PDF 原檔：260601_2454_2379_5347_6415_6515_6223_5274_gs__original.pdf
原始內容
Taiwan Technology: Semiconductors: Day 1 Key takeaways: GS Taiwan Computex & Corporate Day 2026
We are hosting our Goldman Sachs Taiwan Computex & Corporate Day during 1-2 June 2026. During Day 1, within our Taiwan Semiconductor coverage space, we hosted management teams from MediaTek (2454.TW; Buy), Realtek (2379.TW; Neutral), Vanguard (5347.TWO; Sell), Silergy (6415.TW; Buy), WinWay (6515.TW; Buy), MPI (6223.TWO; Buy) and Aspeed (5274.TWO; Buy). We provide our key takeaways from the sessions below.
MediaTek (2454.TW; Buy)
On enterprise ASIC, management reiterated its previously raised 2027 ASIC TAM of US$70-80bn with market share between 10-15%, but they emphasized that even after the upward revision the guidance still sits below customer demand. The key constraints continue to be advanced packaging capacity, substrate, and HBM, though management now appears to have more con fi dence in advanced packaging being less of a bottleneck, noted that HBM is handled by the customer, and highlighted that substrate now has more than one quali fi ed vendor. From a competitive standpoint, MediaTek sees itself as competing on strong performance per total cost of ownership vs. its competitors.
Looking ahead to the next generation, management sees possibilities for GM to be raised by a couple of ppts, as chip design complexity increases signi fi cantly. The next-generation program is adopting only EMIB-T, with tape-out target in 4Q26 and mass production by 4Q27. Management also noted that its foundry partner is willing to support customers on very large packages despite the back-end packaging may not be done by the same vendor, which we believe could help ease investor concerns. With I/O dies, compute dies, and die-to-die interconnect all spec'd up, MediaTek believes the next generation ASIC ASP will at least double vs. the current generation.
For businesses outside of ASIC, in auto, management targets US$1bn of revenue contribution in 2027, ramping with Chinese automakers this year with target to expand to global players, and expects 2027 to have ADAS revenue contribution via partnership with a Japanese vendor. In smartphone, management continues to enhance AI capabilities against a backdrop of an estimated 15% global unit decline in 2026, with its own mainland China customers seeing a steeper decline of around 20%. Finally, management sees AI glasses as a long-term driver, collaborating with Meta, with revenue contribution to begin in 2027 and growing signi fi cantly thereafter.
Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the fi rm may have a con fl ict of interest that could a ff ect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. For Reg AC certi fi cation and other important disclosures, see the Disclosure Appendix, or go to ed as research
Bruce Lu
+886(2)2730-4185 | bruce.lu@gs.com Goldman Sachs (Asia) L.L.C., Taipei Branch
Evelyn Yu
+886(2)2730-4187 | evelyn.yu@gs.com Goldman Sachs (Asia) L.L.C., Taipei Branch
Ryan Huang, CFA |
+886(2)2730-4084 ryan.huang@gs.com Goldman Sachs (Asia) L.L.C., Taipei Branch e92c7a75ab8b4efbba794e6b187208c8
Realtek (2379.TW; Neutral)
Management indicated demand is still outpacing supply and is running ahead of its earlier forecast, with both 1Q26 and 2Q26 coming in above expectations. It characterized this as a happy problem that is increasingly becoming a challenge, given the di ffi culty in securing additional foundry and OSAT capacity. Visibility into 2H26 remains limited: customers have not yet revised their 2H26 orders, despite describing the 1H26 demand as a pull-in, and customer inventory levels do not appear elevated based on management's assessment. The full-year outlook now tracks better than at the start of the year, with management expecting clearer visibility into 2H26 by the end of 2Q26.
From a longer term perspective, management highlighted its Ethernet business as a key driver. Beyond automotive Ethernet solutions, management is targeting the edge-server market via its 100G and above optical module solution, which will be fi rst showcased at Computex this year. Management emphasizes that it is speci fi cally targeting the edge-server market rather than cloud, and believes that the edge-server market will grow rapidly in coming years as customers prefer on-premise datacenters for cybersecurity reasons. Realtek's 100G SerDes engineering sample has successfully taped out, with potentially higher revenue contribution expected in 2027 and beyond.
