“Life isn’t about waiting for the storm to pass… It’s about learning to dance in the rain”
The semiconductor space has gone from ultra-bullish to ultra-bearish in a mere matter of months. While semiconductor demand has become far more diversified in recent years, supply remains subject to the over/ under. With the downcycle in full swing, it’s worth considering ways to weather the cycle, however long it lasts.
I think playing defensive on semiconductor names via Taiwan Semiconductor (TSM) makes the most sense. As the pre-eminent foundry based in Taiwan, TSMC benefits from several tailwinds – 1) It is now the dominant player at leading edge nodes (<7nm), 2) It stands to benefit from a US-China trade war, and 3) Via back-end packaging, continuous logic scaling, and co-design, it can outpace Moore’s Law, perhaps even by a factor of >2.
On Track For a 3Q Beat
TSMC’s August sales came in at NT$91,055mln (+22%MoM; -1%YoY), in line with prior years’ seasonal trends. So far, TSMC has posted ~NT$165bn for 3Q, which makes up 64% of the TSMC’s 3Q18 guidance at the lower bound. To hit the 3Q revised guidance of +9-10% QoQ (ex-virus impact), TSMC has a low bar – a flat to mild MoM decline in September would allow them to meet the lower end. This is especially impressive considering a virus outbreak last month cost them <3%pts of QoQ growth due to shipment delays (shifted from Sept to Dec).
3Q strength is mainly attributable to the 7nm ramp – Apple (NASDAQ:AAPL) has elected to use the 7nm A12 processor for its TFT iPhone 9 in 2H over 10nm given better performance and pricing. As component prices decline, Apple has been able to upgrade to the 7nm processor even at the 64G price point ($649 vs. $849 for 256G). 7nm prospects in 4Q should also be supported by Qualcomm’s (QCOM) 7nm S855 and 5G X50 baseband, Bitmain’s 7nm crypto ASIC, AMD’s (AMD) 7nm CPU, and HiSilicon’s 7nm S980 in 4Q18.
7nm is widely expected to expand meaningfully from ~10% in 2018 to ~30% of revenue in 2019. Because TSMC is virtually uncontested at 7nm, this revenue source will provide a significant barrier to competition from the trailing edge while cushioning the effects of a downturn. GMs at 7nm are ~0.5-1% higher as well, but with its 7nm dominance, I wouldn’t be surprised to see margin expansion going into 2019. For now, high legacy node conversion and cost out efforts should provide an additional margin tailwind.
Last (Pure-Play) Foundry Standing
The big news last month was that GlobalFoundries (GF), AMD’s foundry, had capitulated on its 7nm development. The announcement surprised some, although it should really have been widely expected. GF’s decision was many months in the making – tier two foundries like UMC (UMC) and SMIC had given up on 7nm long before, opting to focus on trailing edge nodes, particularly at 28nm.
Fab economics have also become prohibitive to smaller players due to the sheer size of capex required – TSMC’s 450mm fab, which cost over $10bn, is nothing compared to its latest 3nm fab which will require ~$20bn in capex. The fact that GF capitulated despite government backing is testament to the implicit Darwinism at play in the foundry business.
The revenue impact could be a lot more significant than what the market is pricing in. AMD currently contributes ~2-3% of TSMC’s sales. Post-GF announcement, TSMC will become the exclusive foundry for AMD at 7nm, including its GPU and CPU. Considering AMD allocates a third of its foundry order to TSMC, this should boost TSMC’s revenue line by an incremental ~4-6% into 2019 and 2020. If we factor in the extra value-added services TSMC also provides e.g. packaging, this could easily move up to a sub-10% contribution. And that estimate doesn’t include the potential order gains from GF’s other customers e.g. IBM (IBM) or share gains from Intel (NASDAQ:INTC) (struggling with its 10nm ramp), both of which should drive additional revenue contribution well over the 10% mark.
On paper, this leaves three players in the leading edge (<7nm) race – TSMC, Samsung (OTC:SSNLF) and Intel. Of the three, TSMC is the market leader and now holds a virtual monopoly at 7nm while the rest catch up. Samsung comes in second, with Intel the clear laggard. In fact, Intel has begun outsourcing to TSMC e.g. its 5G modem is slated to shift to TSMC’s 7nm by 2020.
Notably, TSMC is the only pure-play foundry left standing. As Samsung and Intel have other own-brand businesses, its customer base is limited by inherent conflicts of interest and IP concerns. As TSMC pioneered the foundry business and has been around far longer, its customer relationships are also stronger. Both Intel and Samsung have typically been used as a “second-source” by customers who wish to diversify away concentration risk at TSMC.
TAM Expansion at 7nm
It wasn’t too long ago when high-performance computing (HPC) was an afterthought at TSMC as Intel’s x86 dominated PC/server computing. The main contributor to the HPC segment was GPUs (ex-PC) from Nvidia (NVDA) and AMD as well as infrastructure networking.
As we transition to 7nm, TSMC’s HPC TAM (semiconductor used in data centres, servers, networking, storage and gaming) has expanded meaningfully and looks set to become a key growth driver going forward. Via HPC, TSMC is exposed to all the hyped growth drivers of tomorrow (AI, IoT) and then some. Tapeout numbers support the notion that 7nm will be much larger than prior nodes – 7nm has 50+ vs 16nm’s 50 in the first year of production.
