s Semi-Custom segment), we’re increasing our Joe Quatrochi, CFA AMD target price to $34 – driven by upside momentum vs. Associate Analyst |314-875-2055 current estimates. (2) A declining in Intel’s server CPU ship share [email protected] from ~98% currently to 85%, coupled with some downward ASP Jake Wilhelm, CPA ($/unit) pressure amid increased dual-socket competition, leaves us to Associate Analyst |314-875-2502 [email protected] question a negative mid-single CAGR in Intel’s DCP Platform revenue through 2020 – focus on AMD’s insertion / traction in public cloud (~43% of Intel’s DCG revenue).
Please see page 16 for rating definitions, important disclosures and required analyst certifications. All estimates/forecasts are as of 09/09/18 unless otherwise stated. 09/09/18 20:15:26 ET
Wells Fargo Securities, LLC does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of the report and investors should consider this report as only a single factor in making their investment decision. Semiconductors & Semiconductor Cap Equipment Equity Research
Discussion
Intel’s 10nm Delays to Remain a Significant Investor Sentiment Headwind; Persisting Questions over Competitive / Financial Ramifications
Summary Notes:
Intel has noted that PCs using 10nm CPUs would be available for the 2019 holiday season; 10nm server (Ice Lake Xeon) MPUs are expected shortly after (early or mid-2020?). Intel had initially (2015) expected 10nm mass production volume ramps to materialize in second half of 2017.
Performance per Core and / or Performance / $ Matters. Some of our conversations with systems vendors
Intel management has suggested that its initial 10nm vs. 14nm scaling expectations were too ambitious, leaving us / investors to focus on: (1) the achieved logic transistor scaling vs. initial 2.7x target; logic density scaling at 0.37x for 10nm vs. 14nm – hyper-scaling enabling sustained levels of higher scaling relative to pre-14nm trend at 0.5x every two years, and (2) Intel’s 10nm transistor density versus previously noted 100.8 MTr/mm2. This also compares to reports that TSMC and Samsung’s competitive 7nm chips would have transistor densities in the 125+ MTr/mm2 and 115+ MTr/mm2 ranges.
Intel is using quad-patterning immersion lithography for 10nm, noting that this would be the last process node prior to moving to EUV.
TSMC commenced commercial production of 7nm FinFET based chips in August 2018; GlobalFoundries announced that it was abandoning its 7nm efforts, resulting in AMD’s decision to move all of its 7nm CPUs and GPUs to TSMC.
TSMC outlined 5nm plans in early-May 2018 with volume production commencing sometime in the 2020 timeframe. The company will be using EUV lithography (ASMLs’ Twinscale NXE:3400 EUV step + scan systems; the company’s CLN5 technology). In June it was reported that TSMC would invest $25B for 5nm production.
Samsung is using EUV at 7nm with risk production expected to commence in late-2018; volume production expected to ramp in 1H2019.
We think 2019-2020 will be an inflection year for the semiconductor industry as we / investors gauge the industry’s ability to effectively / efficiently scale to 7nm and beyond as we move from multi- patterning immersion to EUV lithography. Samsung is planning to use all EUV lithography at 7nm, while TSMC plans to implement some EUV lithography in non-critical layers at 7nm+ and then full EUV at 5nm.
In addition to leveraging EUV to enable the next level of chip scaling technologies such as Gate-All- Around (GAA) FETs, Multiple-Gate Field-Effect Transistor (MuGFET), nanowires, etc. will become increasingly visible in 2020 and beyond.
While Intel’s 10nm struggles have become very visible (vs. claiming a 3-year leadership positioning in early-2017), the extent / timing of the required efforts to resolve the technological challenges (i.e., achieving high volume yields), the competitive consequences, and financial model ramifications remain significant unknowns. Offsetting this, however, is the view that Intel’s portfolio breadth and economies of scale to address increasing customized hyper-scale demands, and the dynamic in which overall system performance is increasingly driven by solutions outside / adjacent of the CPU, will remain core competitive advantages. With this in mind, we would highlight the following thoughts:
Competitive headwinds as we see investors now considering AMD’s ability to garner a 20%+ share in the server CPU market (note: AMD was able to achieve a 25%+ server CPU ship share in mid-2006 with Opteron). Additionally, Intel’s ability to drive a solutions selling motion (e.g., Cascade Lake + Optane DC Persistent Memory, network connectivity logic) will be important as we / investors question increased spend competition from the need for GPU acceleration, FPGAs, and customized ASICs. The server market that has become increasingly concentrated by hyper-scale vendors with a robust appetite for the highest performing and increasingly customized solutions. Intel recently (early- August) highlighted that custom CPUs accounted for ~50% of its cloud service provider volumes in 2017 (vs. ~18% of volume in 2013).
2 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
The capital intensity of next-generation process node transitions has been widely discussed; however, the unknown / un-quantified model implications remain a key point of investor uncertainty. Concern over gross margin pressures will likely persist - increased competition (price compression), revenue mix shifts, and capital intensity (ongoing 14nm capacity investments, 10nm ramp, and Intel’s significant construction-in-progress balance). From a capital intensity perspective, we would note that IBS has estimated that the 10/7nm process node is 1.3x more capital intensive than the 16/14nm process node, while the 5nm process node could be as much as 3.1x more capital intensive than the 16/14nm node.
In addition to Intel’s competitive positioning versus AMD’s 7nm Rome EPYC processors and competing against NVIDIA’s focus on participating in the inferencing piece of the AI market, we think investors will be focused on the impact of Intel’s slower / delayed transition to the 10nm process node on the company’s custom Foundry business, which the company has highlighted as remaining “a strategically important part of our customer engagement portfolio.”
“Our [Intel’s] ability to advance Moore’s Law to make products less expensive and more capable year in and year out, is really our core competitive advantage.” This statement was made by Stacy Smith, Intel’s EVP of Manufacturing, Operations, & Sales at Intel’s Technology & Manufacturing Day event in early-2017. Mr. Smith retired from Intel later in August 2017. At the same time, Intel noted that the time between process node transitions has gotten longer, resulting in Intel’s focus on Hyper- Scaling utilizing multi-patterning manufacturing processes. In early-2017 Intel highlighted a strong competitive positioning at 10nm, reflective of the company’s Hyper-Scaling architecture development to enable greater density increases. The company compared its 100.8 MTr/mm2 (their proposed density calculation metric) versus TSMC and Samsung’s 10nm process nodes at 48.1 and 41.6 MTr/mm2. However, Intel’s elongated, and now increasingly questionable, transition to 10nm with Ice Lake, and anticipated increasing questions over Intel’s plans post 10nm raises competitive concern and investor angst over Intel’s business model over the next several years. The key issue is that longer times between process nodes in early 2017 was noted as being three years, which now appears to be closer to 4.5 to possibly as long as 5+ years.