Vanguard (5347.TWO; Sell)
For 2026, against a mature node TAM growth of around 15% YoY, management expects Vanguard to outpace the overall market at around 20% YoY, driven by strong PMIC demand from AI and geopolitical wafer reallocation out of mainland China. AI revenue exposure spans both fabless and IDM PMIC clients across US and Japan, and management sees AI demand to increase signi fi cantly thanks to rising power requirement. On the mega-trend, management sees steady mature-node demand growth against steadily declining supply, with leading foundries shrinking mature-node capacity, creating a more favorable pricing environment.
Capex is guided to be NT$60-70bn in 2026, with over 90% on 12-inch expansion and around 10% of 8-inch maintenance/de-bottleneck. On 12-inch, management expects capacity to ramp to 20kwpm+ by end-2027, around 30kwpm in early 2028, and 44kwpm by end-2028, with maximum P1 capacity trimmed to 44kwpm from 55kwpm, as interposer manufacturing is now included, which requires more complex processes. Vanguard's foundry partner will consign the interposer manufacturing equipment directly to Vanguard's 12-inch fab, which is expected to lower Vanguard's overall 12-inch investment. Overall, management still targets a long-term GM target of 35%, while the 12-inch fab ramp up is expected to dilute GM by mid-to-high single digits in 2027, mid-single-digits in 2028, and low-single-digits in 2029.
Silergy (6415.TW; Buy)
Management commented that all fi ve product segments beat expectations in 1Q26, with the industrial business showing particular strength due to datacenter-related demand. Management stated that datacenter accounts for around 15% of 1Q26 revenue and grew about 90% YoY, spanning eSSD, optical modules, and server-motherboard components such as PSU; server motherboard was a low-single-digit share of revenue in 2025 and is expected to reach mid-single digits in 2026, with eSSD and optical modules accounting for a much larger revenue share. For e92c7a75ab8b4efbba794e6b187208c8
full-year 2026, management now guided that datacenter will account for 15-20% of total revenue and auto revenue will grow 30-50% YoY (at least 30% YoY, with 50% dependent on new-product ramp); it also expressed con fi dence in 2H26 growth to outpace 1H26.
On margin, management expects inventory impairment loss to reverse in coming quarters as products are being sold, and it expects 2H26 GM to be better than 1H26 with 1Q26 representing the GM trough. From a long-term perspective, management sees structural improvement in GM supported by higher 12-inch revenue contribution and rising datacenter/auto contribution. Management expects Gen-4 products (auto and datacenter are main applications) to reach 20% or more of revenue by the end of 2026, driven by 8-inch supply tightness which pushes customers to migrate toward 12-inch solutions, where Silergy's capacity is about 3x its 8-inch capacity. Management views this as a good opportunity to gain share, as 12-inch solution o ff ers superior performance with competitive cost.
Overall, Silergy sees auto and datacenter as the two primary growth drivers going forward. For auto, management sees growth to be driven by 1) higher semiconductor content from EV/hybrid EV, 2) expansion of Silergy's addressable content per vehicle, and 3) mainland China localization policy with smaller vendors exiting the industry due to tight 8-inch capacity. In datacenter, management focuses on general server motherboard components for mainland China customers and optical module, where the 800G to 1.6T transition could approximately double the content value. On the overall cycle, management views double-booking as still at an early stage and sees demand continuing to increase while supply declines as capacity shifts to AI.
WinWay (6515.TW; Buy)
Management spent considerable time introducing the company's business model and industry positioning, as we believe a larger number of investors at this year conference were relatively new to WinWay and less familiar with its role within the AI testing ecosystem. Beyond the company overview, the key message from the session was management's tone towards its aggressive capacity expansion plan. WinWay reiterated its plan to more than double its pin capacity in 2026 (spring-probe capacity is expected to increase from 6mn pins per month in 1H26, vs. 3.5mn at end-2025, to 9mn by end-2026) given strong demand from customers especially from the AI/HPC segment; the company also continues to target annual capacity doubling through 2030. At the same time, management maintained its positive stance on its long-term socket TAM growth outlook (recalled that the company sees the 50-60% growth as too conservative, see note). The company is also targeting a 50-60% in-house pin production ratio, which should support both supply security and future margin expansion.
