報告_MS_ABF載板擴產與供給吃緊_20261020

PDF 原檔：報告_MS_ABF載板擴產與供給吃緊_20261020_original.pdf

校正紀錄

• 2026-10-26：本檔為 MS 於 2026-10-19 以「Correction」重發、2026-10-20 版 PDF 之文字（原版 2026-10-14，見 aliases）。原版 PDF 抽字大量缺漏字母 h（如 t e、wit ），以本版可讀文字取代；兩版數字、評等、目標價相同。本版差異：①美系公司科技元件估計明確標示為亞洲供應鏈查核估計，Exhibit 6-11 標題與出處改為 Asia supply chain estimates；②新增「Clarifying AMD unit assumptions」段。
• 圖片沿用原版抽圖（報告_MS_ABF載板擴產與供給吃緊_20261014_NNN.png，同一組 exhibit，已於 2026-10-15 驗證），本版重複抽圖未保留。
• 本版抽字仍有少數缺漏字母 l（如 Supp y、comp exity），原文如此，未改。

圖片清單（已驗證 2026-10-15）

ingest 時建立的「眼見為憑」圖片索引；lib/ 嵌圖只從這裡挑真資料圖。小圖（<40KB）亦已逐張 Read。

原始內容

M Idea October 20, 2026 02:30 AM GMT

Greater China Technology Hardware | Asia Pacific

Correction: ABF Substrates: More Capex, More Tightness

New capacity announcements do not overturn our bu ish view. AI demand and increasing substrate comp exity shou d keep the market undersupp ied from 2027, supporting further pricing and margin expansion. We raise earnings estimates and price targets for Unimicron and NYPCB and reiterate OW on both.

We have made alterations to the text and exhibits relating to tech component estimates involving US companies and their Asia-based supply chains to make clear that these estimates are from Asia-based supply chain checks. We also note that our Asia supply chain model estimate cut for AMD s 2026 GPU shipments was in large part due to a prior modelling error. This article was originally published on 14 Oct 2026 8:00 AM GMT, and is being republished on 19 Oct 2026. To receive an electronic copy of the original version of this publication, please send an e-mail to Equity_Research_Publishing@morganstanley.com with the publication title and date.

Clarifying AMD unit assumptions: Please note the changes in our ABF substrate model do not represent e uivalent reductions in our company-level AMD GPU shipment forecasts (covered by Joseph Moore). For 2026, our October 14 model estimates AMD GPU substrate units of 0.6mn versus 2.4mn in the July 6 model, primarily because the earlier figure incorrectly included approximately 1.4mn Venice CPU units in the GPU category and also incorporated additional substrate-planning buffers. For 2027, our model estimates 1.3mn units versus 2.7mn previously, reflecting the incorporation of our Asia semiconductor team's updated CoWoS assumptions, which changed from approximately 2.3mn AMD GPU shipments as of June 23 to 1.3mn as of September 10; accordingly, the October 14 model does not represent an additional reduction beyond the revised CoWoS assumptions already published. These substrate-unit assumptions are inputs to our ABF supply/demand analysis and, given differences in scope, planning buffers and update timing, should not be interpreted as one-for-one forecasts of finished AMD GPU shipments. For context, Joe Moore's company-level AMD GPU shipment estimates as of August 5 were approximately 0.7mn for 2026 and 1.1mn for 2027. See Joe's report here for further context.

More capacity, but limited near-term relief: Suppliers are stepping up investment, but several newly announced projects are not scheduled to ramp until late 2028-30 or beyond. Our updated supply-demand model continues to indicate undersupply from 2027 through 2030, with tighter supply in CY27-28, which should only be more favorable to pricing. Therefore, capacity additions announced in the past few months do not, in our view, diminish the magnitude of pricing gains over the next two years.

CY 7 pricing should accelerate, amid tighter supply-demand in CY 7- 8, but

Morgan Stanley Taiwan Limited+

Howard Kao

Equity Analyst

H ward.Ka @m rganstanley.c m

+886 2 2730-2989

Irene Yen

Equity Analyst

Irene.Yen@m rganstanley.c m

+886 2 2730-2869

Morgan Stanley Asia Limited+

Andy Meng, CFA

Equity Analyst Andy.Meng@m rganstanley.c m

Greater China Technology Hardware

Asia Pacific Industry View In-Line

+852 2239-7689

Exhibit 1 What's changed

Source: FactSet, Morgan Stanley Research (e) estimates. Pricing as of the close on ctober 12, 2026.

Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.

For analyst certification and other important disclosures, refer to the Disclosure Section, located at the end of this report.

= Analysts employed by non-U.S. affiliates are not registered with FINRA, may not be associated persons of the member and may not be subject to FINRA restrictions on communications with a subject company, public appearances and trading securities held by a research analyst account.

M Idea

visibility on the magnitude remains limited: We see ABF substrate prices rising 25%+ in CY27, below buy-side expectations of ~50%. Our supply/demand outlook points to undersupply next year, with larger substrates, higher layer counts, and greater manufacturing complexity further constraining usable output. Raw material cost pass-through was a key driver of CY26 price increases, we expect supply tightness to play a greater role in CY27. This should allow pricing to outpace cost inflation, driving faster margin expansion. We remain constructive on pricing, but the magnitude and timing of increases depends on customer-level negotiations.