Intel initially announced in July 2015 that it expected its first 10nm product (Cannon Lake) to launch in second half of 2017. The launch date was continually pushed back over quarter after quarter with the company working to move the narrative away from node size after promoting it as a performance measure for years toward ”Hyperscaling” which keeps Intel on Moore’s Law by slowing down the number of shrinks but improving performance more per shrink – the company’s 14nm, 14nm+, and 14nm++ technology transitions. Intel has now been working with its 14nm process node and various iterations of it since 2014. Intel believes that it will be able to rely on its 14nm in the interim as it slows the ramp of 10nm. It’s important to note that Intel’s 14nm chips are similar in performance to competitors’ 10nm from other foundries and its 10nm is approximately equivalent to a 7nm. However, it has now become increasingly apparent that Intel could lag AMD’s Rome 7nm EPYC processors, via TSMC, by up to 12 months, while at the same time investors will question TMSC’s ability to execute to 5nm vs. Intel’s 10nm+ and 10nm++ progression.
Intel has begun shipping low volumes of its 10nm processor with volume ramp delayed from previously expected 2H18 to 2019 as yields didn’t meet expectations (SemiAccurate recently reporting a 8%-10% vs. a desired initial 50%-60% yield at this point). During the company’s 2Q18 earnings call the company noted that the 10nm desktop CPU based systems would be available for the 2019 holiday season and that data center (server) CPUs would be available shortly after – implying early / mid-2020. The company cited high transistor density and heavy multi-patterning use as drivers of the low yield. Intel’s 10nm process node is the last transition prior to requiring the move to EUV. During the 1Q18 earning call, Intel noted that it was probably too aggressive in the architectural assumptions for 10nm vs. 14nm with an expected 2.7x logic transistor density improvement (vs. 2.1x and 2.5x for 22nm vs. 32nm and 14nm vs. 22nm, respectively).
SemiAccurate reported that the shipped 10nm product (shipped in a Lenovo IdeaPad 330 abroad) had significant issues – with massively inferior price/performance compared to the company’s 14nm product – and was shipped mainly as a PR win for the company. SemiAccurate’s recent (8/2/18) article suggest that Intel’s 10nm challenges could result in a decision to abandon the process node to 12nm, which would be a major sentiment blow given Intel’s strong track record of industry firsts – strained silicon transistors, gate-first process flow, high-k metal gate, FinFET, self-aligned double and quad-patterning.
At this point, Intel doesn’t believe that problems that it has encountered at 10nm will impact 7nm production; however, the ramifications of a slipping 10nm transition raises increased questions over this timing relative to TSMC’s 5nm process node. This is mostly due to the fact that Intel believes that 10nm will be the last size that it produces without EUV and the fact that the scaling factor from 14nm to 10nm is much more aggressive than 10nm to 7nm (typical industry scaling factor of 1.5-2 vs. Intel at 2.7).
Wells Fargo Securities, LLC | 3 Semiconductors & Semiconductor Cap Equipment Equity Research
However, we think that with TSMC’s continued progress on its node ramps puts Intel’s long-time manufacturing dominance at risk.
As shown in the Process Technology Comparison table below, we believe that Intel is in the process of losing a meaningful portion of its competitive advantage in terms of chip density.
TSMC announced in April that it has started high volume production of chips with its first-gen 7nm. TSMC will use existing lithography tools instead of moving to EUV and its high-density SRAM cell will measure 0.0272 microns.
Samsung said in early July that its Arm Cortex-A76 chips will use 7nm process technology with a ramp in second half of 2018, while it expects its 5nm mid-2019 utilizing EUV. Samsung’s 7nm process has a 27m pitch, 54nm gate pitch, and high-density SRAM cells at 0.02622 microns.
In late August it was announced that GlobalFoundries was abandoning its efforts to move to the 7nm process node. The company had initially planned to introduce a 7nm chip using ArFi with quadruple patterning sometime in 2019. The company’s 7LP process has a gate pitch of 56nm, a fin pitch of 30nm, and SRAM cell size of 0.02692 microns.
The logic fabrication industry has seen significant consolidation as the costs of production of ever smaller nodes increases. Gartner for example estimated that it will cost approximately $270MM to design a 7nm chip vs. $80MM for a 16/14nm chip and $30MM for a 28nm planar chip. As seen in the chart below, each move in transistor size exponentially increases design cost with IBS estimating that the move from 7nm to 3nm could increase costs by up to 5x. We believe this will result in companies staying at each node size longer. Process Technology Timeline Comparison 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018E 2019E 2020E
Intel 90nm 65nm 45/40nm 32/28nm 22/20nm 14nm 10nm
TSMC 90nm 65nm 45/40nm 32/28nm 22/20nm 16nm 10nm 7nm 5nm
Source: Company Data; Wells Fargo Securities, LLC
Intel vs. TSMC & Samsung Intel TSMC Samsung 22nm 22FFL 14nm 10nm 20nm 16nm 10nm 7nm 5nm 14nm 10nm 7nm 2020 High-Volume Production Timing 2H2011 Late-2017 1H2014 2H19 / 2020 2H2014 2H2015 1H2017 1H2018 2H2014 1H2017 1H2019 (target) Fin Pitch (nm) 60 45 42 34 Planar 48 35 27 20 (?) 48 47 27 Transistor Gate Pitch (nm; CPP) 90 108 70 54 90 90 66 54 44 78 68 54 Minimum Metal Pitch (nm; MMP) 80 90 52 36 64 64 44 36 32 64 48 36 Logic Cell Height (nm) 840 540 399 272 576 480 360 270 n/a 576 360 243 Logic Transistor Density (MTr/mm2) 15.3 19.4 37.5 100.8 28.0 29.0 48.1 116.7 210+ (?) 30.5 51.6 127.3 Logic Transistor Density (vs. Intel Comp.) 1.83x 0.77x 0.48x 1.16x 2.08x 0.81x 0.51x 1.26x SRAM Cell (um2; High-Density) 0.0920 0.0880 0.0500 0.0312 0.0810 0.0740 0.0420 0.0270 n/a 0.0640 0.0400 0.0262 Source: Company Data; Apothetech, WikiChip, IC Insights; Wells Fargo Securities, LLC Estimates ** Intel's 10nm trailing TSMC's competive 7nm by 12+ months (vs. Intel's past emphasis of 3+ year leadership); TSMC and Samsung 7nm transistor density (MTr/mm2) comparable or ahead of Intel (varies among different industry sources - some show TSMC and Samsung at 95-100 MTr/mm2 range). ** Intel using 193 immersion lithography at 10nm; TSMC initially using 193 immersion with follow-up use of EUV in non-critical layers ** TSMC's 5nm plans for 2020 launch plan via EUV to become an key focus through 2019 ** Samsung has outlined production plans for 5nm Low Power Early (LPE) process in 2019; 4nm in 2020 timeframe.
