Taiwan Surface Mounting Technology Corp. (6278.TW)

Deep Dive Research Report

Prepared 2026-06-09. All figures in New Taiwan Dollars (NT$) unless stated. No valuation, price targets, or recommendations are contained in this report.

1. What the Company Does

Taiwan Surface Mounting Technology Corp (TSMT, 台表科) is a contract manufacturer. It does not own the products it builds and it does not sell to consumers. It runs the factory floor where bare printed circuit boards (PCBs) and loose electronic components are turned into finished, tested circuit assemblies that go inside someone else's product. The specific process it has spent 35 years industrialising is Surface Mount Technology (SMT): the high-speed, high-precision placement of chips, resistors, capacitors and connectors onto a board, followed by reflow soldering, inspection and functional test. Its own description is plain: "SMT specialist, world class EMS provider."

The company was incorporated in March 1990 and is headquartered in Taoyuan City, Taiwan. It moved from Taiwan's over-the-counter market to a full TWSE listing in August 2010. For most of its life TSMT was, in effect, the outsourced assembly arm of Taiwan's flat-panel-display industry. When AU Optronics, Innolux or a Chinese panel maker needed the control board, the backlight driver board or the touch-module board that sits behind an LCD screen, TSMT mounted the components onto it. It became the dominant SMT house for TFT-LCD control boards, holding a reported share above 70% of that niche. That is the company's origin and still its single largest line of business.

The interesting part of TSMT today is the deliberate migration away from that origin. Panel-related work is mature, cyclical and structurally flat-to-declining, and management has spent the last several years redeploying the same SMT capability into higher-value end markets: automotive electronics, DRAM/memory modules, Mini LED backlight units, networking gear and cable/wiring harness modules. The pitch to investors in 2025-2026 is a mix-shift story - the same factories, increasingly pointed at memory modules and cars instead of television control boards.

The core value proposition is straightforward and unglamorous. A brand or an ODM that designs an electronic product does not want to own, staff and depreciate SMT lines that must run at high utilisation to be economic. TSMT does. It buys and manages the components, engineers the assembly process, runs the placement and soldering, tests the output, handles logistics and provides after-sales support. The customer gets a qualified, repeatable assembly at a known yield without the capital burden. What makes this hard to replicate is not any single machine - SMT equipment is purchasable - but the accumulated process engineering, the qualification record with demanding customers like Apple, the breadth of qualified lines across multiple countries, and the scale needed to run those lines profitably on thin margins.

A concrete example: a panel maker wins an order for high-end notebook displays using Mini LED backlighting. Mini LED packs thousands of tiny LED dies onto a backlight board with tight placement tolerances. The panel maker ships TSMT the boards and the bill of materials; TSMT procures the components, sets up the placement program, mounts and solders the LEDs and driver ICs, runs optical and electrical test, and ships qualified backlight units back into the panel maker's module line. TSMT never sees the end customer, never brands the notebook, and earns a thin assembly margin on enormous volume.

2. Business Segments

TSMT reports as a single SMT-assembly business but discloses a product-line revenue mix that functions as its de facto segmentation. The nine-month-2025 mix was: TFT-LCD 46%, DRAM modules 14%, Mini LED 12%, automotive 10%, cable/wiring 9%, touch pad 4%, other 5%. The most useful way to understand the company is through these product lines, because each has a different end market, a different margin profile and a different growth trajectory.

TFT-LCD assembly (the legacy core, ~46%)

This is what TSMT was built on. It assembles the control boards (the bulk of the line), backlight bars and touch-module boards that sit behind flat-panel displays - for televisions, monitors, notebooks and tablets. The core capability is volume SMT at extreme cost discipline plus deep co-qualification with the world's panel makers; TSMT's reported share of TFT-LCD control-board SMT is above 70%, a position built over decades that a new entrant cannot buy. It exists as the company's foundation because the panel industry was Taiwan's signature export industry and TSMT grew up beside it. Competitively it is strong - near-dominant in its niche - but the niche itself is the problem: panel demand is mature and cyclical, and the segment entered a correction from Q2 2025 as consumer-electronics demand softened. Management is explicit that this is not the growth engine. It is the cash-generative, high-share base that funds the pivot.

