United Microelectronics Corporation (UMC) - Deep Dive Research Report

Prepared 2026-06-03. Ticker: TWSE 2303 / NYSE: UMC. Sector: Technology (Semiconductor Foundry).

Concalls used (most recent four): Q1 2026 (Apr 29 2026), Q4 2025 (Jan 28 2026), Q3 2025 (Oct 29 2025), Q2 2025 (Jul 30 2025). The most recent is 35 days old, within the 90-day window.

1. What the company does

UMC makes chips that other companies design but do not manufacture. It is a pure-play semiconductor foundry: a fabless customer (say MediaTek or Qualcomm) hands UMC a circuit design, and UMC turns that design into physical silicon wafers, etching billions of transistors into discs of silicon inside enormous, ultra-clean factories called fabs. UMC owns no chip brands of its own. Its product is manufacturing capacity, process recipes, and yield - the percentage of good chips that come off a wafer.

Critically, UMC does not chase the bleeding edge. It deliberately stopped developing transistors smaller than 14 nanometers back in 2018, ceding the race for the most advanced logic (5nm, 3nm, 2nm) to TSMC and Samsung. Instead UMC concentrates on "mature" and "specialty" nodes - 22nm, 28nm, 40nm, 65nm and older - the workhorse processes that go into the chips surrounding the flashy AI processor: WiFi and Bluetooth radios, display driver ICs, power management chips, image signal processors, microcontrollers, NAND flash controllers, and the analog/mixed-signal parts that the world buys in colossal volume. Roughly half of UMC's revenue now comes from "specialty" technologies - processes tuned for a specific physical job (high voltage, radio frequency, embedded memory) rather than raw transistor density.

The founding story matters because it explains the strategy. UMC was Taiwan's first semiconductor company, spun out of the government's Industrial Technology Research Institute (ITRI) on May 22, 1980, to localize chip production. It began life as an Integrated Device Manufacturer (IDM) - designing and selling its own memory, logic, and even x86 clone microprocessors. In 1995 it made the pivotal decision to abandon its own products and become a pure-play foundry, and in 1996 it spun off its design teams into what became MediaTek, Novatek, Faraday, and others - companies that are now its customers. Between 1996 and 2000 UMC ran a series of joint-venture fabs with fabless partners, then merged five of them back into one entity in 2000 and listed on the NYSE the same year. For a stretch in the 2000s UMC was a genuine number-two to TSMC and competed at the leading edge. It lost that race. The 2018 decision to stop at 14nm was an admission that it could not out-spend TSMC, and a bet that there is a large, durable, profitable business in everything that is not the leading edge.

The core value proposition, in management's framing across every recent call: be the differentiated, geographically diversified alternative to both TSMC's premium and China's commodity capacity. As President Jason Wang put it on the Q2 2025 call, the goal is to grow "the percentage of revenue contribution in this space" - specialty - while "the percentage of the revenue contribution competing with the Chinese foundries will continue to decline."

A concrete example. A fabless company designs a WiFi 7 radio chip. It needs that radio built on a process that handles high-frequency signals cleanly with low power leakage - UMC's RF-SOI (radio-frequency silicon-on-insulator) specialty platform. The customer and UMC spend months qualifying the design: test wafers, yield tuning, reliability testing. Once "taped out" and qualified, the customer is effectively locked in - re-qualifying the same chip at a different foundry costs a year and millions of dollars. UMC then runs that design in volume, shipping tens of thousands of 12-inch wafers a month, each wafer diced into thousands of radio chips that end up in phones, routers, and laptops. UMC gets paid per wafer, not per chip, so its economics turn on how full its fabs are (utilization) and what it can charge per wafer (ASP).

2. Business segments

UMC reports as a single operating segment - it is one foundry business - so there are no divisional financials to split. But management consistently frames the business along two axes that function like segments: technology platform (mature logic vs. specialty) and end-market application. The most useful way to understand UMC is through these lenses plus its geographic manufacturing footprint, which has become a strategic axis in its own right.