4 | Wells Fargo Securities, LLC
5
Intel10nm Ice Lake vs. AMD 7nm Rome:Server CPU ShareShift Ahead
| Wells FargoSecurities, LLC Intel Notes: • 10nm CPUs: (1) Client (PCs): Cannon Lake (late- 2019), (2) Servers: Ice Lake (late-2019 / 2020?) • Initially (2015) Intel was planning on 10nm production to commence in 2016, which shifted to late-2017 and now into 2019. • Key 10nm Technology notes: Self-Aligning Quad Patterning (SAQP), 193 Immersion (SAQP), Single Dummy Gate, Contact Over Active Gate (COAG) • Hyper-Scaling (inter-node scaling) strategy
implemented with 14nm and expected to continue with 10nm (e.g., 10nm, 10nm+, and 10nm++ variants). Hyper-Scaling enabling better than 0.5x scaling per node transition (e.g., 0.37x Logic Transistor Density at 14nm vs. 22nm; expectation of same scaling for 10nm vs. 14nm).
Competition Notes: • TSMC commenced volume production of 7nm in early-2018 (est. 10%+ and 20%+ of wafer revenue in 3Q18 and 4Q18, respectively); TSMC has outlined plans for 5nm early volume production in 2020. • Samsung reportedly started 7nm (EUV-based) production in mid-2018; volume production by end of 2018. Samsung’s roadmap shows 5nm production commencing in 2019 using Gates-All-
Around (GAA) FinFETs. • In Aug. 2018 GlobalFoundries announced that it was discontinuing 7nm development • AMD ramping 14nm-based EPYC Naples server CPUs in through 2018; focus on 7nm (TSMC) EPYC Rome processors in early-2019
Equity Research Equity
Semiconductors & Semiconductor Cap Equipment Equity Research
As a reference, below we illustrate Intel’s estimated server CPU shipments by process node over the past several years – highlighting the significant lengthening in the transition to the next-gen process node for 22nm to 14nm (thirteen quarters) and now 14nm to 10nm ( currently implied at over nineteen quarters).
Transition in the Datacenter: Intel vs. AMD Thoughts; AMD Server Scenario Analysis Leaves us to Consider 35%+ Upside to 2019-2020 Estimates
Summary Notes:
Intel has focused on changing the narrative with regard to its competitive positioning in the Data Center segment – shifting from an emphasis on the company’s time-to-market dominance (e.g., ~3-year leadership at 14nm) to the ability to leverage inter-node enhancements and portfolio breadth to address AI, networking, 5G/IoT, etc.
Intel had a ~98.4% server MPU shipment share in 2Q18, according to Mercury Research. While our checks with systems vendors have highlighted accelerating customer interest in AMD’s EPYC positioning / roadmap in the server market, we would consider expectations regarding AMD’s ability to garner a 15%+ market share as mixed.
Although process node competitiveness is a key focus (emphasis on the ability to drive higher transistor density), we think AMD’s EPYC processor positioning also highlights the competitive importance / positioning of packaging to address dual-socket server market with a single-socket Multi Chip Package (MCP) design. Intel’s forthcoming Cascade Lake-AP family is expected to be the company’s first MCP design.
Intel disclosed that 43% of the company’s Data Center Group (DCG) revenue in first half of calendar 2018 was driven by public cloud service providers (revenue growing +45% and +41% y/y in 1Q18 and 2Q18, respectively), while another 21% of total DCG revenue was derived from communications service providers (revenue growing +33% and +30% y/y in 1Q18 and 2Q18, respectively).
Intel’s Enterprise & Government DCG revenue grew 7% y/y in 1H2018, accounting for approximately 36% of total DCG revenue, according to our estimates. We estimate that Intel’s Enterprise & Gov’t DCG revenue declined by a 5% CAGR over 2014-2017.
We think Artificial Intelligence and High-Performance Computing (HPC) will continue to become a key competitive battleground – workloads increasingly requiring GPUs where Intel has thus far faltered; NVIDIA focused on expanding its dominant presence in AI training into inferencing.
6 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
Intel’s ability to maintain a strong competitive positioning in cloud service providers will be a key focus. Intel has reported that over 50% of the company’s SKUs sold into cloud service providers are optimized with a trend toward customization.
AMD’s traction in cloud service providers will be an important competitive focus. IDC estimates that server revenue in public cloud totaled $4.44 billion in 1Q18, up 58% y/y; representing 24% of total server revenue. During the 2Q18 earnings call, AMD reported that server CPU shipments into mega data center customers doubled and the company is working on qualifications with multiple cloud providers scheduled for deployments in second half of 2018.
We will be focused on the positioning of Intel’s 14nm Cascade Lake (4Q18 ship) and Cooper Lake (mid/2H2019?) + Optane DC Persistent Memory solutions versus AMD’s 7nm Rome EPYC MPUs into 2019
In an effort to shift / change the narrative in its data center segment from the impact / implications of a breakdown in Intel’s long proven track record of executing on Moore’s Law with the 10nm Ice Lake production volume launch now expected in the 2020 timeframe, at the recent Data-Centric Innovation Summit Intel emphasized the following key themes we think investors should consider:
1. Public Cloud + Customization. We think the impact of public cloud and the importance of customization is an important consideration as we / investors gauge the competitive battleground playing out as microprocessor performance per $ (transistor density) accelerates into 2019 and beyond. IDC estimates that compute (server) revenue in public cloud grew to ~$16.0 billion in 1H2018, up over 50% y/y and accounting for +20% of the total server market (up from ~12% in 2014). Intel has highlighted that over 50% of its server CPU SKUs into public cloud service providers is optimized with an ongoing trend toward full customization (versus 18% in 2013).
2. Platform / Portfolio Breadth: We think Intel’s portfolio, or expanding platform approach is a key competitive advantage as public cloud service providers, communications service providers, and enterprises address the full ecosystem of data-centric computing. This includes Intel’s CPU optimization efforts (e.g., DLBOOST and BFLOAT16) and packaging (Intel Mesh Architecture, memory sub-system architecture, etc.), integration with Optane DC Persistent Memory, silicon photonics, Nervana (ship in 2019), FPGAs, and more.
Intel’s Data Center Group (DCG) is comprised of processors and adjacent solutions for the traditional enterprise data centers, cloud and communications service providers. Intel has stated that it predicts its Data Center Group’s TAM will be greater than $70 billion by 2022. We would highlight Mercury Research’s forecast of total server processor revenue of $36.0 billion in 2022 vs. $17.6 billion in 2017. Key drivers of Intel’s data center growth are the emergence of next-generation computing applications such as artificial intelligence, exascale computing, virtual reality, and the exponential growth of data generated by new data sources like autonomous driving, IoT, and smart factories.
While Intel maintains a virtual monopoly on server processers we think it is important to note that AMD has significantly closed the performance gap between its products and Intel’s with the launch of its EPYC processor line. AMD re-entered the x86 market in June 2017 with its 14nm Naples line (EPYC 7000 series) designed around its Zen core architecture. EPYC processors are roughly comparable to Intel’s latest Skylake 2-way server processors in terms of price/performance which we believe will result in AMD taking at least some share from Intel (~99% market share). It is well-known that AMD plans to jump over 10nm processers, moving from its 14nm Naples EPYC chips to 7nm with the future release of its Rome EPYC processors set to sample at year end/ship early 2019 and Milan 7nm+ processors in 2020. We believe that AMD’s competitive positioning vs. Intel in the server market is significantly underappreciated.