With pricing a segment-wide tailwind, we prefer Unimicron for its stronger CY 7 capacity growth: As supply tightens, the ability to deliver incremental volume should become a key differentiator. We favor suppliers bringing capacity online in CY27 over those whose expansion plans contribute in later years. Unimicron's Guangfu II ramp should add 10-15% to capacity in CY27, compared with 5-10% growth at NYPCB through debottlenecking, giving Unimicron greater scope to benefit from both pricing and volume growth. Unimicron is also our preferred play heading into earnings, as more demanding expectations for NYPCB leave less room for an upside surprise. But we remain OW on NYPCB given its networking and AMD CPU exposure, alongside its proactive pricing strategy.

We reiterate Overweight on Unimicron and NYPCB: Reflecting stronger earnings trajectories, we raise our EPS estimates for both companies. Based on upside to our revised price targets, our order of preference is Unimicron (NT$1,775 PT; 24x CY28e P/E; 41% upside), followed by NYPCB (NT$2,000 PT; 22x CY28e P/E; 39% upside).

Potential risks: 1) Weaker demand for PCs, general servers, and AI servers; 2) meaningful capacity expansion plans; 3) T-glass shortage; 4) CoWoP displacing ABF substrates.

M Idea

Supp y/Demand Update: Undersupp y Deepens in CY27-28

Our updated model points to a wider supply shortfall in CY 7- 8, despite additional capacity: Incorporating the latest expansion plans and semiconductor demand forecasts, we now expect ABF substrate demand to exceed available supply by 9.3% in CY27 and 13.7% in CY28, versus 6.2% and 9.4% previously. This represents a widening of 3.1ppt and 4.3ppt, respectively, relative to our prior estimates. We maintain our view that CY26 is a broadly balanced market, followed by increasingly tight supply over the next two years.

The main demand upgrade comes from other ASICs, including Chinese AI chips: We raise our shipment assumptions for this category to 6.4mn units in CY27 and 8.7mn in CY28, from 4.6mn and 6.2mn previously. Our Asia supply chain forecasts for Nvidia GPUs and server CPUs remain largely unchanged, while we lower supply chain estimates for AMD GPUs, but also keep CY27 Trainium and TPU shipments largely unchanged (note, these are separate from Joe Moore's company forecasts). The tighter substrate outlook therefore reflects stronger demand across a broader set of ASIC suppliers, rather than higher shipment assumptions for every major AI platform. We also raise our substrate content assumptions for select chips based on our latest supply chain checks, lifting our substrate demand estimates across some segments, including server CPUs and networking.

Higher supply assumptions provide limited relief in CY 7- 8: We now include additional capacity announcements that have been announced recently, including two additional Kinsus factories, scheduled to come online in CY28 and CY29, beyond the CY27 expansion already in our model. Several larger projects, however, contribute later. NYPCB's new site starts ramping in 2H29, while SEMCO's Sejong expansion begins mass production in September 2028 and its project in Vietnam is scheduled to begin operations by June 2030. These additions increase medium-term supply but do not prevent the modeled shortfall from worsening through CY28.

Increasing substrate complexity adds to the pressure on usable output: Larger body sizes, higher layer counts, and more demanding manufacturing processes constrain the output available to meet demand, even as suppliers expand installed capacity. With the supply gap widening in CY27-28, we expect tightness to become a more important pricing driver than raw material cost pass-through. This supports our forecast for 25%+ price increases in CY27 and further increases in CY28, alongside continued margin expansion.

M Idea

Exhibit 2 Expecting ABF substrate undersupply gap t widen by 2030 vs. ur pri r estimates

S urce: M rgan Stanley Research (e) estimates.

Exhibit 3 Increased demand assumpti ns cause undersupply t persist thr ugh 2030

S urce: M rgan Stanley Research (e) estimates.

Supply-side developments

Unimicron 3037.TW, OW, covered by Howard Kao)

On September 30, Unimicron announced that it has ac uired a Hukou site from Bridgestone Taiwan for ~NT$10.089bn (~US$318mn). Based on our checks, this is for capacity expansion plans in 2030 or beyond.

For CY26, Unimicron has announced capex of NT$53.7B.

Over the next few years, Unimicron already has several new production facilities coming online, including:

• Guangfu II: ramp in 2027
• Yangmei II: ramp in 2028
• Yangmei III: ramp in 2029
• Yangmei IV: ramp in 2029+
• Bridgestone: ramp in 2030+

M Idea

NYPCB (8046.TW, OW, covered by Howard Kao)

On August 6, NYPCB announced that its Board of Directors had approved a capex budget of NT$46.8bn (~US$1,474mn) to invest in building a new ABF substrate production site. Our checks indicate that this production site is located in the southern part of Taiwan, and will start to ramp up production in 2H29. Given the size of this investment, we estimate that it will approximately double its capacity exiting CY30, relative to its CY28 capacity.

Kinsus 3189.TW, Not Covered)

In our previous model, we had assumed only one factory (k6b) coming online in CY27, increasing capacity ~30%. However, recent checks suggest that Kinsus intends to bring two more factories online, one in CY28 (k6c) and one in CY29 (k6d). This implies that capacity would increase another ~25% in CY28 and ~20% in CY29, bringing Kinsus' total ABF substrate capacity to almost double exiting CY29, relative to its CY26 capacity.

Ibiden 4062.T, covered by Shoji Sato)

Back in February, Ibiden first announced that it will invest capex of ¥500bn in the electronics field planned through F3/29, for ABF substrate expansion. Sato-san's latest view is that Ibiden will need a new factory building to install the e uipment and facilities that will be purchased with the ¥500bn. As a result, Sato-san expects Ibiden during its 2Q results briefing (October 30) to announce plans to make a large capital investment, including for a factory building with a long lead time, and raise its medium-term earnings forecast (report here).