DRAM / memory modules (the re-rating engine, ~14% rising toward 20%+)

This is the segment doing the heavy lifting in the current investment narrative. TSMT performs the SMT assembly of DRAM memory modules - mounting memory chips and supporting components onto module boards, increasingly DDR5 generation. The core capability is the precision and yield demanded by memory-module assembly plus the qualification to run it at server-grade reliability. It exists as a distinct line because the economics are unusually favourable right now: module revenue tracks memory pricing, and the memory cycle turned sharply up. Industry forecasters cited by management expected DRAM contract prices up roughly 90-95% into Q1 2026 and NAND up 70-75% into Q2 2026, which lifts both the average selling price and the margin of the assembled module. Management's stated target is for DRAM modules to exceed 20% of total revenue in 2026, up from 14% in 2025 and around 7% in 2024. Within the group this is the growth bet and the margin lever, and it is the line management most wants investors to value.

Mini LED (the premium display option, ~12%)

Mini LED backlight assembly serves high-end notebooks and tablets, and this is where TSMT's Apple relationship sits. The capability is placing thousands of miniature LED dies on a backlight board at high yield - materially harder than conventional backlight assembly. It exists as its own line because it is a premium, faster-growing slice of the display market with better economics than commodity TFT-LCD work, and because the customer set (premium device makers) is distinct. Competitively TSMT is well positioned through its installed qualification with Apple's supply chain. Strategically it is the "good" part of the display franchise - display-adjacent but premium and growing - a hedge against the decline of commodity panel work.

Automotive electronics (the second growth leg, ~10%)

TSMT assembles automotive modules: camera modules, battery-management-system (BMS) boards, fingerprint/biometric sensors, and increasingly sensing and electromechanical-control modules. The capability that matters here is automotive-grade quality discipline - the certification, traceability and zero-defect orientation that car supply chains demand, which is a different and slower qualification than consumer work. It exists as a separate line because the automotive customer base, qualification regime and product longevity are entirely different from displays. This business grew from roughly 10% of "other" in 2021 to about 30% of that bucket, then declined about 15% year-on-year in 2025 on weak auto-market conditions. Management frames it as the second growth leg alongside DRAM and expects it to "return to a growth trajectory" from 2026 as new sensing and control modules ramp.

Cable / wiring modules and touch pads (~9% and ~4%)

The cable/wiring line assembles wiring and interconnect modules; touch pads serve notebook and peripheral makers. These are smaller, more commoditised lines that round out factory loading and customer relationships rather than drive the story.

3. Products and Business Detail

The product catalogue maps onto the segments above but is worth naming concretely. On the display side TSMT builds TFT-LCD control boards (the timing-controller and driver electronics behind a panel), LED backlight bars and driver boards, LCD TV inverter and power boards, and touch-module substrates. It assembles Mini LED backlight units for premium notebooks and tablets. In memory it assembles DRAM/DDR modules. In automotive it builds camera modules, battery-management-system boards, fingerprint and biometric sensor modules, and dashboard/control electronics. It also does FPCB chip-on-film (COF) related assembly for displays, cable/wiring modules, touch pads, and networking and set-top-box assembly out of its India operations. Across all of these the actual product is the same in kind - a populated, soldered, tested circuit assembly - and the differentiation is which end market it is qualified for and at what reliability grade.

What makes the work hard is less the placement machine and more the surrounding discipline: process engineering to hit yield on a new board design, materials procurement and inventory management across volatile component markets, optical and electrical inspection, and customer-specific qualification. Apple-grade and automotive-grade work in particular require quality systems, traceability and audit records that take years to build a track record in. TSMT operates an ISO/IEC 17025 CNAS-accredited laboratory to support that testing and qualification.