While we think it is important to note that SemiAccurate’s research has left us to consider some anti-Intel bias, we have found their research / published articles on Intel’s 10nnm challenges as relatively accurate. That said, research from SemiAccurate has been highlighting the increasingly precarious positioning of Intel’s roadmap versus AMD’s EPYC processors, including:
OEMs have been reporting significantly lower than expected sales of Intel’s Purely chip do to the chips poor price/performance compared to Intel’s Broadwell. Cascade Lake-AP/SP (14nm on Purely) which is set for release to hyper-scalers in 4Q18 and to the rest of the market in 1Q19 will be at more than a 50% per-socket disadvantage to AMD’s Rome EPYC CPUs. SemiAccurate believes that Rome will beat or tie Cascade Lake on a single-threaded basis and will have >2x the cores (reports noting 28 vs. up to 64 cores). Cooper Lake (14nm++ on Whisky Lake platform) – due in late 2019 and according to SemiAccurate when normalized to Purely is approximately 40% faster. This results in a significantly higher TCO and is still due out almost a full year after Rome.
Wells Fargo Securities, LLC | 7 Semiconductors & Semiconductor Cap Equipment Equity Research
Ice Lake (mid-2020) - 10nm+ on Intel Whitley platform using the same LGA 4189 socket as Cooper Lake (enabling quick follow-on transition). SemiAccurate has noted that it expects Ice Lake to be much slower than AMD’s Rome assuming that Intel’s 10nm process is back on track. Additionally, AMD’s 7nm+ Milan is scheduled to be released before Ice Lake with SemiAccurate reporting that Milan will widen the gap for AMD over Ice Lake by at least mid-double digit percentages. SemiAccurate reports that they have access to internal Intel documents which state Intel internally believes that they will not be technologically competitive in the server market until after their Sapphire Rapids product which is set for delivery in 2022 – thereby giving AMD at least four years of performance leadership in Intel’s core market. Additionally, they report that Intel expects that they could lose much greater than the 15-20% market share that former CEO Brian Krzanich admitted was at risk.
AMD EPYC Share Gain Potential – 15%+ Now in Focus? Intel has publically admitted that AMD will take at least some market share from the company with management stating that its goal was to defend its position and not let AMD capture “15-20% market (shipment) share”. AMD CEO Lisa Su has stated that the company is targeting 4-6% of the server market in exiting 2018 with a mid-term target of a double-digit market share. As we think about AMD’s server CPU share aspirations, we would highlight:
We currently model AMD’s Enterprise, Embedded, & Semi-Custom (EESC) segment revenue at $2.430B, $2.750B, and $3.029B in 2018, 2019, and 2020, respectively (approximately 2%-4% above current consensus estimates). As shown below, our scenario analysis based on AMD’s ability to garner an mid-teens shipment share in server CPUs in 2019 and 2020 implies $2.5-$3.0 billion in server CPU revenue; 35%+ non-GAAP EBIT and EPS upside versus current street estimates.
Our EESC estimates currently reflect a server CPU revenue estimate of $0.43M, $1.2B, and $1.95B in 2018, 2019, and 2020, which is based on CPU shipment share estimates at ~2%, ~6%, and ~8%, respectively – using Mercury Research server CPU forecast estimates. We model AMD’s blended EPYC server CPU ASP ($/unit) at approximately $600.
Based on our server estimates, our current model would imply -20%+ CAGR in implied Semi-Custom + Embedded revenue – a level we view as conservative as we see AMD as remaining favorably positioned for next-gen game console upgrades and continued growth in embedded. Our scenario analysis below includes Semi-Custom + Embedded revenue estimates that range from -10% to +20% vs. our 2018 revenue estimate.
As summarized in our Intel vs. AMD Server CPU analysis summary below, a mid-teens server CPU shipment share would leave us to estimate an implied revenue in the ~$2.5-$3.0B/annum range in 2020, or 35%+ upside to our estimates (all-else held equal). With a flat Semi-Custom + Embedded revenue estimate from 2018, we would arrive at an implied AMD revenue upside of 20%+.
As a reference, AMD’s announced EPYC customers include:
OEM / ODM: Cloud Customers: HP Enterprise Microsoft Azure Dell JD.com Cisco Baidu Supermicro Alibaba Cray Tencent Lenovo Packet Sugon Yahoo! Japan Penguin Computing Hivelocity Asus Dropbox Tyan Gigabyte ASI
** AMD reported that it had +50 platforms in the market based on EPYC exiting 2Q18, an increase from 40 EPYC- based platforms in the market exiting 1Q18
** As we focus on AMD’s EPYC ramp into 2019, and in particular with the 7nm Rome EPYC processors looking into mid-2019, we think a key focus will be the company’s ramp in cloud providers.
** Intel has disclosed that cloud service providers (CSPs) accounted for ~43% of total 1H2018 DCG segment revenue; we estimate trailing 12-month cloud revenue grew at more than 35% y/y.
8 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
Wells Fargo Securities, LLC | 9 Semiconductors & Semiconductor Cap Equipment Equity Research