SEMCO 009150.KS, covered by Shawn Kim)

On September 29, SEMCO announced investment of KRW4.27 trillion (~US$3,178mn) in Sejong and KRW2.51 trillion (~US$1,488mn) in Vietnam, for AI server semiconductor substrate capacity expansion.

• Sejong: Investment period from September 2026 to May 2028, with mass production starting in September 2028. Largest single-product investment to date, serving as key hub for mass production of high-value-added, highly demanding products, proactively responding to the increasing demand for AI server substrates.
• Vietnam: The Vietnamese subsidiary aims to complete construction and begin operations by June 2030. It will be operated as a strategic hub to respond to global volume demand and enhance production efficiency through a stable supply system.

With this investment, Samsung Electro-Mechanics plans to actively respond to the growing demand for high-value-added semiconductor substrates in high-growth markets such as AI servers, HPC, and automotive electronics. In particular, this investment is being pursued based on medium- to long-term cooperation with global big tech and semiconductor customers. The investment includes customer funding support and medium- to long-term volume guarantees; thus, the company expects to strengthen the competitiveness of its FCBGA business based on a stable demand foundation.

LG Innotek 011070.KS, covered by Ryan Kim)

M Idea

According to news reports (link), LG Innotek signed a memorandum of understanding (MOU) with Haiphong, Vietnam, for an investment to expand its semiconductor substrate plant. Under the agreement, LG Innotek will push to build a semiconductor substrate plant in Haiphong.

• Construction will begin in July 2026 and is scheduled to be completed in May 2027.
• The site spans approximately 330,000 s uare meters (98,000 pyeong), e uivalent to 45 soccer fields.
• According to a news report, LG Innotek will invest approximately $1 billion, or about 1.53 trillion won, for its Phase 1 expansion plan, which mainly includes RF-SiP (radio fre uency system-in-package) and FC-CSP (flip chip-chip scale package).
• ABF substrates will be part of the Phase 2 expansion plan, but no details or investment size has been disclosed, yet.

According to a press release from LG Innotek on September 30, LG Innotek started mass production of ABF substrates for PC chipsets targeting big tech clients last year. It initiated high-volume mass production of FC-BGA for PC CPUs in the third uarter of this year. Simultaneously, the company is rapidly boosting revenue by securing additional major clients. Starting next year, it plans to se uentially enter the AI accelerator and AI server markets.

• Revenue from LG Innotek's package solution business has maintained an upward trajectory for three consecutive years. With the package solution business demonstrating tangible results alongside its optical solution business, the company has strengthened its capacity to generate stable earnings.
• LG Innotek plans to rapidly expand its market share in the emerging semiconductor substrate market, aiming to build its package solution business, including semiconductor substrates, into a business generating US$743 million in operating profit by 2031.
• In the future, LG Innotek plans to develop its future businesses, including semiconductor substrates and physical AI solutions, to US$5.94 billion by 2030. The company aims to have such future new businesses account for 25% of total revenue, thereby completing a well-balanced business portfolio.

The expanded plant will produce major semiconductor substrates including radio fre uency system-in-package (RF-SiP), flip chip chip scale package (FC-CSP), and flip chip ball grid array (FC-BGA). "The Gumi site will serve as a 'mother factory' responsible for developing new semiconductor substrate technologies and producing new models and high-value-added products, while the expanded plant in Vietnam will be operated as a production base for general-purpose semiconductor substrates," an LG Innotek official explained.

AT&S ATSV.VI, Not Covered)

On June 15, AT&S announced that it had signed key terms with AMD and Marvell to add AI/HPC IC substrate capacity in Kulim, Malaysia. The company said the €1.5-2bn investment (approximately US$1.68bn to US$2.24bn) is fully supported and financed by long-term customer commitments, pending final execution.

M Idea

Shinko private)

On September 30, Shinko announced that it has entered into a purchase agreement with Japan Display Inc. (JDI) to ac uire its Mobara fab. In addition, it has entered into a sale agreement with FICT Ltd. for the transfer of a portion of the fab to be ac uired.

• Shinko will position the fab as a new core plant for advanced semiconductor packages, and strengthen its production systems for flip-chip type packages through the determination and implementation of capital investment plans for the fab in conjunction with capacity expansion at its plants in the Nagano area. Through these initiatives, Shinko aims to respond accurately to strong market demand while pursuing sustainable business growth and enhancing corporate value.
• The JDI Mobara fab has a total land area of 373,483 s uare meters, with Shinko using the eastern area of the Fab (140,738 s uare meters, or ~38%) and FICT planning to use the western area (232,745 s uare meters, or ~62%).

Toppan 7911.T, Not Covered)

According to news, Toppan broke ground in March 2024 on a new factory on Pesawat Drive near Singapore's Jurong Lake District, built through its local unit Advanced Substrate Technologies (AST), a JV with Broadcom. It makes FC-BGA substrates, which is Toppan's term for them. It is Toppan's first FC-BGA plant outside Japan. The plant opening ceremony took place on September 30, 2026.

• AST operates a facility of approximately 70,000 m² in the vicinity of Jurong Lake District and employs more than 200 highly skilled engineers and technicians.
• Focus on high-demand, large-size, high-layer-count substrates
• While the investment size is not disclosed directly by the company, Nikkei sized the investment at around ¥50 billion (~US$317mn).