The manufacturing footprint is the most strategically active part of the business. TSMT's production base is heavily in China - plants in Chongqing, Xianyang, Hefei, Suzhou (the Junling/峻凌 Electronics subsidiary), Dongguan and Xiamen - which historically accounted for around 84% of output. Over the last few years it has deliberately built capacity outside China to serve customer "production dispersion" and tariff-driven nearshoring:

• India (Chennai + Gujarat): The Chennai phase-2 expansion was completed in Q4 2025 with move-in expected Q1 2026. A new greenfield plant on purchased land in the Sanand industrial zone, Gujarat, is targeted for completion in Q4 2026. India serves networking, set-top-box and smartphone-assembly customers.
• Vietnam: A phase-2 expansion is planned to capture notebook and related supply-chain shifts out of China.
• Mexico: A nearshoring base for North American customers, primarily automotive and networking; management has flagged the customer set as still unstable due to tariff-policy uncertainty, with major-customer validation expected to complete around Q1 2026.

Total SMT line count has scaled from roughly 220-230 lines (2023) upward through the expansion program, and 2026 capital expenditure was guided at roughly US$80-100 million, directed at India, Vietnam and Mexico while China shifts to equipment upgrades rather than expansion.

Milestones that changed the business: the TWSE listing in 2010; the build-out of automotive electronics from a rounding error in 2021 to a double-digit revenue line; the entry into and scaling of DRAM module assembly (from ~7% in 2024 toward a 20%+ target in 2026); the Mini LED ramp tied to premium notebooks; and the multi-country capacity diversification (Vietnam/Mexico from 2024, India phases through 2026).

4. Customers

TSMT's customers are not consumers; they are brands, ODMs, panel makers and memory-module makers who outsource assembly. The disclosed Q3 2025 customer mix was: American clients (Apple) about 34%, China panel makers about 16%, Taiwan panel makers about 15%, and "others" (primarily DRAM module customers) about 34%. Named accounts that appear in the company's and analysts' descriptions include Apple (Mini LED and premium display assembly), Taiwan panel makers AU Optronics and Innolux, and Chinese panel/appliance makers such as Haier and Hisense.

The buying decision inside these customers sits with sourcing and operations - supply-chain and manufacturing-engineering functions that qualify a contract manufacturer's lines for a specific product, then place volume. The criteria are yield, on-time delivery, cost, geographic footprint (can you build it where I need it for tariff reasons) and quality record. Sales cycles are long for new programs because they require qualification - for automotive and Apple work especially, a new line or new product must pass audit and reliability testing before it sees volume, which can take months to over a year.

Why customers choose TSMT comes down to qualified capacity at scale and cost. In TFT-LCD control boards TSMT's >70% share means a panel maker's path of least resistance is to use the incumbent that already runs its boards at known yield. In Mini LED and automotive the draw is a track record of passing demanding qualification. In DRAM modules it is the ability to ramp module assembly volume quickly into a hot memory cycle. And increasingly the draw is footprint: a customer trying to move production out of China for tariff reasons can use TSMT's India, Vietnam or Mexico lines rather than re-qualifying a new vendor.

Switching costs are real but asymmetric. Once a board is qualified on TSMT's line, moving it to a competitor means re-qualifying - re-running the audits, reliability tests and yield ramp - which is costly and risky, especially for automotive and Apple-grade work. That installed-base lock-in protects the legacy TFT-LCD position. But for commodity work the switching cost is lower, and a large customer with multiple qualified vendors can shift volume on price.

Concentration is a genuine feature. Apple at roughly a third of revenue is the single defining relationship; the top customer groups (Apple plus Taiwan and China panel makers) account for the majority of revenue. This is both a quality signal - Apple does not qualify weak suppliers - and a risk: a program loss, an Apple design change away from a TSMT-built component, or a cyclical air-pocket at the panel makers hits revenue hard.

Contract structure is predominantly program-based and order-driven rather than long-dated take-or-pay. Revenue is therefore exposed to the volume swings of the underlying end products and to memory pricing on the DRAM line, which is why management's quarterly guidance moves with consumer-electronics demand and component prices rather than tracking a contracted backlog.