Another key focus during the meeting is CPO opportunity. Volume ramp could potentially start in 2H26 following its customer's mass production timeline, with WinWay remains primarily focused on sockets for CPO chips rather than test equipment. Management highlighted several incremental opportunities from CPO adoption, including optical engine (OE) die test sockets and higher-content test sockets driven by larger chip sizes. Management indicated that CPO products could carry meaningfully higher content than non-CPO versions, primarily due to signi fi cantly longer testing times.
e92c7a75ab8b4efbba794e6b187208c8
MPI (6223.TWO; Buy)
Management guided 2026 revenue growth at around 30% YoY with GM of 56% or above. It expects 2Q26 revenue to be up 20% with GM near 58%, due to a greater mix of lower-margin product, followed by 10-15% QoQ revenue growth in 3Q26/4Q26 and GM recovery back towards 1Q26 level on higher ASIC contribution.
On CPO, management expects the Insertion 2 quali fi cation result to come out in 3Q26 and sees an opportunity to be the sole provider for Insertion 2 double-sided prober once quali fi ed. Management also noted that only MPI and another foreign competitor have the capability to o ff er optical/electrical testing solutions, with MPI's solution appearing more automated among the two vendors. Insertion 3 is already shipping small volumes to fabless and OSAT clients with no further quali fi cation needed. MPI's Insertion 3 solutions target the die level with clients potentially adopting both probe card and socket based solutions. Overall, as CPO equipment is margin accretive, management believes a 60% GM is achievable under the scenario where FX remains favorable, UTR maintains at a high level, CPO revenue contribution increases, and PCB self-su ffi ciency ramps up.
On CPU, management is in discussion with a leading CPU provider, with initial shipment in 4Q26 targeting legacy products, alongside quali fi cation for its latest N2 products, using all-MEMS solutions. We believe MPI may be gaining share from another foreign competitor that is refocusing on the ASIC supply chain, with the customer being keen to secure su ffi cient probe-card supply. Overall, we expect MPI to continue expanding its MEMS capacity to better capture both ASIC and non-ASIC opportunities. Management reiterated its previous capacity guidance, with MEMS expected to reach 3.5mn pins per month and VPC to reach 2mn pins per month by 4Q26, and 2027 MEMS capacity additions is expected to be no less than 2026.
Aspeed (5274.TWO; Buy)
Management reiterated its 3Q26 guidance with revenue of NT$4.1-4.3bn and GM of 6768%. While supply constraints remain the key bottleneck into 3Q26, management expects gradual improvement in substrate and backend capacity availability into 4Q26 and 2027, supported by the addition of a second OSAT partner in 4Q26 and the quali fi cation of E-glass supply for AST2700. Importantly, Aspeed indicated that it is likely to see another round of price hikes in 2H26, including AST2700, as backend/material costs continue to rise, with the key objective of maintaining corporate GM at the similar levels.
Another highlight is the announcement of its AST1840 chip, developed jointly with Lattice, which integrates FPGA functionality directly into Aspeed's chip. Notably, management believes AST1840's long-term shipment opportunity could eventually be comparable to the traditional BMC market, with volume production expected by end-2027. Historically, customers required a separate FPGA device alongside the BMC or SMC to handle I/O expansion and system management functions. AST1840 consolidates these functions into a single chip, reducing board complexity, shortening development cycles, and lowering software maintenance requirements for customers. We view this as a meaningful evolution of Aspeed's platform management roadmap and a potential catalyst for content expansion across future server designs.
Management remains highly constructive on underlying demand trends, noting that e92c7a75ab8b4efbba794e6b187208c8
customer lead times have now extended beyond 36 weeks and that many CSP customers e ff ectively view a large portion of their 2027 capacity as already committed. We continue to believe this re fl ects a structural rather than cyclical demand environment, driven by ongoing AI infrastructure deployment and increasing server complexity. Notably, management indicated that it may need to revisit its long-term BMC TAM assumptions, with a potential update expected by the end of 3Q26.