Daeduck 353200.KS, Not Covered)

Daeduck announced on July 20, 2026, that it is investing ₩497 billion (~US$370M) manufacturing facilities and related infrastructure for semiconductor products, running through December 31, 2027.

ABF capex is stepping up, but announced investment still falls short of implied

demand. We estimate that ABF substrate makers have announced ~US$14.3bn of investment year to date, materially above the segment's aggregate annual capex in recent years, although these commitments span multiple years. Our tally is not exhaustive. LG Innotek has yet to disclose the size of its planned ABF capacity investment, while Shinko has not uantified the additional spending re uired to expand capacity following its announced ac uisition of a JDI fab.

Our supply chain checks suggest that each US$1 of investment can support US$0.500.60 of annual revenue at full ramp. On this basis, the US$14.3bn announced so far could support US$7.2-8.6bn of incremental annual revenue. This compares with US$11.4bn of incremental revenue implied by our industry forecast. We model the ABF substrate market reaching US$20.9bn in 2030, up from US$9.5bn in 2026. Despite the step-up in M Idea capex announcements, the investments we can currently uantify appear insufficient to meet our demand forecast, pointing to a need for further capacity expansion over the coming years.

Exhibit 4 Ann unced ABF Substrate capacity investments

S urce: M rgan Stanley Research.

Exhibit 5 Maj r ABF substrate supplier capex (US$ mn)

Major ABF Substrate Vendors' Capital Expenditure (US$ M)

S urce: C mpany data, M rgan Stanley Research.

Stronger ASIC demand from hyperscalers

We have updated our ASIC demand forecasts in our Asia team's ABF model to align with the estimates from our Asia semiconductors team. As shown in the exhibits below, our revised assumptions point to a higher estimate for Amazon's Trainium processors in 2026, as well as marginally lowered shipment estimates for Amazon's Trainium processors and Google's TPUs in 2027, while the other estimates stay largely unchanged. These ASICs are characterized by large die sizes and high layer-count package substrates, and thus we believe the slightly lowered 2027 volume estimates do not derail our constructive outlook for the substrate suppliers.

M Idea

Exhibit 6 Changes t Asia supply chain estimates f r Amaz n Trainium

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

Exhibit 7 Changes t Asia supply chain estimates f r G gle TPUs

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

Other ASIC chips

For "Other" ASIC chips, we increase our estimates. Included in this section, based on Asia supply chain forecasts, are ASIC chips from Microsoft, other US companies, and Chinese AI chips, such as those from Huawei, Cambricon, Hygon, T-Head, etc. For Chinese AI chips, we also included the latest estimates from our semiconductor team, which drives the total "Other ASIC chips" category to reach ~13.4mn units by 2030, up from ~11.7mn units previously.

M Idea

Exhibit 8 Seeing increased demand fr m Chinese AI chips

"Other ASIC" chip, includes China AI chips (M units)

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

Changes to our GPU and Nvidia CPU forecasts

We also update our Asia supply chain demand forecasts for Nvidia and AMD GPUs. Our revised estimates generally point to lowered demand for AMD GPUs in 2026-30 compared to our previous assumptions, while our estimates for Nvidia GPUs remain largely unchanged. Our forecasts for Nvidia's server CPU portfolio (including Grace, Vera, and future platforms) also stay largely unchanged.

Exhibit 9 Changes t Asia supply chain estimates f r Nvidia GPUs

Exhibit 10 Changes t Asia supply chain estimates f r AMD GPUs

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

M Idea

Exhibit 11 Changes t Asia supply chain estimates f r Nvidia CPUs

Nvidia CPU (M units)

S urce: M rgan Stanley Asia-Pacific Research (e) estimates.

M Idea

Earnings Ca endar and Comps Tab e

Exhibit 12 Earnings Calendar

S urce: C mpany data, M rgan Stanley Research.

3Q preview: Morgan Stanley estimates vs. consensus

Unimicron

We expect a slightly stronger gross margin of 27.6%, ~30bps above the Street estimate, reflecting a more favorable product mix and improving pricing environment. On operating expenses, our estimate is roughly the same as consensus, resulting in an operating margin of 19.4%, ~70bps above the Street estimate. However, our non-op income estimate of NT $661mn is materially higher than the Street's estimate of NT$1,495mn in non-op loss. This leads to our EPS estimate of NT$5.21 being ~37% higher than the Street.

Exhibit 13 Unimicr n 3Q preview: We are 30bps ab ve the Street n 3Q GM

S urce: C mpany data, Visible Alpha, M rgan Stanley Research estimates.

NYPCB

Our estimates are modestly ahead of consensus, driven primarily by stronger operating profitability. Our operating margin estimate of 26.2% is 50bps above consensus (25.7%), reflecting stronger gross margin (29.2% vs. 28.9% consensus) and lower-than-expected operating expenses. As a result, our operating profit forecast exceeds consensus slightly (~3%), which, when combined with lower taxes, flows through to EPS of NT$5.84, around M Idea

7% above the Street estimate of NT$5.44. Overall, we assume earnings will come in better than expected, driven more by better profitability than by top-line performance.

Exhibit 14 NYPCB 3Q preview: we are 7% ab ve the Street n 3Q EPS

S urce: C mpany data, Visible Alpha, M rgan Stanley Research estimates.