5. Competitive Landscape

TSMT sits in the EMS (electronics manufacturing services) layer of the supply chain, the same broad category as Taiwan's giant contract manufacturers but at a fraction of their size and with a narrower SMT/module focus. Competition has to be split by where it competes.

In its core TFT-LCD control-board niche TSMT has few genuine peers of scale - its >70% share reflects decades of co-qualification with panel makers and the unattractive economics of a new entrant trying to win a mature, declining, low-margin business. This is the closest thing to a moat the company has, but it is a moat around a shrinking pond.

In the broader EMS/SMT arena into which TSMT is diversifying - automotive modules, networking, memory modules, premium display assembly - it competes against far larger generalist EMS/ODM players. The relevant named competitors are Taiwan's Hon Hai/Foxconn, Pegatron, Wistron and USI (Universal Scientific Industrial / 環旭電子), and global EMS firms Jabil and Flex, plus private specialists like Zollner in automotive. TSMT's advantage against these giants is focus and flexibility on SMT and module assembly plus its panel-industry heritage; its disadvantage is scale - in any business where the customer wants full box-build, system integration or the lowest possible cost at vast volume, the giants win.

In DRAM module assembly the competitive set is memory-module makers and their assembly partners; TSMT competes on assembly yield and the ability to scale into the cycle, not on owning memory.

Barriers to entry are moderate and segment-specific. The equipment is buyable; what is not buyable quickly is the qualification record (Apple-grade, automotive-grade), the multi-country qualified footprint, and the scale needed to run thin-margin lines economically. For the legacy panel niche the barrier is high (incumbency + re-qualification cost). For commodity assembly the barrier is low and competition is intense, which shows up in the thin gross margins (around 12-13%) and net margins (around 5%). This is not a high-moat business in aggregate; it is a scale-and-qualification business with one protected legacy niche and several competitive growth fronts.

The structural shift in the landscape is the China+1/nearshoring wave. Tariffs and supply-chain de-risking are pushing customers to dual-source outside China, and TSMT's India/Vietnam/Mexico build-out is a direct play to win that re-shored volume against competitors making the same bet.

Market caps are approximate peer-size references as of June 2026 and move with the market; they are not valuation inputs for the subject company.

6. Industry

TSMT operates in electronics contract manufacturing, specifically the SMT-assembly layer that converts components and bare boards into finished circuit assemblies. Demand for this work is derived demand: it rises and falls with the production volumes of the end products its customers sell - televisions, notebooks, tablets, cars, servers and networking gear - and with the component-price cycle, most visibly memory.

The demand drivers are therefore several distinct cycles bolted together. Display/consumer electronics is a mature, cyclical market that entered a correction in 2025 as consumer demand softened. Automotive electronics content per vehicle is on a long secular rise (cameras, sensors, BMS, control modules) even as unit volumes wobble with the auto cycle. Memory is the most violent cycle of all - the 2025-2026 DRAM/NAND upcycle, with contract prices forecast up 90-95% (DRAM, into Q1 2026) and 70-75% (NAND, into Q2 2026), directly lifts the revenue and margin of TSMT's module assembly because module ASP tracks chip prices. Networking and server demand is being pulled by data-centre and connectivity buildout.

The EMS industry globally is large - hundreds of billions of dollars of outsourced manufacturing - and structurally favoured by the OEM preference to avoid owning low-margin assembly capacity. TSMT sits low in the value chain: it earns an assembly margin between the component makers (which capture more value) and the brand OEMs (which capture the most). That position caps profitability - thin margins are the industry norm - but provides volume and capital-light-relative-to-brand exposure to whatever end markets it serves.

The defining current industry dynamic is geographic re-shoring. For two decades EMS concentrated in China; tariffs, US-China tension and customer de-risking are now pushing assembly into India, Vietnam, Mexico and elsewhere. TSMT, with ~84% of capacity historically in China, is both exposed to this shift and investing to ride it. Regulation here is less about product approval (though automotive and Apple qualification act like de facto certifications) and more about trade policy - tariffs are explicitly cited by management as destabilising the Mexico customer base.