2.1 Mature & advanced logic (the 22/28nm engine)

What it does. Standard digital logic manufacturing on UMC's finest nodes - 28nm and its derivative 22nm - down through 40nm and 65nm. This is the growth core. The 22/28nm bucket has been the single most-discussed number on every recent call. In Q2 2025 it hit a record 40% of total revenue; by Q1 2026, 22nm alone reached a record 14% of revenue and the 28/22 bucket sat around 34%. The applications are display driver ICs, networking chips, microcontrollers, ISPs, and increasingly AI-adjacent silicon.

Core capability. 22nm is UMC's crown jewel because it is the most advanced planar (non-FinFET) logic node still in volume production, and UMC has tuned it for low power and low leakage. Management said on the Q1 2026 call that "over 50 customers will have complete tape-out on our 22nm platform." That installed base of qualified designs is the moat - each one took a customer a year to qualify and is sticky.

Why it stands apart. This is where UMC competes most directly with both TSMC's mature lines and China's SMIC. It is the margin-and-growth bet: as commodity 40/65nm fades, 22/28 is the node UMC is steering customers toward.

2.2 Specialty technologies (~50% of revenue)

What it does. Process platforms designed for a physical function rather than density: RF-SOI (radio chips), embedded high-voltage (display drivers, OLED), embedded non-volatile memory / eNVM (microcontrollers, secure elements), BCD (Bipolar-CMOS-DMOS, for power management), and CMOS image sensors. Management said specialty is roughly 50% of total revenue.

Core capability. Specialty processes are co-developed with customers over years and are far less price-elastic than commodity logic. UMC announced a 55nm BCD platform on the Q3 2025 call serving "mobile and consumer... automotive and industrial." This is the part of UMC that SemiAnalysis flagged as insulated from Chinese overcapacity because it carries "significant single-source business."

Why it stands apart and how it fits. Specialty is the defensive moat and the margin stabilizer. When commodity logic pricing gets crushed by Chinese capacity, specialty is where UMC retreats to defend ASPs. Management's entire pricing narrative is "differentiate the technology offering... improve the ASP resilience" (Q2 2025).

2.3 Advanced packaging, FinFET via Intel, and silicon photonics (the optionality)

What it does. Three forward bets that are small today but reframe the "sleepy mature-node" story: (1) advanced packaging - 2.5D interposers with deep-trench capacitor (DTC), wafer-to-wafer 3D stacking, and bridge dies for chiplet integration; (2) a 12nm FinFET node co-developed with Intel at Intel's Arizona fab, giving UMC a FinFET offering without building its own FinFET line; (3) silicon photonics on 12-inch wafers (IMEC-licensed PDK targeted for 2027) for optical interconnect.

Why it exists separately. Each is a route back toward higher-value work UMC abandoned in 2018. The Intel deal is capital-light re-entry into FinFET; advanced packaging lets UMC participate in AI/chiplet trends; photonics is a long-dated option on optical I/O. Management called the Intel program "too important, especially for UMC. We are putting all the possible resources" (Q1 2026).

How it fits. This is the strategic-option bucket - immaterial to current revenue, central to the multi-year re-rating thesis.

2.4 Geographic manufacturing footprint (the "China+1" segment)

What it does. UMC runs ~12 fabs across Taiwan, Singapore, China (Xiamen), and Japan (Mie). The newest is Singapore Fab 12i Phase 3, completed in 2025. Management explicitly markets geographic diversity as a product: a customer worried about Taiwan-Strait risk or US tariffs can have its chips built in Singapore or Japan.

Why it's strategic. Wang on Q3 2025: "UMC has a geo-diversified manufacturing site across the globe." This is increasingly a reason customers pick UMC over a Taiwan-only or China-only competitor. It is the cash-cow footprint plus the Singapore ramp as the near-term growth lever.

3. Products and business detail

The product catalogue is process platforms, not chips. UMC sells access to qualified manufacturing recipes:

• Logic nodes: 22nm, 28nm (high-performance and high-density variants), 40nm, 55/65nm, 90nm, 0.11µm, and a long tail of 8-inch processes down to 0.5µm. 22nm is the flagship; below-40nm processes are now more than half of total wafer revenue.
• RF-SOI: Radio front-end and switch processes for 5G/WiFi. UMC is a major supplier here; this rides the 5G/6G RFIC wave and is built with "scalable 3D wafer-to-wafer stacking and TSV... in mass production for the extremely small form factor for the 5G and 6G RFIC" (Q2 2025).
• Embedded high-voltage (eHV): The dominant process for display driver ICs and OLED panels.
• Embedded non-volatile memory (eNVM): For microcontrollers, smartcards, and secure elements.
• BCD / power management: Including the new 55nm BCD platform (announced Q3 2025) for mobile, consumer, automotive, and industrial power.
• CMOS image sensors (CIS): For cameras across phones, automotive, and industrial vision.
• Advanced packaging: 2.5D interposer with deep-trench capacitor, wafer-to-wafer 3D stacking, bridge die and DTC for chiplets - "more than 35 new tape-outs in 2026" expected (Q1 2026).
• 12nm FinFET (with Intel): PDK/IP delivery in 2026, customer tape-out 2027, production "later 2027," for DTV, WiFi connectivity, and high-speed interface chips.
• Silicon photonics: 12-inch PIC platform, PDK v1.0 targeted 2027.

Manufacturing. UMC operates roughly 12 fabs with combined capacity above 400,000 12-inch-equivalent wafers per month. The four 300mm (12-inch) fabs are the strategic core: Fab 12A (Tainan, Taiwan - the 22/28nm flagship), Fab 12i (Singapore - now with Phase 3 added in 2025), Fab 12X (Xiamen, China - ~32,000 wafers/month, runs at full capacity), and Fab 12M (Mie, Japan - the former Fujitsu fab acquired in 2019 as United Semiconductor Japan, running ~40/65nm specialty). The remainder are 8-inch (200mm) fabs in Hsinchu and Singapore that run mature specialty and analog. 2026 capacity grows only ~1.2% year-on-year - this is a tightly disciplined capacity story, not a build-out, with the Singapore Phase 3 ramp the main addition (meaningful volume in H2 2026).

Geographies of sale. Asia is roughly two-thirds of revenue (~67%), North America around a quarter (rose to ~25% in Q3 2025), and Europe under 10% (~8%). By customer type, fabless design houses are ~80% and IDMs ~19-20%.

Milestones that changed the business: the 1995 pivot from IDM to foundry; the 1996 spin-off of MediaTek/Novatek; the 2000 five-way merger and NYSE listing; the 2018 decision to halt below-14nm development; the 2019 Mie (Japan) fab acquisition; the 2023 buyout to majority control of the Xiamen JV; the 2024 Intel 12nm partnership announcement; and the 2025 completion of Singapore Fab 12i Phase 3.

4. Customers

Who buys. UMC's customers are the world's fabless chip designers plus some IDMs: MediaTek, Qualcomm, Broadcom, Realtek, Novatek, Texas Instruments, and Intel are named publicly, alongside hundreds of small and mid-size design houses. Several of the largest - MediaTek, Novatek - are companies UMC itself spun off in 1996, an unusually deep historical tie. Geographically the base skews Asian (~67%) with a growing North American share (~25%).

Who makes the decision and on what criteria. Inside a fabless customer, the choice of foundry is made by the VP of operations/manufacturing together with the design team, and it is made per product. The criteria: does the foundry have the right specialty process, what is the yield, what is the wafer price, can they get capacity allocation, and - increasingly - where geographically are the fabs. The sales cycle is long: qualifying a new design on a UMC process runs six months to over a year before volume production.

Why they choose UMC. Three specific reasons recur. First, specialty process fit - for an RF-SOI radio or an embedded-high-voltage display driver, UMC has a tuned, proven platform. Second, geographic diversification - UMC explicitly sells Singapore and Japan capacity as a hedge against Taiwan and China concentration, and management says this is winning business ("acknowledge... the geolocation benefit," Q2 2025). Third, price relative to TSMC for work that does not need TSMC's leading edge.

Switching costs. High and structural. Once a design is taped out and qualified on a UMC process, moving it to another foundry means re-qualifying from scratch - new test wafers, new yield ramp, new reliability sign-off - a year of work and real money, and a fresh risk of field failures. This is why UMC's "over 50 customers... taped out on 22nm" matters: each is an annuity. Management leans on this: the strategy of pushing customers up to 22nm is partly about deepening that lock-in.