We continue to see a more favorable setup for Aspeed's earnings outlook, driven by 1) emerging agentic AI demand, which management believes is creating another wave of general-purpose server deployment and could drive AI-related general server demand from ~2.1mn units in 2025 to ~9mn units in 2026, 2) accelerating AI ASIC and TPU deployments, 3) faster-than-historical BMC adoption for AST2700, and 4) continued pricing support amid supply constraints. Management expects AST2700 to become the fastest-ramping BMC generation in company history, potentially surpassing 50% mix within two years versus the historical 2.5-3-year transition cycle, supported by the EOL of DDR4, tighter cybersecurity requirements, and migration toward next-generation server platforms. With AST2700 ASPs at US$22-25 vs. ~US$16 for AST2600, we continue to expect meaningful blended ASP expansion over the next several years.
Looking further ahead, management said that AST2800 development has already begun with CSP customers. Built on 6nm technology, AST2800 is expected to deliver substantially higher performance, lower power consumption, next-generation memory support, and enhanced security features. Importantly, management expects AST2800 ASPs are likely to be 2x of AST2700 levels (as cost is around 2x higher) while maintaining similar gross margins. With fi rst samples expected in 1Q28 and volume production targeted for 2029, we view AST2800 as another major content expansion opportunity that should extend Aspeed's growth runway well into the next decade.
e92c7a75ab8b4efbba794e6b187208c8
Investment Thesis, Valuation and Risks
Investment Thesis - Mediatek
MediaTek is a leading global IC design house specializing in smartphone AP (application processor). We are positive on MediaTek, with the global smartphone market and its revenue to resume growth momentum from 2025E onwards. We think MediaTek is in a good position to transition to an AI play from a traditional smartphone AP provider, starting from AI smartphones, to enterprise ASICs and smart auto (partnering with Nvidia) solutions, etc., in 2025E-26E. We expect MediaTek to deliver decent multi-year growth, with revenue/earnings to grow by 16%/21% CAGR from 25-27E driven largely by 1) market share gains, especially in the premium segment (in particular, high-end 5G fl agship SoCs), 2) smartphone upgrade cycle driven by GenAI demand, 3) new TAM (total addressable market) into enterprise ASIC and auto/computing markets.
Price Target Risks and Methodology - MediaTek (2454.TW)
Valuation: We have a 12m TP of NT$5,000. Our TP is based on a target P/E multiple of 25x (1.8 stdv above its 5-year trading average) applied to our 2H27E-1H28E EPS.
Key risks to our views: (1) Weaker-than-expected end demand especially with smartphones, (2) Higher foundry cost to impact its margin outlook, (3) Intensifying competition would result in change in pro fi tability as competition would normally lead to change in pricing dynamics, and (4) Slower ramp in ASIC would result in changes in operating leverage
Investment Thesis - Realtek
Realtek is an IC design house specializing in communications network ICs, computer peripheral ICs, as well as multimedia ICs. While we still hold a positive view towards its organic TAM (total addressable market) growth driven by ongoing technology migration and dollar content increase, we see limited potential for a strong upward earnings revision in the near-to-mid term given the lack of a new growth engine. While Realtek has recently tapped into the AI server segment through its new SSD (solid-state drive) controller, we believe the contribution should not be meaningful in the near term.
In addition, we believe Realtek's revenue growth momentum is likely peaking in 1H26, due to an earlier-than-expected order pull-in from customers partially due to concerns over further memory price hikes. We are now expecting weaker momentum for the rest of 2026 post a robust 1H26. With the shares now trading largely inline with its through-cycle avg. of 17x, we have a Neutral rating on Realtek.
Price Target, Risks and Methodology - Realtek
Valuation : We have a 12m TP of NT$560. Our TP is based on a P/E multiple of 17x (in line with its 5-year trading avg.) applied to our 2H26E-1H27E EPS.