Cyclically the business is exposed on multiple fronts at once: a display downturn, an auto downturn and a memory downturn would compound; conversely a memory upcycle (as in 2026) can offset display weakness. That is precisely the balancing act visible in the numbers - flat-to-modest top-line growth as DRAM strength offsets consumer/panel weakness.

7. Growth Triggers

Extracted from management commentary across the relevant reporting periods (TSMT's broker-hosted investor conferences and results disclosures). Forward-looking items only.

• DRAM modules to exceed 20% of total revenue in 2026, up from 14% in 2025. Repeated across the Q3 2025 conference (Dec 18, 2025) and the Q1 2026 conference (May 8, 2026).

  Management stated the 2026 target is for "DRAM modules to comprise over 20% of total revenue," driven by DDR5 penetration and the memory upcycle. (Q3 2025 conference, Dec 18, 2025)

• Memory-price upcycle as an ASP/margin tailwind. Management cited industry forecasts of DRAM contract prices up ~90-95% into Q1 2026 and NAND up ~70-75% into Q2 2026, supporting both module revenue and average selling price. (Q1 2026 conference, May 8, 2026)

• Automotive electronics returning to a growth trajectory in 2026 via new sensing and electromechanical-control modules, after a ~15% YoY decline in 2025. (Q3 2025 conference, Dec 18, 2025; reiterated Q1 2026 conference, May 8, 2026)

• India - Chennai phase-2 completed Q4 2025, move-in Q1 2026, expanding capacity for networking/STB/smartphone work. (Q3 2025 conference, Dec 18, 2025)

• India - new Gujarat (Sanand) greenfield plant targeted for completion Q4 2026. (Q3 2025 conference, Dec 18, 2025)

• Vietnam phase-2 expansion to capture notebook and related supply-chain relocation out of China. (Q3 2025 conference, Dec 18, 2025)

• Mexico major-customer validation expected to complete around Q1 2026, enabling a nearshore base for North American automotive/networking customers (currently small-scale, tariff-sensitive). (Q3 2025 conference, Dec 18, 2025)

• 2026 capital expenditure of roughly US$80-100 million directed at the India/Vietnam/Mexico expansions. (Q3 2025 conference, Dec 18, 2025)

8. Key Risks

Memory-cycle dependence cuts both ways. The DRAM module line is the re-rating story, but its revenue and margin track memory prices, which are the most volatile in electronics. The 2026 upcycle (DRAM +90-95%, NAND +70-75%) is a tailwind today; a roll-over in memory pricing would deflate ASP-driven revenue and compress the margin gain just as quickly, and the segment is being scaled to 20%+ of the company precisely as prices peak. This is a high-probability moderate-to-significant drag risk on any memory downturn.

The largest segment is in structural decline. TFT-LCD assembly is ~46% of revenue and management openly says it is not the future. Panel demand is mature, entered a correction from Q2 2025, and faces long-run pricing and volume pressure.

Management noted that "from Q2 2025 the panel industry entered a correction period, and related demand was relatively weak." (Q3 2025 conference, Dec 18, 2025) The risk is that legacy decline outpaces the ramp of DRAM and automotive, leaving the company running to stand still - which is essentially what the "flat 2026" guidance implies.

Customer concentration around Apple and the panel makers. With Apple-linked work near 34% of revenue and panel makers another ~31%, a lost program, an Apple design change away from a TSMT-built component, or a cyclical air-pocket at AUO/Innolux/Chinese panel makers would hit revenue disproportionately. High impact, moderate probability in any given year.

China manufacturing concentration into a tariff world. Roughly 84% of capacity has been in China while the customer base is increasingly demanding non-China supply. The India/Vietnam/Mexico build is the mitigation, but until it scales the company is exposed to tariffs and to losing re-shored volume to faster competitors. Management has explicitly flagged the Mexico customer base as "unstable" due to tariff policy.