Intel vs. AMD Server CPU Analysis Total Server CPUs 2012 2013 2014 2015 2016 2017 1H2018 2018E 2019E 2020E 2021E Total CPUs Shipped 17,669 18,015 19,351 21,274 23,437 24,893 13,028 27,881 29,414 30,885 32,738 - Y/Y % Change 3.0% 2.0% 7.4% 9.9% 10.2% 6.2% 11.4% 12.0% 5.5% 5.0% 6.0% Revenue ($M) $10,108 $10,174 $12,417 $14,710 $15,530 $17,424 $9,995 $21,466 $21,741 $23,285 $24,188 - Y/Y % Change 7.2% 0.7% 22.0% 18.5% 5.6% 12.2% 27.2% 23.2% 1.3% 7.1% 3.9% - Blended ASP ($/unit) $572 $565 $642 $691 $663 $700 $767 $770 $739 $754 $739 - Y/Y % Change 4.1% (1.3%) 13.6% 7.8% (4.2%) 5.6% 14.2% 10.0% (4.0%) 2.0% (2.0%) 2020E AMD Server CPU Shipment Share (Low to High) Intel Server CPUs 2012 2013 2014 2015 2016 2017 1H2018 6% 8% 10% 12% 14% 16% 18% 20% 22% 24% Intel Server CPU Shipments $540 $1,001 $1,334 $1,668 $2,001 $2,335 $2,668 $3,002 $3,336 $3,669 $4,003 - Multi Socket 816.9 806.5 910.1 828.4 842.6 745.1 309.8 $550 $1,019 $1,359 $1,699 $2,038 $2,378 $2,718 $3,058 $3,397 $3,737 $4,077 - Dual Socket 12,785.4 12,834.8 13,650.4 15,148.8 15,138.9 16,575.4 9,491.2 $560 $1,038 $1,384 $1,730 $2,075 $2,421 $2,767 $3,113 $3,459 $3,805 $4,151 - Single Socket 3,146.3 3,583.3 4,390.1 5,124.5 7,373.5 7,410.0 3,049.1 $570 $1,056 $1,408 $1,760 $2,113 $2,465 $2,817 $3,169 $3,521 $3,873 $4,225 Total Shipments 16,748.6 17,224.6 18,950.6 21,101.7 23,355.0 24,730.5 12,850.1 $580 $1,075 $1,433 $1,791 $2,150 $2,508 $2,866 $3,224 $3,583 $3,941 $4,299 - Dual + Single Socket 15,931.7 16,418.1 18,040.5 20,273.3 22,512.4 23,985.4 12,540.4 $590 $1,093 $1,458 $1,822 $2,187 $2,551 $2,916 $3,280 $3,644 $4,009 $4,373 $600 $1,112 $1,482 $1,853 $2,224 $2,594 $2,965 $3,336 $3,706 $4,077 $4,447 Intel Server CPU Revenue $610 $1,130 $1,507 $1,884 $2,261 $2,638 $3,014 $3,391 $3,768 $4,145 $4,522 - Multi Socket $864 $858 $971 $884 $898 $797 $334 $620 $1,149 $1,532 $1,915 $2,298 $2,681 $3,064 $3,447 $3,830 $4,213 $4,596 - Dual Socket $8,134 $8,157 $10,194 $12,499 $12,991 $15,201 $9,016 $630 $1,167 $1,557 $1,946 $2,335 $2,724 $3,113 $3,502 $3,891 $4,281 $4,670 - Single Socket $787 $896 $1,124 $1,281 $1,625 $1,385 $549 $640 $1,186 $1,581 $1,977 $2,372 $2,767 $3,163 $3,558 $3,953 $4,349 $4,744 Total Revenue $9,784 $9,910 $12,289 $14,664 $15,514 $17,383 $9,899 $650
AMD Server CPU ASP ($/unit) ASP Server CPU AMD $1,205 $1,606 $2,008 $2,409 $2,811 $3,212 $3,614 $4,015 $4,417 $4,818 - Dual + Single Socket $8,920 $9,052 $11,318 $13,779 $14,616 $16,586 $9,565 $660 $1,223 $1,631 $2,038 $2,446 $2,854 $3,261 $3,669 $4,077 $4,484 $4,892 - Intel Total DCG Platform Revenue $11,019 $11,271 $13,300 $14,828 $15,913 $17,445 $9,916 2020E Intel Server CPU Shipment Share (Low to High) Intel Server CPU ASP ($/unit) 76% 78% 80% 82% 84% 86% 88% 90% 92% 94% - Multi Socket $1,057 $1,064 $1,067 $1,068 $1,065 $1,070 $1,079 $570 $13,379 $13,731 $14,083 $14,436 $14,788 $15,140 $15,492 $15,844 $16,196 $16,548 - Dual Socket $636 $636 $747 $825 $858 $917 $950 $585 $13,731 $14,093 $14,454 $14,815 $15,177 $15,538 $15,900 $16,261 $16,622 $16,984 - Single Socket $250 $250 $256 $250 $220 $187 $180 $600 $14,083 $14,454 $14,825 $15,195 $15,566 $15,937 $16,307 $16,678 $17,048 $17,419 Total Revenue $584 $575 $648 $695 $664 $703 $770 $615 $14,436 $14,815 $15,195 $15,575 $15,955 $16,335 $16,715 $17,095 $17,475 $17,855 - Dual + Single Socket $560 $551 $627 $680 $649 $691 $763 $630 $14,788 $15,177 $15,566 $15,955 $16,344 $16,733 $17,123 $17,512 $17,901 $18,290 $645 $15,140 $15,538 $15,937 $16,335 $16,733 $17,132 $17,530 $17,929 $18,327 $18,725 AMD Server CPUs 2012 2013 2014 2015 2016 2017 1H2018 $660 $15,492 $15,900 $16,307 $16,715 $17,123 $17,530 $17,938 $18,346 $18,753 $19,161 - Shipments 920 790 400 172 82 163 178 $675 $15,844 $16,261 $16,678 $17,095 $17,512 $17,929 $18,346 $18,763 $19,179 $19,596 AMD Shipment Share 5.5% 4.6% 2.1% 0.8% 0.4% 0.7% 1.4% $690 $16,196 $16,622 $17,048 $17,475 $17,901 $18,327 $18,753 $19,179 $19,606 $20,032 - Revenue $324 $264 $129 $47 $17 $41 $95 $705 $16,548 $16,984 $17,419 $17,855 $18,290 $18,725 $19,161 $19,596 $20,032 $20,467 - Blended ASP ($/unit) $352 $334 $321 $271 $205 $250 $536 $720 $16,900 $17,345 $17,790 $18,234 $18,679 $19,124 $19,569 $20,013 $20,458 $20,903
Intel Server CPU ASP ($/unit)Intel ASP Server CPU $735 $17,252 $17,706 $18,160 $18,614 $19,068 $19,522 $19,976 $20,430 $20,884 $21,338 Source: Mercury Research; Wells Fargo Securities, LLC Estimates $750 $17,604 $18,068 $18,531 $18,994 $19,457 $19,921 $20,384 $20,847 $21,311 $21,774
AMD Server CPU Scenario Analysis
2018 Estimates 2019 Estimates 2020 Estimates 2017 Current Est. Current Est. Server-Driven Est. Range Current Est. Server-Driven Est. Range Actual Wells Street Wells Est. Street Low Mid High Wells Est. Street Low Mid High Client CPUs (PCs; Mercury) $1,466 $1,937 $2,180 $2,084
Implied Trad'l Graphics $1,212 $2,357 n/a $2,285 n/a No Changes $2,568 n/a No Changes Approx. Crypto (AMD Disclosed %) $299 $310 $85 $50 Total Graphics (Implied; Mercury) $1,511 $2,667 $2,370 $2,618 Total Computing & Graphics $2,977 $4,604 $4,409 $4,550 $4,662 $4,550 $4,550 $4,550 $4,701 $5,249 $4,701 $4,701 $4,701
Server CPUs $41 $427 $1,202 $1,179 $1,672 $2,680 $1,952 $2,255 $2,801 $3,739 n/a n/a n/a Semi-Custom + Embedded $2,235 $2,003 $1,548 $1,803 $2,003 $2,204 $1,076 $2,003 $2,204 $2,404 Enterprise, Embedded, & Semi-Custom $2,276 $2,430 $2,326 $2,750 $2,653 $2,982 $3,676 $4,884 $3,029 $2,979 $4,258 $5,005 $6,143 TOTAL REVENUE $5,253 $7,035 $6,735 $7,300 $7,315 $7,532 $8,226 $9,434 $7,730 $8,112 $8,959 $9,706 $10,844 - Revenue Upside vs. Current Street Estimate 3.0% 12.5% 29.0% 10.4% 19.6% 33.7% Gross Margin 34.1% 37.5% 37.3% 39.7% 39.3% 40.0% 40.5% 41.5% 41.3% 41.5% 42.0% 43.0% 44.0% Operating Expenses % of Revenue 29.8% 27.3% 27.5% 27.0% 27.4% 27.0% 26.0% 25.0% 26.6% 26.3% 26.5% 25.5% 24.5% Operating Income $224 $688 $662 $923 $874 $979 $1,193 $1,557 $1,133 $1,233 $1,389 $1,699 $2,115 - EBIT Upside vs. Current Street Estimate 12.1% 36.5% 78.2% 12.7% 37.8% 71.5% Other Income (Expense) ($104) ($92) ($143) ($80) ($124) ($80) ($76) ($205)8.0% ($76) Tax Rate 14.2% 6.3% 6.8% 8.0% 8.3% 8.0% 8.0% 8.4% Non-GAAP EPS $0.09 $0.50 $0.46 $0.67 $0.64 $0.72 $0.89 $1.18 $0.84 $0.88 $1.05 $1.29 $1.63 - Non-GAAP EPS Upside vs. Current Street Estimate 12.1% 38.8% 84.1% 63.3% 46.5% 84.1% - Fully Diluted Shares 1,039.0 1,120.3 1,122.7 1,151.0 1,146.0 1,151.0 1,154.0 1,137.1 1,154.0 Server CPU Estimates: Total Server CPUs Shipped (Mercury Est.) 27,317 31,440 35,735 35,735 39,563 39,563 - AMD Server CPUs Shipped 163 722 2,074 2,144 2,859 4,288 3,157 3,956 4,748 5,934 - AMD Server CPU Ship Share 0.6% 2.3% 5.8% 6% 8% 12% 8.0% 10% 12% 15% - AMD Server CPU ASP ($/unit) $250 $591 $579 $550 $585 $625 $618 $570 $590 $630 Semi-Custom + Embedded Y/Y % Chg. vs. 2018 Est. -10% 0% 10% 0% 10% 20% Source: Company Data; Mercury Research; FactSet; Wells Fargo Securities, LLC Estimates