The emergence of crypto demand is a prime example of TAM expansion. From virtually zero in 2016, crypto revenue contribution now runs anywhere from MSD to HSD. At 7nm, mining giants such as Bitmain and Canaan Creative (~85-90% share of mining hardware) require leading edge nodes to stay ahead of the mining race. With the ETH mining now also set for an ASIC shift, crypto revenue is set to grow at the foundry level from increased content per box. As GPUs require a larger die than ASICs, each GPU can only contain one while ASICs can have hundreds. This, combined with higher manufacturing yields (ASICs have lower complexity and defect rates), should result in higher revenues per miner post-ASIC shift.
(Source: Bitmain; Note: Third row denotes number of hash chips per ASIC)
These drivers are only going to go upward as new use cases develop along with parallel attachments. The latter is a key driver, yet rarely mentioned. For instance, hardware accelerators are increasingly being used at data centres with additional attach rates set to be driven by AI applications. Estimates for this market alone range from $8bn to $10bn by 2020 (from $2bn to $3bn in 2018). As “everyone’s foundry”, TSMC is a prime beneficiary.
Trade War Beneficiary
The semiconductor sector has seen largely one-way impact from the US-China trade war as aside from added demand uncertainty, the US doesn’t really depend on China for its semis.
The supply side impact, however, is far more interesting for two reasons. Firstly, Chinese semi players are now enjoying shorter qualification times as China seeks to accelerate domestic replacement. This should provide some offset to any trade war-related demand uncertainty at the foundry level. Secondly, Chinese foundries with US customers e.g. SMIC are disproportionately at risk of seeing order shifts to other geographies. SMIC’s 30-40% US revenue contribution for instance is at risk of shifting to its non-China peers.
TSMC stands to benefit from this shift, along with its Taiwanese peers.
2x More Than Moore?
Perhaps the most exciting new disclosure out of TSMC is its ability to not only defy but also outpace Moore’s Law by a factor of two. Per TSMC, it expects to scale on 3 axes – via 3D, co-design and continuous logic scaling.
TSMC’s competitive advantage here is its back-end capabilities, particularly in packaging. Its fan-out packaging (Integrated Fan Out/ “InFO”), for instance, is industry-leading. By moving back-end IC packaging to the wafer-level, InFO enables lower costs (via the elimination of substrate) and higher yields (via faster speed and improved thermal performance). Its success in back-end packaging extends to non-wafer services such as silicon IP development, design services, and photomask production.
Its 2.5D/3D packaging technology (Chip on Wafer Substrate/”CoWoS) has also been adopted by high-end chips such as GPUs and FPGAs. CoWoS connects ICs side-by-side via a silicon interposer and delivers higher performance albeit at a higher cost.
Along with advances in packaging, wide bandgap materials such as Silicon Carbide (SiC) and Gallium Nitride (GaN) are allowing foundries to overcome the physical limitations of silicon and boost transistor density.
Per TSMC CEO Mark Liu – “In the past, die area, performance, and power were the key metrics for the semiconductor industry…what we need now is to reduce energy consumption.” With ~10x better conduction and switching properties, wide bandgap materials will enable compact, high-power density, cost-effective, energy-efficient power components operating at higher temperature, voltage and frequency conditions.
Semiconductor Name with Counter-Cyclical Properties
Due to the diversity of its customer base (>400 customers worldwide) as well as its diversified mix across the semi universe, TSMC weathers the semiconductor boom-bust cycles quite well. In fact, its only top-line contraction over the last decade came during the GFC period.
While communication remains its core segment (Apple contributes ~20+% to revenue), its exposure to handsets remains relatively diversified as well, with contributions from different geographies (including Chinese smartphone makers) and parts (processors, modems etc.). High performance computing has grown from ~8% to ~10% in 2017, but with 7nm options limited to only TSMC and Samsung, most HPC players are likely to be allocating more to TSMC, which should further diversify the mix.
An additional source of TSMC’s defensive quality most miss is its “copy exactly” policy where fab operations using the same technology are required to be identical e.g. configuration, equipment, process recipe and production flows. This allows TSMC to recover quickly from temporary disruptions such as this year’s earthquake – back up from other fabs can seamlessly take over during down time. This policy also allows TSMC to flex capacity during peak/troughs. From a cost perspective, this also enables lower procurement costs as well as smooth knowledge transfer.
With key tailwinds such as a shift away from Chinese foundries, increasing 7nm share, expanding foundry TAM, and a short-term boost from the upcoming iPhone ramp, TSMC could well serve as a shelter should fears of a semiconductor down-cycle be realized. TSMC may look slightly pricey at these levels, but I don’t think the magnitude of recent developments has been fully appreciated by the market. Assuming TSMC achieves >$3 EPS by 2020 (implying ~15x forward PE), valuations really aren’t demanding. I like the stock here and am accumulating.
Disclosure: I am/we are long TSM.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Editor’s Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.
Read more from source here…