Morgan Stanley vs. consensus

Exhibit 15 Unimicr n: M rgan Stanley vs. c nsensus estimates

S urce: Visible Alpha, M rgan Stanley Research estimates.

Exhibit 16 NYPCB: M rgan Stanley vs. c nsensus estimates

S urce: Visible Alpha, M rgan Stanley Research estimates.

Exhibit 17 ABF substrate peer c mparis n table

S urce: FactSet (f r AT&S and Kinsus), M rgan Stanley Research estimates (f r Unimicr n, NYPCB, Ibiden and SEMCO). NC = N t c vered by M rgan Stanley Research.

Related Reports:

Nan Ya PCB: Announces preliminary August profits (2 Oct 2026)

Greater China Technology Hardware: ABF substrates - Kinsus announces preliminary M Idea Greater China Technology Hardware: ABF substrates - AT&S announces expanded Greater China Technology Hardware: ABF Substrates - Implications of AT&S' F1Q26/27 Results (4 Aug 2026)

Nan Ya PCB: Margin improvement monthly in 2Q; total 2Q net profit 4% above MSe (30

August profit (23 Sep 2026) collaboration with Marvell (23 Sep 2026) Jul 2026)

Unimicron: 2Q26 post-results briefing takeaways (29 Jul 2026)

M Idea

The State of the Current Cyc e

The ABF substrate market has gone through a few cycles over the past two decades. We believe we are now still in the early innings of an up-cycle.

There was a prolonged down-cycle from 2012-18 as the PC TAM shrank amid the rise of smartphones and tablets. During this tough period, companies were not able to generate profits and had to compete on price to fill their idle capacity. As a result, during those years, there was no or very limited capacity expansion, and capital expenditures by the major ABF substrate suppliers were kept at a minimum.

Even as general server demand picked up, lifting utilization rates across the group going into 2020, ABF substrate suppliers remained cautious on the longer-term demand outlook and remained disciplined, hesitating to expand capacity. Hence, capex turned up just slightly in 2019-20, increasing ~US$600mn vs. ~US$2.2bn on average per year in 2012-18.

This underinvestment led to sudden undersupply, as the shock from Covid-19 drove strong work-from-home demand for PCs, as well as increased data usage and storage for general computing driven by online entertainment. This ultimately led to a significant increase in ABF substrate market value from 2020-22, reflected by the sudden surge in demand along with meaningful pricing adjustments owing to the lack of supply.

During this 2020-22 period, much capacity expansion was set in motion as ABF substrate suppliers and customers came to agree that necessary investments were needed for future growth. Some subsidies were even given to these suppliers to facilitate this capacity expansion. This is why capex jumped from ~US$2.9bn in 2020 to US$4.1bn in 2021 and US$5.5bn in 2022; even in 2023-24, capex remained elevated, as capacity expansion plans continued.

Unfortunately, demand did not play out as expected. As the world came out of the pandemic, demand for PCs and online entertainment fell sharply. This normalization of demand, coupled with capacity expansion from practically every major ABF substrate supplier, resulted in the ABF substrate market once again becoming oversupplied. This led to a sharp decline in ABF substrate market value in 2023-24, as volumes normalized, driving prices down to normalize against the new supply/demand dynamic.

Fortunately, cloud demand remained strong for general computing, and with AI demand starting to pick up, the down-cycle this time around was much shorter (two years) vs. the prior down-cycle (six years).

In 1H24, the ABF substrate market troughed and we saw an upward inflection since then, but the pace of the recovery has been moderate and bifurcated. Midrange/low-end markets such as PCs have shown moderate demand recovery, with high-end server and AIrelated end-demand recovering and growing at a faster pace. This is why we turned the page on our bearish stance in 1H25, but did not turn more bullish until early 2026.

M Idea

Periods

Description

Peri d I (2010-2012): PC-centric equilibrium ABF substrate demand is primarily driven and d minated by CPU and GPU demand f r deskt ps, n teb ks, discrete GPUs, with a meaningfully smaller exp sure t wards server CPUs.

Peri d II (2012-16): PC decline and demand diluti n

Peri d III (2016-18): The quiet inflecti n - servers and early HPC

Gl bal PC demand entered a structural d wn-cycle as c nsumer behavi r changed, with smartph nes and tablets abs rbing m re c nsumer time spent. This resulted in l wer v lumes and increased pricing c mpetiti n as suppliers f ught t fill idle capacity.

During these years, the PC market matured, while cl ud c mputing started t take ff. At the same time, server CPU substrates quietly grew larger with m re layers. We als started t see early accelerat rs being ad pted in high perf rmance c mputing.

Peri d IV (2018-2022): WFH demand driven structural sh rtage

Peri d V (2022-2024): P stpandemic c rrecti n and new capacity arrival The C vid-19 pandemic dr ve massive w rk fr m h me demand f r c nsumer electr nics and cl ud c mputing. This sudden surge in PC demand and increased time spent n gaming and nline entertainment, bundled with a c ntinu us increase in substrate package size and layer c unt, resulted in a structural sh rtage, which als resulted in l nger lead times and raised prices. It was during this peri d (2021-22) that ABF suppliers gl bally started t invest in new capacity.

As the w rld came ut f the C vid-19 pandemic, WFH demand faded and PC and cl ud c mputing demand were negatively affected. But at the same time, the new investments and capacity additi ns were just starting t sl wly c me nline, resulting in a dr p in utilizati n rates and pricing.