10 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
Intel Positioning in Artificial Intelligence / High-Performance Computing (HPC) Will Also Remain a Key Focus Intel’s positioning for artificial intelligence / high-performance computing will remain a key competitive focus. The growth / importance of GPUs versus CPUs will likely leave investors to question a shifting dollar spend away from Intel. Intel’s efforts in AI include an anticipated re-engagement in the GPU market (2020+ story), the company’s Nervana Systems Neural Network Processors and the recent acquisition of Vertex.ai, Movidius Myriad VPUs, data center adoption of Stratix FPGAs, and more. At the early August Data-Centric Summit, Intel disclosed that its sales of chips into AI opportunities surpassed $1B/annum. IDC currently estimates that spending on Accelerated Servers will grow from $6.47 billion in 2017 to $10.3 billion in 2018 and $25.6 billion by 2022, representing a +32% 2017-2022 CAGR. This compares to the total server market estimated to grow from $70.6 billion in 2017 to $89.0 billion by 2022, or rather implying non-accelerated server spending to remain flat over this timeframe. AI-focused server spending is estimated to grow from $4.6 billion in 2017 to $6.7 billion in 2018 and $17.6 billion by 2022, a +31% CAGR. Within this, cloud-based AI server spending is estimated to grow from $1.08 billion in 2017 to $9.08 billion by 2022, a +53% CAGR over this timeframe. The expanding role of GPUs in HPC / supercomputing is becoming increasingly visible. For example, the recently deployed IBM-based 200 petaflops Summit supercomputer at Oak Ridge National Laboratory (ORNL) that incorporates 4,608 compute servers, each with two 22-core IBM Power9 processors and six NVIDIA Tesla V100 GPU accelerators – or equating to 27,648 Volta Tensor Core GPUs that account for 92% of the systems performance. The charts below illustrate the expanding growth / role of GPUs in high-performance computing. As we consider the aforementioned growth dynamics in accelerated / AI-focused servers, investors will likely be continually focused on Intel’s positioning relative to NVIDIA – i.e., spending shifting focus from solely driving performance via increasing core density to the need to leverage GPU-based acceleration. We think investors will continue to look at the expansion of GPUs in the HPC and cloud markets as a key competitive advantage. The June 2018 Top500 list showed NVIDIA accelerator share at 87% vs. 78% a year ago and that the number of NVIDIA GPU cores deployed across the top 500 supercomputers increased to 5.68MM, up +400% y/y. NVIDIA accelerators are used in 110 of the top 500 systems (vs. 91 a year ago) which compares to 7 systems for Intel who discontinued its Xeon Phi product line in 2H2017. We believe that NVIDIA is well positioned to capitalize on Intel’s 10nm transition issues via its dominate position in the GPU market - focus on NVIDIA’s anticipated move to 7nm via TSMC in 2019, while Intel has plans to launch its first discrete GPU in 2020.
Wells Fargo Securities, LLC | 11 Semiconductors & Semiconductor Cap Equipment Equity Research
Intel’s Focus on Expanding its AI-Focused Portfolio. Intel has highlighted a broadening portfolio focused on Artificial Intelligence, which include endpoint solutions such as Movidius and Mobileye, a focus on edge offerings with Atom and the Altera Stratix FPGAs. The company is focused on AI in the data center with Xeon CPUs and Nervana Neural Network Processors (NNP). Intels Below we provide a quick summary of Intel’s efforts to expand its AI portfolio – focus on building on the 90%+ share of Xeon CPUs have in AI Inferencing; participating in expanding edge-to-data center use cases utilizing customized and programmable silicon.
Nervana Acquired: ~$400 million; closed in 3Q16 – first commercial shipments expected in 2019.
Product: Neural Network Processors – Nervana’s NNP L-1000 (codenamed: Lake Crest) ASICs.
Optimized across memory and bandwidth to provide 3-4x training performance over first generation NNP processors. Reports highlight on- and off-chip high-speed mesh and integration of HBM (high-bandwidth memory). Pre-acquisition reports noted that Nervana was planning to roll- out a 28nm chip in 2017.
Memory interface development focused on competing against NVIDIA’s NVLink; Next Platform reported in early-2018 that the Nervana NNP would provide 5-6x the performance of NVIDIA’s Pascal GPUs.
Movidius Acquired: Reports suggest ~$400 million (no Intel disclosure); close in 2H2016
Product: Myriad X and Myraid 2 Vision Processing Units (VPUs; SoCs); computer vision technology that was behind Google’s Project Tango 3D-sensor technology; focus on integration with Intel’s RealSense technology.
Intel’s focus is to place the Movidius VPUs in low-powered edge devices (robotics, drones, video cameras, PCs, etc.) for vision processing and inferencing.
Soft Acquired: Reports suggest ~$250 million (no Intel disclosure).
Machines Product: VISC technology; industry reports suggested that Soft Machines new multi-core processors enabled parallelized performance (multiple CPU cores working together) to support as much as 4x the performance per watt when compared to traditional single-threaded CPUs. Reports in February 2016 noted that the company was working on plans to tape out an SoC codenamed Mojave at 16nm FinFET; 64-bit ARMv8 processor core codenamed Shasta.
eASIC Acquired: Announced in mid-July 2018; clos in 3Q18
Product: Structured ASICs; focus on complementing Intel’s broadening portfolio to address the evolution / adoption of purpose-built ASICs, Vision Processing Units (VPUs), and FPGAs.