Capex into a flat year. TSMT is committing US$80-100m to new plants in 2026 while guiding revenue roughly flat with 2025. If the re-shored volume or the DRAM/automotive ramp disappoints, the company adds depreciation and underutilised lines against a stagnant top line - a classic EMS margin trap. Moderate probability, moderate impact.

Thin-margin commodity economics. Gross margin sits around 12-13% and net margin around 5%. There is little cushion; a modest cost shock, FX move (NTD strength against the USD/RMB it earns and spends in), or price concession from a large customer flows quickly to the bottom line. The market also noted the dividend is not comfortably covered by free cash flow during the heavy-capex phase, a constraint on capital returns.

9. Walk the Talk

The five reporting periods used for this assessment are: Q1 2025 results (commentary around the FY2024/early-2025 conference cycle, ~March-May 2025), Q2 2025 results (mid-2025), Q3/9M 2025 (broker-hosted online conference, Dec 18, 2025), FY2025 annual results (board approval and disclosure March 10, 2026), and Q1 2026 (online investor conference with Chinese and English sessions, May 8, 2026). A note on availability: TSMT is a Taiwanese mid-cap that holds broker-invited investor conferences rather than a fixed quarterly English webcast, so the depth of public transcript varies by period. The Dec 18, 2025 (Q3) and May 8, 2026 (Q1 2026) conferences are the best-documented; the others are reconstructed from results disclosures and analyst memos. Where I lack a verbatim quote I say so rather than invent one.

The through-line is a management team that has been consistent and roughly accurate on the big strategic claims, and appropriately conservative on near-term guidance. Early in 2025 management framed the year around a recovery and the rising contribution of DRAM and automotive, guiding for full-year revenue growth approaching 10%. That number was essentially delivered: 2025 revenue came in at NT$49.9 billion, up 10.13%, the company's return to growth after two declining years. Hitting a ~10% growth target almost exactly is a point in management's favour on accuracy.

On the DRAM mix-shift, management has been consistent and so far credible. The DRAM line moved from roughly 7% of revenue (2024) to 14% (9M 2025), and the stated 20%+ target for 2026 was repeated at both the December 2025 and May 2026 conferences rather than quietly raised or dropped. That is the behaviour of a team tracking a real ramp, not floating a slogan.

"DRAM modules and automotive-related products" were named as the two growth engines, with DRAM "targeted to exceed 20% of total revenue" in 2026. (Q3 2025 conference, Dec 18, 2025; reiterated May 8, 2026)

On near-term guidance the team has been conservative and honest about weakness. Into Q4 2025 management guided revenue down 6-7% quarter-on-quarter on soft consumer electronics, and was candid that the panel business had been in correction since Q2 2025. For 2026 it set a deliberately low bar - "relatively conservative," roughly flat with 2025 - rather than promising that DRAM and automotive would drive a step-change. The Q1 2026 result (cumulative revenue +1.13% YoY, EPS NT$2.17, down both sequentially and year-on-year) is consistent with that conservative framing; management did not over-promise and then miss, it set expectations low and delivered in line.

The area to watch is the capacity build. Management has repeatedly guided to new plants in India (Chennai phase-2 Q4 2025/move-in Q1 2026; Gujarat Q4 2026), Vietnam phase-2, and Mexico (validation Q1 2026). An earlier (2023) version of this story guided Vietnam and Mexico to mass production "by Q1 2024," so the multi-country expansion has been a multi-year, repeatedly-extended program. The cadence of completion dates slipping outward across conferences is the one place where management's timelines have proven optimistic, and it is the right thing to hold the next several conferences accountable against.

Net read: this is a management team that does broadly what it says on strategy and is conservative-to-accurate on financial guidance, with a credibility soft spot on capacity-expansion timelines, which have repeatedly slid.