Peri d VI (2024-2026):

Bifurcated rec very PC demand n rmalized at the pre-C vid levels, with s me replacement demand driven by Wind ws upgrade, while the rec very and gr wth f AI and general server demand f r cl ud picks up at a meaningfully higher gr wth traject ry.

Peri d VII (2026-30): AI diffusi n Demand f r AI pr liferates fr m training t inference, driving ABF substrate market gr wth t re-accelerate ver the next five years after a tw -year d wn-cycle f ll wing the C vid-19 pandemic. This drives increased demand f r AI c mputing, netw rking, and cl ud c mputing, while demand f r PCs remains s ft.

Exhibit 18 We are entering the AI diffusi n era

S urce: M rgan Stanley Research (e) estimates.

M Idea

Exhibit 19 We n w expect ABF substrate demand gr wth t accelerate - t a 22.2% CAGR ver 2025-30e

S urce: Prismark, M rgan Stanley Research (e) estimates.

Breaking down ABF substrate end demand

The end market mix for ABF substrates has shifted dramatically over the past decade. PCs were once the dominant demand driver, accounting for approximately 70% of global ABF substrate market value in 2015, supported primarily by PC CPUs and GPUs. Since then, however, the industry's demand profile has undergone a structural transformation. We estimate that PC-related applications represented only ~23% of total ABF substrate market value in 2025, and this share is expected to decline further to below 10% by 2030.

In contrast, demand from servers, AI GPUs, AI ASICs, and networking applications has expanded rapidly. We estimate that these segments collectively accounted for only around 10% of the ABF substrate market in 2015, but their share increased to ~60% by 2025. Looking ahead, we expect they will represent more than 80% of total ABF substrate market value by 2030.

In our view, this structural shift in end-market demand has been the primary driver of ABF substrate industry growth over the past decade and will become even more pronounced over the next five years. As AI infrastructure deployment accelerates, we expect continued expansion in server, AI accelerator, and networking demand to drive meaningful acceleration in ABF substrate market growth.

M Idea

Exhibit 20 PC-related ABF demand is seen declining fr m ~70% in 2015 t <10% by 2030, while server, AI GPU, AI ASIC, and netw rking-related ABF demand sh uld rise fr m ~10% in 2015 t 75%+ by 2030

S urce: M rgan Stanley Research (e) estimates.

This report references U.S. Executive Order 14032 and/or entities or securities that are designated thereunder. U.S. persons ma be prohibited from bu ing certain securities of entities named in this report. Readers are solel responsible for ensuring that their investment activities are carried out in compliance with applicable laws.

This report references export controls and/or entities that ma be subject to export control restrictions. Readers are solel responsible for ensuring that their investment or trade activities are carried out in compliance with applicable laws.

This report references U.S. Executive Order 14105 and/or entities that ma be in scope of such order. U.S. persons ma be prohibited from engaging in certain transactions or otherwise require certain other transactions be notified to the U.S. Department of Treasur . Readers are solel responsible for ensuring that their investment or trade activities are carried out in compliance with applicable laws.

M Idea

Unimicron: Estimate Changes

We raise our 2026, 2027 and 2028 EPS estimates 62%, 30%, and 28%, respectively, primarily driven by our assumptions for better margin profiles.

Exhibit 21 Estimate Changes

S urce: M rgan Stanley Research (E) estimates.

M Idea

Price Target and Va uation Methodo ogy

Our 1 -month price target increases ~ 1% to NT$1,775 (from NT$1,465), driven by our increased earnings estimates for CY 6- 8: We believe the company's earnings have returned to a growth track from 2026 onward and will show stronger growth in 2027, driven by the high-end ABF substrate market's reversion to undersupply and Unimicron's dominant supply share in AI ASICs and also server CPUs, which puts it in a favorable position to benefit from rising AI investments.

We believe the overall ABF substrate market bottomed in 2H24, and see growth accelerating from 2H26 onward with the ramp of multiple next-generation, large body size ABF substrates for new server GPUs, CPUs, ASICs, and networking chips. New technologies such as embedded passives and Intel's EMIB-T should also increase the manufacturing complexity of the package substrates, which adds to the under-capacity as usable output decreases.

Similar to our analysis of other tech hardware companies within our coverage, we use a residual income (RI) valuation model to value Unimicron - we think it derives the most accurate value of the firm as it takes into account cost of e uity. Our key assumptions include:

• Cost of e uity of 9.2% (unchanged)
• ° Risk-free rate of 1% (10-year Taiwan government note yield);
• ° E uity risk premium of 8.7%;
• ° Beta of 1.0;
• Medium-term growth rate of 14% (slightly lowered from 15%)
• Terminal growth rate of 3% (unchanged, similar to the rest of the companies under our tech hardware coverage).

Our bull and bear case values increase by similar magnitudes, to NT$3,030 and NT $595, respectively.

Exhibit 22 Unimicr n residual inc me (RI) m del

S urce: M rgan Stanley Research (E) estimates.

M Idea

Exhibit 23 Unimicr n P/E chart

S urce: TEJ, M rgan Stanley Research estimates.

M Idea

Financia s

Exhibit 24 Quarterly Estimates

S urce: C mpany data, M rgan Stanley Research (E) estimates.

M Idea

Exhibit 25 C ns lidated Financial Summary

Income Statement

Cash Flow Statement

Balance Sheet

S urce: C mpany data, M rgan Stanley Research (E) estimates.