Concentration. UMC does not disclose named single-customer percentages, but the base is broad - hundreds of designers plus the long IDM tail - which limits single-customer risk relative to a TSMC (where Apple/Nvidia dominate). The flip side: in commodity logic, customers can and do dual-source against SMIC, so concentration risk is lower but pricing power is also lower outside specialty.

Contract structure. A mix. Most business is recurring volume on qualified designs (sticky but not formally long-term), some spot/short-cycle, and a growing layer of longer-term capacity agreements UMC has pursued to underwrite fab investments. Revenue is therefore moderately predictable - utilization swings with the cycle, but the qualified installed base puts a floor under volumes. UMC also flagged tariff-driven inventory pull-forward inflating Q2/Q3 2025 demand, a reminder that near-term order flow can be noisy.

5. Competitive landscape

UMC sits in the middle of a barbell. Above it is TSMC, which controls roughly 70% of the entire foundry market and all of the leading edge. Below and beside it is a crowded mature-node field where Chinese foundries are adding capacity aggressively. UMC's strategy is to climb out of the commoditized middle into defensible specialty.

The named competitors:

• TSMC - the giant. Competes with UMC on mature 22/28nm logic, where TSMC also has large capacity. UMC cannot win on prestige or leading edge; it competes on price and on being a credible second source. TSMC's ~70% share makes it the gravitational center of the industry.
• SMIC (China) - UMC's most direct threat. SMIC is expanding mature-node capacity rapidly with state backing and grew ~16% in 2025 while non-Chinese mature foundries grew ~8%. SMIC competes hardest on commodity logic and price. UMC's defense is explicitly to "reduce exposure to more commoditized market segments" and move to specialty where SMIC is weaker. SMIC has overtaken UMC in overall foundry share (~5.3% vs UMC ~4.35%).
• GlobalFoundries (US) - the closest peer in strategy: also abandoned the leading edge, also specialty-focused (RF-SOI, automotive, US/EU/Singapore footprint). The two compete head-to-head in RF-SOI and for "China+1" customers. SemiAnalysis groups UMC and GF together as specialty foundries shielded by single-source business.
• Samsung Foundry - #2 overall (~7.2%) but focused on advanced nodes; overlaps UMC less directly, mostly at the 28nm boundary.
• Vanguard, PSMC, Tower (now Intel-owned), Nexchip, HuaHong - the long tail of mature/specialty foundries. Nexchip (China) grew ~24% in 2025, intensifying the commodity squeeze.

Where UMC wins: specialty processes with single-source lock-in (RF-SOI, eHV, BCD), the 22nm planar sweet spot, and geographic diversity (Singapore + Japan + Taiwan + China) that no Chinese competitor and few others can match. Where it is exposed: commodity 40/65nm logic, where Chinese capacity drives prices down and UMC has no structural advantage.

Barriers to entry are very high in absolute terms - a 12-inch fab costs billions and process know-how takes years - but the relevant threat is not a new entrant; it is subsidized expansion by existing Chinese players. That is a barrier-erosion story, not a barrier-breach story, and it is the central competitive risk. SemiAnalysis estimated global mature-node operating profit (ex-TSMC/Samsung) fell ~23% in 2024 under exactly this pressure.

This is not a clean moat narrative. In commodity logic, margins are being competed away and UMC says so by steering out of those segments. The genuine moat is narrower: the specialty single-source business and the qualified-design installed base.

6. Industry

What drives demand. UMC's chips are the connective tissue of electronics - radios, display drivers, power management, sensors, microcontrollers. Demand tracks smartphone units, PCs, networking gear, automotive electronics content, and industrial. Unlike the leading edge, which rides AI training/inference, mature-node demand is broad and unit-driven. The newer tailwind is that AI systems still need a halo of mature-node chips around every advanced processor - power delivery, connectivity, sensing - which management repeatedly cites ("AI-related segment to remain as the primary growth driver," Q4 2025).

Size and trajectory. The global foundry market hit a record ~$169-170 billion in 2025 (top-10 foundries), up ~26% year-on-year, though that growth is hugely concentrated in TSMC. TSMC reached ~69.9% share; UMC was ~4.35%, third-to-fifth depending on the quarter, behind SMIC (~5.3%) and ahead of GlobalFoundries (~3.87%). Mature-node foundry specifically grew far more slowly (~8% ex-China) and is where overcapacity fears concentrate.