Intel’s press release notes that it is focused on taking advantage of the company’s Embedded Multi-Die Interconnect Bridge (EMIB) – combining Intel’s FPBAs (Altera Stratix) and structured ASICs (eASICs) in a system-in-package solution.
Vertex.ai Acquired: Intel announced acquisition of Vertex.ai in August 2018
Product: Focus on providing tools to run algorithms more efficiently; focus on integration with Intel’s nGraph tool.
Source: Company Data; Wells Fargo Securities, LLC
12 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
Intel Capital Intensity versus Depreciation Expense – Wells Model Implying Potential Significant Depreciation Expense Increase in 2019-Beyond We think the capital intensity of Intel’s progression to 10nm, along with continued capacity investments to meet demand for 14nm (note: there have been recent reports of tight 14nn client CPU supply looking into the 2018 holiday season), and the yet-to-be discussed move to 7nm and beyond through the use of Extreme Ultra Violet (EUV) systems, will become a key investor focus as we look into 2019. As shown in the first chart below, Intel has guided its 2018 capital expenditures at approximately $15 billion. As we look forward to the implications of increasing capital intensity on Intel’s model, we highlight the following thoughts / considerations:
1. Depreciation vs. Capital Expenditures: Well’s Waterfall Model Implies Notable Acceleration in Depreciation Expense in 2019-Beyond (Note: Historical R2 = 0.99+ vs. Reported Depreciation). Put simplistically, we think investors are increasingly focused on the difference between Intel’s capex spending and reported depreciation expense, or rather how we should be thinking about the margin implications looking into 2019 and beyond.
In an effort to try to model the depreciation expense implications of Intel’s capex we created an annual depreciation waterfall analysis using Intel’s Property, Plant, & Equipment (PP&E) breakdown since 1970. We arrive at a model that: (1) Exhibits a high historical correlation (R2 = 0.99+) with Intel’s annual depreciation, and (2) leaves us to focus on implied significant increase in depreciation expense to be in 2019 and beyond.
2. Intel’s Significant Increase in Construction in Progress (CIP) Balance = Yet to be Depreciated Capex. We would highlight Intel’s significant increase in the Construction in Progress balance within PP&E as this balance reflects capex investments from which depreciation is not yet recognized. As shown in the chart on page 14, Intel’s CIP balance grew to $15.812 billion, a notable increase from the $9.4 billion and $10.9 billion balance exiting 2015 and 2016, respectively (note: Intel does not provide this disclosure on a quarterly basis).
While it is unclear as to how much of the CIP balance reflects machinery & equipment (most notably EUV systems) versus land / buildings, we would note that this balance includes the company’s Fab 42 in Chandler, AZ, which reflects a $7 billion investment that was initially targeted at 14nm production, but subsequently deferred for 7nm / 450mm production (note: February 2017 announcement – press release). In late-August it was reported that Fab 42 could be opened in 2020-2021 timeframe. Industry reports have noted that Intel’s 10nm production would be done out of Fab 28 located in Kiryat Gat, Israel. In early-2018 the Israel Ministry of Finance reported that Intel would spend $5 billion on Fab 28 in preparation of 10nm production (from 22nm production).
Wells Fargo Securities, LLC | 13 Semiconductors & Semiconductor Cap Equipment Equity Research
14 | Wells Fargo Securities, LLC
Intel Corporation Equity Research
While not necessarily a reflection of capital intensity, we think investors will continue to consider the increasing costs associated with Intel’s eventual move to 7nm and 5nm. As shown below, IBS research estimates that 5nm advanced chip design costs would be over 1.8x more expensive than 7nm, or more than 3.1x more capital intensive than 10nm. Despite the inherent cost complexities of next-generation leading-edge processes, IBS has also estimated a rapid growth of 7nm in foundries – growing from no revenue in 2017 to nearly $5 billion in 2017 and $49.8 billion estimated in 2019. This compares to 16nm/14nm/12nm foundry revenue estimated to decline to $9.7 billion in 2018, or -11% y/y.
For Intel, we think investors will be focused on the company’s deployment of EUV as a key consideration in the company’s anticipated move beyond 10nm, as well as the model implications of increasing capital intensity. During the company’s 2Q18 earnings call, management noted that the “…amount of reuse of equipment and tools from one node to the next is extremely high…with the exception of 10nm going to 7nm, litho changing quite a bit with EUV…”.
Wells Fargo Securities, LLC | 15 Semiconductors & Semiconductor Cap EquipmentSemiconductors & Semiconductor Cap Equipment Equity Research
Required Disclosures
This is a compendium report, to view current important disclosures and other certain content related to the securities recommended in this publication, please go to https://www.wellsfargoresearch.com/Disclosures or send an email to: [email protected] or a written request to Wells Fargo Securities Research Publications, 7 St. Paul Street, Baltimore, MD 21202.
Additional Information Available Upon Request
I certify that: 1) All views expressed in this research report accurately reflect my personal views about any and all of the subject securities or issuers discussed; and 2) No part of my compensation was, is, or will be, directly or indirectly, related to the specific recommendations or views expressed by me in this research report.
Wells Fargo Securities, LLC does not compensate its research analysts based on specific investment banking transactions. Wells Fargo Securities, LLC’s research analysts receive compensation that is based upon and impacted by the overall profitability and revenue of the firm, which includes, but is not limited to investment banking revenue.
STOCK RATING 1=Outperform: The stock appears attractively valued, and we believe the stock's total return will exceed that of the market over the next 12 months. BUY 2=Market Perform: The stock appears appropriately valued, and we believe the stock's total return will be in line with the market over the next 12 months. HOLD 3=Underperform: The stock appears overvalued, and we believe the stock's total return will be below the market over the next 12 months. SELL SECTOR RATING O=Overweight: Industry expected to outperform the relevant broad market benchmark over the next 12 months. M=Market Weight: Industry expected to perform in-line with the relevant broad market benchmark over the next 12 months. U=Underweight: Industry expected to underperform the relevant broad market benchmark over the next 12 months.
VOLATILITY RATING V=A stock is defined as volatile if the stock price has fluctuated by +/-20% or greater in at least 8 of the past 24 months or if the analyst expects significant volatility. All IPO stocks are automatically rated volatile within the first 24 months of trading.
As of: September 9, 2018
48% of companies covered by Wells Fargo Securities, LLC Equity Wells Fargo Securities, LLC has provided investment banking Research are rated Outperform. services for 41% of its Equity Research Outperform-rated companies.
50% of companies covered by Wells Fargo Securities, LLC Equity Wells Fargo Securities, LLC has provided investment banking Research are rated Market Perform. services for 29% of its Equity Research Market Perform-rated companies.
1% of companies covered by Wells Fargo Securities, LLC Equity Wells Fargo Securities, LLC has provided investment banking Research are rated Underperform. services for 21% of its Equity Research Underperform-rated companies.