-9,100 Other adjustments

0 Other adjustments

85,833 146,283 Other adjustments

Financial Ratios

105,298 167,511 271,568 413,104 Gross Profit

836 Operating Profits M Idea

NYPCB: Estimate Changes

We raise our 2026, 2027 and 2028 EPS estimates 20%, 46%, and 43%, respectively, primarily driven by higher assumptions for AI semiconductors volumes and higher margin profiles.

Exhibit 26 Estimate Changes

S urce: M rgan Stanley Research (E) estimates.

M Idea

Price Target and Va uation Methodo ogy

Our price target increases ~ 9%, to NT$ ,000, reflecting our raised earnings estimates for CY 6- 8: We believe the company's earnings have returned to a growth track from 2026 onward and will show stronger growth in 2027, driven by the high-end ABF substrate market's reversion to undersupply, NYPCB's dominant supply share for networking chips, and its proactive pricing strategy, which puts it in a favorable position to benefit from rising AI investments.

We believe the overall ABF substrate market bottomed in 2H24, and growth should accelerate from 2H26 onward with the ramp of multiple next-generation, large body size ABF substrates for new server GPUs, CPUs, ASICs and networking chips. New technologies, such as embedded passives and Intel's EMIB-T, should also increase manufacturing complexity of the package substrates, adding to the overall undercapacity as usable output decreases.

Similar to our analysis of other tech hardware companies within our coverage, we use a residual income (RI) valuation model to value NYPCB - we think it derives the most accurate value of the firm given that it takes into account cost of e uity. Key assumptions (all unchanged):

• Cost of e uity of 9.3%;
• ° Risk-free rate of 1% (10-year Taiwan government note yield);
• ° E uity risk premium of 8.7%;
• ° Beta of 1.0;
• Medium-term growth rate of 17% - fair given our analysis of the substrate market over the next couple of years;
• Terminal growth rate of 3% (similar to the rest of the companies under our tech hardware coverage).

Our bull and bear case values increase by similar magnitudes, to NT$3,575 and NT $1,0 0, respectively.

Exhibit 27 NYPCB residual inc me (RI) m del

S urce: M rgan Stanley Research (E) estimates.

M Idea

Exhibit 28 NYPCB P/E chart

S urce: TEJ, M rgan Stanley Research estimates.

M Idea

Financia s

Exhibit 29 Quarterly Estimates

S urce: C mpany data, M rgan Stanley Research (E) estimates.

M Idea

Exhibit 30 C ns lidated Financial Summary

Consolidated Income Statement

Consolidated Cash Flow Statement

Consolidated Balance Sheet

2026E 2027E 2028E NT$mn (Year End Dec 31)

97,041 139,167 Operating Cashflow

(67,996) Net Profits

68,510 Working capital change

812 Other adjustments

600 Other adjustments

90.18 Other adjustments

Consolidated Financial Ratios

S urce: C mpany data, M rgan Stanley Research (E) estimates.

84,990 Margins (%)

117 Gross margin

33,091 Operating margin

27,298 Operating profits M Idea

Risk Reward - Unimicron (3037.TW) Risk Reward - Unimicron (3037.TW)

Price hikes and spec upgrades to support higher earnings estimates; OW

NT$1,775.00 PRICE TARGET

Base case, residual income (RI) valuation model, which we think derives the most accurate value of the firm given that it takes into account cost of equity. We use a cost of equity of 9.2% [risk-free rate of 1% (10-year Taiwan government note yield), equity risk premium of 8.7%, and a beta of 1.0], a medium-term growth rate of 14%, and a terminal growth rate of 3%.

NT$1,303.00

Source: Refinitiv, Morgan Stanley Research

BULL CASE

BASE CASE

OVERWEIGHT THESIS

• ▪We expect the ABF substrate market to revert to undersupply from 2027 onward, with the undersupply gap reaching 26% by 2030.
• ▪This is primarily driven by AI demand, with next-gen AI/networking chips adopting larger body size ABF substrates.
• ▪In the near to medium term, the short supply of T-glass will be a key watch point, but we believe this will mainly affect lowend ABF and BT substrates, as suppliers prioritize supply to the AI chip vendors.
• ▪Our price target implies 41x 2027e P/E or 24x 2028e P/E, above the 10-20x range at which it traded in the previous up-cycle. We deem this fair, given AI chip substrate spec upgrades and suppliers' ability to raise prices.

Source: Refinitiv, Morgan Stanley Research

Risk Reward Themes

Pricing Power:

Positive

Secular Growth:

Positive

View descriptions of Risk Rewards Themes here

BEAR CASE

70x 2027e P/E

Demand for AI servers, general servers and PCs is much stronger than expected: Pricing increases more than expected, 30%+ in 2026-27, driving stronger margin expansion for Unimicron.

41x 2027e P/E

Entering up-cycle driven by AI chip upgrades: Robust demand for AI GPUs and ASICs, supported by continued data center infrastructure investments by CSPs and neo cloud, leads to undersupply in the high-end ABF market from 2027 onward. PC demand is sluggish, while general servers still show double-digit unit growth in 2026. ASP has bottomed and started to inflect upward, driven by raw material price hikes and substrate spec upgrades.

14x 2027e P/E

Demand for AI servers and general servers softens, with PC demand weakening further: The result is greater ABF oversupply; thus, there are pricing declines of ~15-20% in 2026-27, leading to weaker margins for Unimicron over the next two years.