Where UMC sits in the supply chain. UMC is upstream of OSAT (assembly/test) and the fabless designers. It is a manufacturing utility for everything that is not the leading-edge processor. Its position is structurally durable - the world will need 22-65nm chips for decades - but structurally lower-margin than the leading edge and increasingly contested.

Import-substitution dynamics. This is the defining industry force. China is building mature-node capacity to substitute imported chips, both for cost and for self-sufficiency under US export controls. That floods the commodity end. Symmetrically, Western customers want "China+1" supply (and the US/EU want non-China capacity), which is the tailwind UMC's Singapore/Japan footprint captures. UMC sits on both sides of this: threatened by Chinese commodity capacity, helped by Western diversification demand.

Regulation. US export controls, potential US tariffs on chips, and the CHIPS-Act-style subsidy regimes shape where capacity gets built and who can sell to whom. UMC flagged tariff uncertainty repeatedly (Q2/Q3 2025) and noted some 2025 demand was pull-forward ahead of feared tariffs. Its US-adjacent Intel 12nm partnership is partly a regulatory hedge - "we're investing into the U.S., so we're definitely going to present our case" on tariff exemptions (Q3 2025).

Cyclicality. Deeply cyclical. The industry runs inventory cycles; UMC's utilization swung from 69% (Q1 2025) to 79% (Q1 2026) over the period studied - a ~10-point swing that flows straight to margins because fabs are high-fixed-cost. Mature-node utilization is the single best real-time gauge of where UMC is in the cycle.

Net: tailwinds are AI-halo content, geographic diversification demand, and 22nm migration; headwinds are Chinese overcapacity in commodity logic, tariff/FX uncertainty, and a high-fixed-cost cost base in a depreciation-peak phase.

7. Growth triggers

All items sourced to the four concalls.

• Singapore Fab 12i Phase 3 ramp to meaningful volume in H2 2026. Completed in 2025; production began January 2026; higher-volume ramp from H2 2026 into 2027. Repeated across all four calls.

  "The 12 IP3 production ramp will start in January 2026, and it will ramp up with a higher volume starting in second half of 2026. And that milestone schedule remains." (Q3 2025, Oct 29 2025)

• 22nm tape-out acceleration as the primary growth lever. Over 50 customer tape-outs; 22nm hit a record 14% of revenue in Q1 2026; management guides "high-teen percentage" growth in 22nm in H2 2026 vs H1. Repeated and intensifying across all four calls.

  "Over 50 customers will have complete tape-out on our 22nm platform for a very diverse range of applications, including display driver IC, network chips, and microcontrollers." (Q1 2026, Apr 29 2026)

• 22/28nm combined to grow double-digit year-over-year in 2026. (Q3 2025, Oct 29 2025) - "we expect the overall 22- and 28-nanometer revenue to achieve double-digit year-over-year growth in 2026."

• Intel 12nm FinFET partnership reaching first tape-outs in 2027. PDK/IP delivery in 2026, customer tape-out 2027, initial production "later 2027," for DTV, WiFi, and high-speed interface chips. Repeated across all four calls with consistent timeline.

  "We are on track with our 12 nm cooperation with Intel, which should start to see tape out in 2027." (Q4 2025, Jan 28 2026)

• Advanced packaging scaling: 35+ new tape-outs in 2026. 2.5D/3D stacking, bridge die, DTC; >10 customers engaged. Trigger grew between calls (Q4 2025 said ">20 tape-outs"; Q1 2026 said ">35"). (Q4 2025 Jan 28 2026; Q1 2026 Apr 29 2026)

• H2 2026 wafer price increase. UMC sent customers letters announcing a second-half 2026 price adjustment citing raw material, energy, logistics, and geopolitical cost pressures. (Q1 2026, Apr 29 2026)

  "We recently sent out a letter to our customers, talking about the price increase to happen in the second half of 2026." (Q1 2026)

• 55nm BCD specialty platform readiness for mobile, consumer, automotive, and industrial power. (Q3 2025, Oct 29 2025)

• Silicon photonics PDK v1.0 in 2027 on 12-inch wafers (vs competitors at 8-inch), targeting pluggable optics and future co-packaged optics. (Q4 2025 Jan 28 2026; Q1 2026 Apr 29 2026)

• Full-year 2026 to outperform the addressable market, with H2 stronger than H1 (a break from normal seasonality). Repeated Q3 2025, Q4 2025, Q1 2026.