Key risks including: (1) worse/better-than-expected electronics end demand; (2) slower/faster progress with automotive Ethernet; (3) pace of WiFi spec migrations; (4) company's execution and market shares; and (5) geopolitical and FX movements.
e92c7a75ab8b4efbba794e6b187208c8
Investment Thesis - Vanguard (5347.TWO)
Vanguard (VIS) is a pure foundry player in Taiwan focusing on legacy technologies. We expect VIS to grow its revenue into 2026E/2027E driven by 1) better-than-expected supply/demand dynamics with stable pricing, as we believe most of the expansion is mainly concentrating on 12-inch mature nodes such as 22/28nm instead of 8-inch capacity, 2) stronger-than-expected share gains especially in PMIC (power management IC), bene fi ting from the ongoing order shift from mainland China, which should continue to support its growth into 2026, and 3) customer demand coming in ahead of expectation for its incoming 12-inch fab with customer secured rates and LTAs (long-term agreements) progressing better than we expected, exceeding 60% of LTA rate.
However, the company is now trading +2.0 stdv above its 10-year avg. P/E of 18x, based on FY27E EPS. With VSMC likely to pressure Vanguard's pro fi tability starting 2027 and the lack of meaningful like-for-like price hikes for its 8-inch wafer business, we do not see meaningful upside potential to its earnings in 2027E. With stretched valuation and limited earning upside potential, we thus are Sell rated on Vanguard.
Price Target Risks and Methodology - Vanguard (5347.TWO)
Valuation: We are Sell rated on Vanguard with a 12m TP of NT$138. Our TP is based on a target P/E multiple of 20x (+0.5 stdv above its 10-year avg. forward P/E) applied to our 2027E EPS.
Key risks to our views: (1) Stronger pace of end-demand recovery; (2) Faster ramp with better pro fi tability from the 12-inch Singapore fab; (3) smaller-than-expected price erosion with potential price hike; and (4) an easing competitive landscape
Investment Thesis - Silergy (6415.TW)
Silergy is a leading domestic PMIC design house, and we remain constructive on its long-term growth outlook as the company continues to expand into higher-end applications such as servers/datacenter, HPC, and automotive. We expect Silergy to return to its long-term revenue growth trajectory of 20-30% YoY over the coming years, supported by continued dollar-content expansion in both the datacenter and automotive segments. As revenue contribution from these businesses increases, we believe Silergy's gross margin pro fi le should improve structurally, driven by the rising penetration of margin-accretive Gen 4 solutions. Overall, Silergy is currently trading below its historical upcycle valuation of 30x FY27E P/E, which we view as an attractive entry point. We therefore maintain a Buy rating on the stock.
Price Target Risks and Methodology
Valuation: Our 12m TP is NT$650, based on a target P/E of 30x (inline with its historical upcycle P/E) applied to our 2027E EPS.
Key risks to our views: 1) Slower end-demand recovery, 2) Worse-than-expected cost controls, 3) slower progress with new addressable markets, 4) an intensifying competition landscape, and 5) geopolitical risks.
Investment Thesis - WinWay (6515.TW)
WinWay is a leading Taiwan-based socket provider specializing in test sockets and e92c7a75ab8b4efbba794e6b187208c8
burn-in sockets, which are used in Final Test (FT), System-Level Test (SLT), and burn-in test. The company is the 2nd largest socket supplier globally in terms of revenue, with c.8% of market share (in 2024). 56% of its total revenue comes from the socket segment.
We like WinWay as we expect its revenue/earnings to further accelerate at 78%/109% CAGRs in 2025-28E, mainly driven by 1) continued dollar content expansion in next-gen AI chips, 2) ongoing share gains as shipment volumes ramp following its new penetration into its US AI GPU customer's SLT market beyond the FT market, and 3) even stronger volume growth from its US AI ASIC customers across both AI accelerators and CPUs, 4) robust CPU socket demand, driven by agentic AI demand to general server CPU, and 5) stronger revenue contribution from MEMS probe cards supported by CPU and networking demand. The company is now trading above its 3-yr. average forward P/E of 31.5x; however, with increasing ASP thanks to rising chip complexity and shorter product cadence driving frequent socket upgrade demand, we continue to see upside potential to its revenue and earnings growth and have a Buy rating on the name.