10. Shareholder Friendliness Index

Dividends. TSMT has paid a cash dividend for 27 consecutive years, an unusually long unbroken record. Over the last three distribution years the cash dividend per share was NT$7.0 (paid 2023), NT$5.0 (paid 2024) and NT$5.5 (paid 2025), against a five-year average of about NT$5.4. The trend is therefore down-then-stabilising rather than steadily growing: the NT$7.0 in 2023 reflected a peak-profit year and was not sustained, with the payout resetting to the NT$5.0-5.5 range. The payout ratio runs around 58% of earnings, a moderate, sustainable level, though analysts have flagged that the dividend is not comfortably covered by free cash flow during the current heavy-capex phase. With FY2025 EPS of NT$9.39 and the FY2025 dividend to be set at the March 2026 board, the next dividend will reveal whether management lifts the payout on the stronger year or holds it to fund the India/Vietnam/Mexico build.

Buybacks and dilution. TSMT runs recurring buybacks, but they are employee-incentive buybacks (shares repurchased to transfer to employees), not capital-return buybacks that retire stock. The most recent completed round expired Feb 26, 2026, with 32.68% of the authorised amount executed at an average price of NT$96.04; a fresh authorisation to buy up to 10,000 lots (about 10 million shares) in a NT$66.5-148 range began in January 2026, again earmarked for employee transfer. Because these shares are recycled to employees rather than cancelled, the buybacks offset dilution rather than shrink the share count; the share base (capital of about NT$2.92 billion, roughly 292 million shares) has been broadly stable, with no large issuance and no net retirement. (Note: the buyback figures above are sourced from exchange filings and financial news across 2025-2026; no MoatMap database block was supplied, so the three-year picture is drawn entirely from public filings.)

Verdict: Returns Capital, modestly - a 27-year dividend streak with a ~58% payout makes it a consistent dividend payer, but the buybacks fund employee stock rather than retire shares, so capital return is via dividends only and is being balanced against a heavy capacity-expansion phase.

11. Insider Activities

The listing venue is the Taiwan Stock Exchange, where insider holdings and transfers are disclosed through the Market Observation Post System (MOPS, 公開資訊觀測站) under "Director/Supervisor Equity Holdings" (董監事持股) and insider share-transfer pre-declarations (內部人持股轉讓申報). MOPS is a form-POST-driven portal that is not directly fetchable, and aggregator searches returned no specific individual director or officer open-market transactions for 6278 over the last 12 months. Individual insider transaction detail for TSMT could not be located in primary or aggregator form within the search budget; the following reflects what is verifiable from filings and disclosures.

What is verifiable:

• Aggregate insider ownership is high and stable. Directors, supervisors and major shareholders collectively hold about 33.88% of TSMT. A founder/insider block of roughly a third of the company is a meaningful alignment signal and there is no evidence in the search of a large block being sold down.

• Company buyback (corporate, not individual insider). TSMT completed a treasury-share buyback that expired Feb 26, 2026 (32.68% executed, average NT$96.04) and authorised a further repurchase of up to ~10 million shares (NT$66.5-148) from January 2026, both earmarked for transfer to employees rather than cancellation. This is a corporate action to fund employee incentives, not an open-market purchase by an individual insider, so it is not a conviction signal in the way a director's personal open-market buy would be - but the company choosing to repurchase in the NT$66-148 band, with an average fill near NT$96, does indicate the board viewed those levels as acceptable for accumulating its own shares.

No individual open-market buys or sells by named directors or officers were identified in the period, and therefore none can be characterised as bullish cluster-buying or as concerning insider selling.

Net assessment: With high, stable aggregate insider ownership (~33.88%) and a company-level employee buyback executed in the mid-NT$90s, the verifiable signal is neutral-to-mildly-constructive - there is no evidence of insider distribution, but the absence of locatable individual open-market purchases means there is no strong conviction buy signal either. A reader who wants transaction-level granularity should pull the 6278 董監事持股 and 內部人持股轉讓 records directly from MOPS, which were not machine-accessible for this report.