  "UMC is confident that 2026 will be another growth year as tape-outs on our 22 nm platform accelerate." (Q4 2025, Jan 28 2026)

8. Key risks

• Chinese mature-node overcapacity crushing commodity pricing. Mechanism: SMIC, Nexchip, HuaHong add subsidized 28-65nm capacity faster than demand grows; commodity wafer prices fall; UMC's non-specialty logic gets squeezed on both volume and ASP. This is high-probability, moderate-to-serious drag, and it is already happening - SemiAnalysis estimated mature-node operating profit ex-TSMC/Samsung fell ~23% in 2024. UMC's own mitigation (exit commoditized segments, push specialty) is an admission of the threat. Wang, Q2 2025: the share "competing with the Chinese foundries will continue to decline" - by design, because that competition is unwinnable on price.

• New Taiwan dollar appreciation eroding margins. Mechanism: UMC bills largely in USD but reports in NT$ and has a NT$ cost base; a stronger NT$ directly compresses reported margin. Management quantified it precisely: "every 1% move appreciation of NT dollars against U.S. dollars, it will erode our gross margin of about 0.4 to 0.5 percentage points," and a 6%+ NT$ move cut Q2 2025 margin by ~3 points (Q2 2025). High-probability, moderate, and largely outside management's control.

• Depreciation-peak margin headwind. Mechanism: the Singapore Phase 3 and prior 12-inch investments carry heavy depreciation that hits the P&L before the revenue ramps. Management said it is in "the peak of our depreciation increase cycle" (Q1 2026) with continued quarter-over-quarter increases, though the rate of increase should fall "below 10%" in 2026 vs low-20s% in 2025 (Q2 2025). High-probability, time-limited drag.

• Cyclical utilization swings. Mechanism: fabs are high-fixed-cost; a demand air-pocket drops utilization and margins fast. Utilization ran 69% to 79% across the studied period - operating leverage cuts both ways. Some 2025 demand was tariff-driven pull-forward (Q2 2025), which risks a payback air-pocket. Moderate probability, moderate impact.

• Strategic stranding at 14nm. Mechanism: by halting below-14nm, UMC depends on the Intel 12nm partnership to access FinFET. If that program slips or underdelivers, UMC has no internal path back to advanced logic and stays confined to a slower-growing, more contested segment. Low-probability-of-failure but high-consequence; management's "all the possible resources" language (Q1 2026) signals how much rides on it.

• Customer dual-sourcing in commodity logic. Because switching costs are high only after qualification, customers increasingly qualify a second source (often SMIC) on commodity designs, capping UMC's pricing power outside specialty. Structural, moderate.

• Geopolitical / tariff tail risk. Mechanism: US chip tariffs or Taiwan-Strait disruption could hit either UMC's cost structure or its Taiwan-concentrated capacity. Management flagged tariff uncertainty repeatedly (Q2/Q3 2025). Low-probability, high-consequence; partly hedged by the Singapore/Japan/US-adjacent footprint. (Historical note: UMC pleaded guilty in a 2018 US DOJ trade-secret case tied to Micron/Fujian Jinhua and paid a $60M fine - a reminder of the compliance and geopolitical exposure of a Taiwan foundry serving Chinese customers.)

9. Walk the talk

Concalls assessed: Q2 2025 (Jul 30 2025), Q3 2025 (Oct 29 2025), Q4 2025 (Jan 28 2026), Q1 2026 (Apr 29 2026). The pattern that emerges is a management team that is operationally conservative and consistently delivers on the controllable items (utilization, shipments, the 22nm ramp, capex discipline, project milestones), while being appropriately vague about the items it cannot control (pricing, FX).

Start with Q2 2025. Management guided Q3 shipments to "increase by low single-digit percentage" with utilization "in the mid-70% range," and flagged that 22/28nm at 40% of revenue was a record. They also told investors the Singapore Phase 3 ramp would start in 2026 and contribute revenue in H2 2026, and that the Intel PDK would be ready "in June 2026" with customer tape-out in 2027.