Price Target Risks and Methodology - WinWay (6515.TW)
Valuation: We are Buy rated on WinWay. Our 12-month TP of NT$15,000 is based on a target P/E of 40x (1.26x of 3-year average forward P/E of 31.5x) applied to our 2028E EPS, and discounted back to 2027E at 13.6% CoE. Our key assumptions for CoE include: (1) 1.5x beta (sourcing Bloomberg), (2) a 4.25% risk-free rate, and (3) a market risk premium at 6.25% (in line with GS house view).
Key risks to our views: 1) softer AI/HPC demand, 2) slower penetration into new TAM, and 3) intensifying competition.
Investment Thesis - MPI (6223.TWO)
MPI is a leading Taiwan-based probe card provider specializing in CPC (Cantilever Probe Card), VPC (Vertical Probe Card), and MEMS (Micro-Electro-Mechanical Systems) Probe Card, which are used in the chip probe (CP) stage. The company is the 4 th largest probe card supplier with c.7% market share (2024), and 57% of its total revenue comes from the probe card segment.
We like MPI as we expect its revenue/earnings CAGR to further accelerate to 69%/97% in 2025-28E, driven by 1) continued share gain in vertical probe card (VPC) market via AI ASIC, and 2) potential share gain in MEMS probe card market through penetration into new both AI and non-AI customers, along with AI ASIC upgrade. MPI is also aiming to increase its PCB self-su ffi ciency to enable shorter product delivery lead time, which would provide it with more advantages over peers. The company is now trading above its average AI cycle forward P/E; however, with increasing ASP thanks to rising chip complexity and shorter product cadence driving frequent probe card upgrade demand, we continue to see upside potential to its revenue and earnings growth. We have a Buy rating on the name.
Price Target Risks and Methodology - MPI (6223.TWO)
Valuation: We are Buy rated on MPI. Our 12-month TP of NT$8,800 is based on a target P/E of 40x (1.7x of 3-year average forward P/E) applied to our 2028E EPS, and discounted back to 2027E at 12.4% CoE. Our key assumptions for CoE include: (1) 1.3x e92c7a75ab8b4efbba794e6b187208c8
beta (sourcing Bloomberg), (2) a 4.25% risk-free rate, and (3) a market risk premium at 6.25% (in line with GS house view).
Key risks to our views: 1) softer AI/HPC demand, 2) slower penetration into new TAM, and 3) intensifying competition.
Investment Thesis - Aspeed (5274.TWO)
Aspeed is a leading Taiwan-based IC (integrated circuit) design company specializing in BMC (baseboard management controller) SoC, which is used to remotely monitor the physical state of servers. The company is the largest BMC supplier, with c.70% of the global BMC market share, and 90% of its total revenue comes from the BMC segment.
We like Aspeed as we expect its revenue/earnings CAGRs to further accelerate to 31%/37% in 2025E-27E (vs. 5.9%/4.9% during 2019-2023), driven mostly by AI server demand, especially starting 2025E. Furthermore, we also expect server architecture changes will bring more business opportunities to Aspeed to drive up its TAM expansion within the server market. The company is now trading at the higher end of its historical P/E range; however, with more TAM expansion and more dollar content growth within the AI theme, we continue to see upside potential to its revenue and earnings growth. We have a Buy rating on the name.
Price Target Risks and Methodology - Aspeed (5274.TWO)
Valuation: Our 12-month TP of NT$22,000 is based on a target P/E of 40x (in line with its 10-year historical trading average) applied to our 2028E EPS, and discounted back to 2027E at 13.6% CoE. Our key assumptions for CoE include: (1) 1.5x beta (sourcing Bloomberg), (2) a 4.25% risk-free rate, and (3) a market risk premium at 6.25% (in line with GS house view).
Key downside risks to our view and TP: 1) A softer-than-expected server market demand recovery, 2) slower-than-expected BMC penetration within AI servers, 3) intensifying market competition.
e92c7a75ab8b4efbba794e6b187208c8