NT$595.00

NT$3,030.00

NT$1,775.00

M Idea

Risk Reward - Unimicron (3037.TW)

KEY EARNINGS INPUTS

INVESTMENT DRIVERS

• 5G smartphone demand
• 5G base station demand
• Automotive demand

GLOBAL REVENUE EXPOSURE

Source: Morgan Stanley Research Estimate View explanation of regional hierarchies here

MS ALPHA MODELS

4/5

MOST

3 Month

Horizon

Source: Refinitiv, FactSet, Morgan Stanley Research; 1 is the highest favored Quintile and 5 is the least favored Quintile

RISKS TO PT/RATING

RISKS TO UPSIDE

• Better-than-expected ABF substrate demand from PC and server customers
• Capex cuts; halt to its capacity expansion plan
• Continued yield issues of alternative technology that doesn't require substrate (e.g., CoWoP)

RISKS TO DOWNSIDE

• Sudden demand shortfall
• Technological change that would not require ABF substrates
• Intensifying competition
• Yield issues or production hiccups when ramping new capacity

OWNERSHIP POSITIONING

Source: Refinitiv, Morgan Stanley Research

MS ESTIMATES VS. CONSENSUS

Source: Refinitiv, Morgan Stanley Research M Idea

Risk Reward - Nan Ya PCB (8046.TW) Risk Reward - Nan Ya PCB (8046.TW)

Price hikes and spec upgrades to support higher earnings estimates; OW

NT$2,000.00 PRICE TARGET

Base case, residual income (RI) valuation model, which we think derives the most accurate value of the firm given that it takes into account cost of equity. We use a cost of equity of 9.3% [risk-free rate of 1% (10-year Taiwan government note yield), equity risk premium of 8.7%, and a beta of 1.0], a medium-term growth rate of 17%, and a terminal growth rate of 3%.

NT$1,755.00

Source: Refinitiv, Morgan Stanley Research

BULL CASE

BASE CASE

OVERWEIGHT THESIS

• ▪Price hikes for BT and ABF substrates, driven by raw material cost increases, will likely drive margin improvement for NYPCB.
• ▪We believe the ABF substrate market will become undersupplied again from 2027 onward, driven by AI demand, with next-gen AI/networking chips adopting large-body size ABF substrates.
• ▪NYPCB should benefit from ABF substrate structural tailwinds as well - rising data traffic, big data analysis, AI/ML, and 5G.
• ▪Our price target implies 41x 2027e P/E or 22x 2028e P/E, vs. the 10-30x range in which it traded in the previous up-cycle. We deem this fair given ABF substrate spec upgrades and suppliers' ability to raise prices.

Source: Refinitiv, Morgan Stanley Research

Risk Reward Themes

Pricing Power:

Secular Growth:

Positive Positive

View descriptions of Risk Rewards Themes here

BEAR CASE

74x 2027e P/E

BT prices increase further in 2H26, and ABF becomes undersupplied faster than expected (current expectation is from 2027 onward), driving even stronger margin expansion for Nan Ya PCB.

41x 2027e P/E

Entering up-cycle driven by AI chip upgrades: Robust demand for AI GPUs/ASICs and networking chips, supported by continued data center infrastructure investments by CSPs and neo cloud, leads to undersupply in the high-end ABF market from 2027 onward. PC demand is sluggish, while general servers still show double-digit unit growth in 2026. ASP has bottomed and started to inflect upward, driven by raw material price hikes and substrate spec upgrades.

21x 2027e P/E

ABF demand weakens further, resulting in greater oversupply in the next few years. Thus, ABF pricing decreases 20% or more in 2026-27, leading to weaker margins for Nan Ya PCB.

NT$1,020.00

NT$3,575.00

NT$2,000.00

M Idea

Risk Reward - Nan Ya PCB (8046.TW)

KEY EARNINGS INPUTS

INVESTMENT DRIVERS

• 5G smartphone demand
• 5G base station demand
• Automotive demand

GLOBAL REVENUE EXPOSURE

Source: Morgan Stanley Research Estimate View explanation of regional hierarchies here

MS ALPHA MODELS

Source: Refinitiv, FactSet, Morgan Stanley Research; 1 is the highest favored Quintile and 5 is the least favored Quintile

RISKS TO PT/RATING

RISKS TO UPSIDE

• Better-than-expected ABF and BT demand
• Faster-than-expected pickup in AI and 5G demand
• Stronger-than-expected ASP hike
• Continued yield issues of alternative technology that doesn't require substrate (e.g. CoWoP)

RISKS TO DOWNSIDE

• Sudden demand shortfall, creating headwinds for ABF substrate demand and pricing
• Technological change that would not require ABF substrates
• Intensifying competition

OWNERSHIP POSITIONING

Inst. Owners, % Active

78.1%

Source: Refinitiv, Morgan Stanley Research

MS ESTIMATES VS. CONSENSUS

M Idea

Risk Reward Reference inks

1. View explanation of Options Probabilities methodology -

Options_Probabilities_Exhibit_Link.pdf

1. View descriptions of Risk Rewards Themes - RR_Themes_Exhibit_Link.pdf

2. View explanation of regional hierarchies - GEG_Exhibit_Link.pdf

3. View explanation of Theme/Exposure methodology -

ESG_Sustainable_Solutions_External_Link.pdf

1. View explanation of HERS methodology - ESG_HERS_External_Link.pdf

M Idea

INDUSTRY COVERAGE: Greater China Technology Hardware

M Idea

St ck Ratings are subject t change. Please see latest research f r each c mpany.

• Hist rical prices are n t split adjusted.

© 2026 Morgan Stanley