12. Scenarios

Bull case. The memory upcycle runs hot through 2026 and TSMT's DRAM module line clears the 20%+ revenue target, with rising chip prices lifting both volume and margin so that the segment becomes a genuine second pillar rather than a cyclical kicker. Automotive electronics turns the corner as the new sensing and electromechanical-control modules ramp, restoring it as the third leg. The India plants - Chennai phase-2 now running, Gujarat completing on schedule in Q4 2026 - and the Vietnam and Mexico bases come on line just as customers accelerate their move out of China, so TSMT captures re-shored networking, notebook and automotive volume that competitors cannot serve as quickly. The mix shift away from commodity TFT-LCD lifts blended gross margin off the 12-13% floor, the heavy 2026 capex starts to be absorbed at high utilisation, and a management team that has hit its strategic targets earns the benefit of the doubt. The story becomes "a memory-and-automotive module maker with a stable display cash base," not "a panel assembler in decline."

Base case. The most likely path is the one management has already drawn: 2026 lands roughly flat with 2025. DRAM and automotive grow and lift the mix, but they grow into the teeth of a softening consumer-electronics and panel market, so the two roughly offset and the top line is stable rather than expanding. The memory tailwind helps margins in the first half but is not assumed to persist indefinitely. The capacity build proceeds, with the usual modest slippage in plant timelines, and the new geographies contribute gradually rather than stepping up. The dividend holds in the NT$5-5.5 range, the employee buyback continues, and the company remains a consistent, thin-margin EMS operator slowly improving its end-market mix. Steady, unspectacular, and broadly as guided.

Bear case. The memory cycle rolls over faster than expected, deflating the DRAM module ASP just as TSMT scales the line to 20%+ of revenue, so the segment goes from margin tailwind to drag. At the same time the structural decline in TFT-LCD - still 46% of revenue - accelerates, and a cyclical air-pocket or a program/design loss at Apple or the panel makers (where over 60% of revenue is concentrated) takes a chunk out of the base. The US$80-100m of 2026 capex lands as depreciation against underutilised India/Vietnam/Mexico lines because re-shored volume is slower to materialise or is won by larger competitors, and the Mexico customer base stays unstable under shifting tariff policy. Thin margins, no cushion, and a dividend already stretched relative to free cash flow leave management choosing between cutting the payout and pausing the expansion. The mix-shift story stalls and TSMT looks like a sub-scale EMS caught between a declining legacy and growth bets that arrived a cycle too late.

A note on sourcing and limitations

• Five reporting periods used: Q1 2025, Q2 2025, Q3/9M 2025 (conference Dec 18, 2025), FY2025 (board March 10, 2026), Q1 2026 (conference May 8, 2026). The Dec 2025 and May 2026 conferences are the best-documented; TSMT holds broker-invited conferences rather than fixed quarterly English webcasts, so transcript depth varies and is disclosed as such in Section 9.
• Section 13 (Further Reading) is intentionally omitted: searches of SemiAnalysis, Stratechery and MBI Deep Dives returned no coverage of this company.
• Section 11 flags that individual MOPS insider-transaction records were not machine-accessible within the search budget.
• Some Chinese-source revenue figures contained decimal-placement inconsistencies; segment percentages, dividends, EPS and guidance have been cross-checked across multiple sources where possible. The FY2023/FY2024 revenue line in the chart is approximate.

Sources:

• TSMT corporate site (English)
• TSMT 2024 Annual Report (PDF)
• Fugle - TSMT Q3 2025 investor conference memo (Dec 18, 2025)
• cnyes - TSMT 2025 revenue NT$49.9bn +10%, buyback in progress
• MoneyWeekly - TSMT DRAM module & automotive margin feature
• nStock - TSMT company profile
• 913 - TSMT segment/customer breakdown and catalysts
• MoneyDJ - TSMT Vietnam/Mexico/India capacity expansion
• Wantgoo - TSMT dividend history
• cmoney - TSMT corporate calendar (conferences, AGM)
• Simply Wall St - TSMT overview, dividends, ownership
• StatementDog - TSMT EPS/financials