16 | Wells Fargo Securities, LLC Intel 10nm Ice Lake vs. AMD 7nm Rome: Server CPU Share Shift Ahead Equity Research
Important Disclosure for U.S. Clients This report was prepared by Wells Fargo Securities Global Research Department (“WFS Research”) personnel associated with Wells Fargo Securities and Structured Asset Investors, LLC (“SAI”), a subsidiary of Wells Fargo & Co. and an investment adviser registered with the SEC. If research payments are made separately from commission payments, this report is being provided by SAI. For all other recipients in the U.S. this report is being provided by Wells Fargo Securities.
Important Disclosure for International Clients
EEA – The securities and related financial instruments described herein may not be eligible for sale in all jurisdictions or to certain categories of investors. For recipients in the EEA, this report is distributed by Wells Fargo Securities International Limited (“WFSIL”). WFSIL is a U.K. incorporated investment firm authorized and regulated by the Financial Conduct Authority. For the purposes of Section 21 of the UK Financial Services and Markets Act 2000 (“the Act”), the content of this report has been approved by WFSIL, an authorized person under the Act. WFSIL does not deal with retail clients as defined in the Directive 2014/65/EU (“MiFID2”). The FCA rules made under the Financial Services and Markets Act 2000 for the protection of retail clients will therefore not apply, nor will the Financial Services Compensation Scheme be available. This report is not intended for, and should not be relied upon by, retail clients.
Australia – Wells Fargo Securities, LLC is exempt from the requirements to hold an Australian financial services license in respect of the financial services it provides to wholesale clients in Australia. Wells Fargo Securities, LLC is regulated under U.S. laws which differ from Australian laws. Any offer or documentation provided to Australian recipients by Wells Fargo Securities, LLC in the course of providing the financial services will be prepared in accordance with the laws of the United States and not Australian laws.
Canada – This report is distributed in Canada by Wells Fargo Securities Canada, Ltd., a registered investment dealer in Canada and member of the Investment Industry Regulatory Organization of Canada (IIROC) and Canadian Investor Protection Fund (CIPF). Wells Fargo Securities, LLC’s research analysts may participate in company events such as site visits but are generally prohibited from accepting payment or reimbursement by the subject companies for associated expenses unless pre-authorized by members of Research Management.
Hong Kong – This report is issued and distributed in Hong Kong by Wells Fargo Securities Asia Limited (“WFSAL”), a Hong Kong incorporated investment firm licensed and regulated by the Securities and Futures Commission of Hong Kong (“SFC”) to carry on types 1, 4, 6 and 9 regulated activities (as defined in the Securities and Futures Ordinance (Cap. 571 of The Laws of Hong Kong), “the SFO”). This report is not intended for, and should not be relied on by, any person other than professional investors (as defined in the SFO). Any securities and related financial instruments described herein are not intended for sale, nor will be sold, to any person other than professional investors (as defined in the SFO). The author or authors of this report may or may not be licensed by the SFC. Professional investors who receive this report should direct any queries regarding its contents to Mark Jones at WFSAL (email: [email protected] ).
Japan – This report is distributed in Japan by Wells Fargo Securities (Japan) Co., Ltd, registered with the Kanto Local Finance Bureau to conduct broking and dealing of type 1 and type 2 financial instruments and agency or intermediary service for entry into investment advisory or discretionary investment contracts. This report is intended for distribution only to professional investors (Tokutei Toushika) and is not intended for, and should not be relied upon by, ordinary customers (Ippan Toushika).
The ratings stated on the document are not provided by rating agencies registered with the Financial Services Agency of Japan (JFSA) but by group companies of JFSA-registered rating agencies. These group companies may include Moody’s Investors Services Inc., Standard & Poor’s Rating Services and/or Fitch Ratings. Any decisions to invest in securities or transactions should be made after reviewing policies and methodologies used for assigning credit ratings and assumptions, significance and limitations of the credit ratings stated on the respective rating agencies’ websites.
Wells Fargo Securities, LLC | 17 Semiconductors & Semiconductor Cap EquipmentSemiconductors & Semiconductor Cap Equipment Equity Research
About Wells Fargo Securities Wells Fargo Securities is the trade name for the capital markets and investment banking services of Wells Fargo & Company and its subsidiaries, including but not limited to Wells Fargo Securities, LLC, a U.S. broker-dealer registered with the U.S. Securities and Exchange Commission and a member of NYSE, FINRA, NFA and SIPC, Wells Fargo Prime Services, LLC, a member of FINRA, NFA and SIPC, Wells Fargo Securities Canada, Ltd., a member of IIROC and CIPF, Wells Fargo Bank, N.A. and Wells Fargo Securities International Limited, authorized and regulated by the Financial Conduct Authority.
This report is for your information only and is not an offer to sell, or a solicitation of an offer to buy, the securities or instruments named or described in this report. Interested parties are advised to contact the entity with which they deal, or the entity that provided this report to them, if they desire further information or they wish to effect transactions in the securities discussed in this report. The information in this report has been obtained or derived from sources believed by Wells Fargo Securities Global Research Department (“WFS Research”), to be reliable, but WFS Research does not represent that this information is accurate or complete. Any opinions or estimates contained in this report represent the judgment of WFS Research, at this time, and are subject to change without notice. Certain text, images, graphics, screenshots and audio or video clips included in this report are protected by copyright law and owned by third parties (collectively, “Third Party Content”). Third Party Content is made available to clients by Wells Fargo under license or otherwise in accordance with applicable law. Any use or publication of Third Party Content included in this report for purposes other than fair use requires permission from the copyright owner. Any external website links included in this publication are not maintained, controlled or operated by Wells Fargo Securities. Wells Fargo Securities does not provide the products and services on these websites and the views expressed on these websites do not necessarily represent those of Wells Fargo Securities. Please review the applicable privacy and security policies and terms and conditions for the website you are visiting. All Wells Fargo Securities and SAI research reports published by WFS Research are disseminated and available to all clients simultaneously through electronic publication to our internal client websites. Additional distribution may be done by sales personnel via email, fax or regular mail. Clients may also receive our research via third party vendors. Not all research content is redistributed to our clients or available to third-party aggregators, nor is WFS Research responsible for the redistribution of our research by third party aggregators. Equity Strategists focus on investment themes across the equity markets and sectors. Any discussion within an Equity Strategy report of specific securities is not intended to provide a fundamental analysis of any individual company described therein. The information provided in Equity Strategy reports is subject to change without notice, and investors should not expect continuing information or additional reports relating to any security described therein. For research or other data available on a particular security, please contact your sales representative or go to http://www.wellsfargoresearch.com. For the purposes of the U.K. Financial Conduct Authority's rules, this report constitutes impartial investment research. Each of Wells Fargo Securities, LLC and Wells Fargo Securities International Limited is a separate legal entity and distinct from affiliated banks. Copyright © 2018 Wells Fargo Securities, LLC
SECURITIES: NOT FDIC-INSURED/NOT BANK-GUARANTEED/MAY LOSE VALUE
18 | Wells Fargo Securities, LLC