Move to Q3 2025. Shipments did rise (3.4% wafer-shipment growth) and utilization landed at 78% - at or above the mid-70s guided. That is a kept promise. They then guided Q4 shipments "flat," ASP "firm," gross margin "high 20% range," and utilization "mid-70% range." They reaffirmed the Singapore milestone verbatim ("ramp will start in January 2026... higher volume... second half of 2026... that milestone schedule remains") and pulled the Intel PDK timeline forward - from "June 2026" to "the early PDK will be ready for the first wave of customers in January 2026." Pulling a milestone forward, not pushing it back, is the credibility tell.

"Our wafer shipment will remain flat... Capacity utilization rate will be in the mid-70% range." (Q3 2025 guidance for Q4)

Q4 2025 showed utilization held flat at ~78% - again at the top of or above the guided mid-70s, a modest beat. 22nm hit a record 31% sequential jump to >13% of revenue, validating the repeated "22nm acceleration" story. Capex was cut from $1.8B (2025) to $1.5B (2026) - discipline, not a build-spree, consistent with the "outperform the market" framing rather than a capacity gamble. They guided Q1 2026 shipments "flat," utilization "mid-70% range."

Q1 2026 delivered utilization of 79% - again ahead of the mid-70s guided - and 22nm reached a record 14% of revenue, exactly the trajectory promised across the prior three calls. The Singapore ramp, Intel 2027 tape-out, and advanced-packaging tape-out count (raised from >20 to >35) all advanced as described. They then guided Q2 2026 shipments to "high single digit" growth.

Where they hedge rather than promise: pricing and FX. On the Q3 2025 call, asked about 2026 ASP, Wang deferred - "we will provide more detail in the upcoming January 2026 conference call, as we are going through some discussion with our customers." That is not evasion so much as honesty about negotiations in flight; by Q1 2026 they had concretely announced the H2 2026 price increase. The one genuinely soft spot is margin: management has been candid that depreciation and FX are squeezing gross margin regardless of operational delivery, and they have steered investors toward EBITDA rather than gross margin as the better gauge (Q1 2026) - a subtle reframing that a skeptic should note.

Assessment: this is management that does what it says on the variables it controls. Utilization has met or beaten guidance four quarters running, the 22nm story has compounded exactly as narrated, project milestones have held or pulled forward, and capex has been disciplined. They under-promise on the uncontrollables. This is a credible, conservative team - not a promotional one.

10. Shareholder friendliness index

Dividends. UMC pays an annual cash dividend, and over the last three fiscal years it has declined steadily: NT$3.00 for FY2023 (paid 2024), NT$2.85 for FY2024 (paid 2025), and NT$2.60 for FY2025 (approved at the May 27 2026 AGM). The prior year (FY2022) was NT$3.60, so the trend is four straight years of lower dividends. Notably, FY2024 EPS (NT$3.80) was higher than the year before yet the dividend was cut - UMC is choosing to retain more cash for the Singapore ramp and depreciation cycle rather than maximize payout. The payout ratio remains high (well over half of earnings; trailing per-share metrics on US ADR data even show payout optically above 100% in some windows due to the depreciation-depressed near-term earnings), which tells you the absolute dividend is being managed down deliberately, not because the company lacks cash. The CFO framed it on Q3 2025: "we always try to strike a good balance between the high percentage payout ratio and absolute dividends."

Buybacks and dilution. UMC ran an active buyback in 2026: the board authorized up to 50,000,000 shares (0.40% of shares outstanding) between April 30 and June 29 2026 at NT$52.50-109.50, and execution was underway - cumulative treasury holdings reached ~30.55 million shares by May 20 2026, with a fresh tranche of ~10 million bought in that window (these buybacks are earmarked for employee incentives). Separately, UMC executes small capital reductions to cancel unvested restricted shares (e.g., 1.54 million shares cancelled in 2026, ~0.012% of capital). Net share count is therefore roughly flat-to-slightly-shrinking - there is no meaningful dilution, and the buyback plus dividend together return real capital, but the buyback is modest in scale (<0.5% of shares).

Verdict: Returns Capital (moderately). UMC consistently pays a large dividend and runs buybacks with no dilution, but the steadily declining dividend - cut even in a higher-EPS year - shows capital is being conserved for the investment cycle rather than maximized for shareholders right now.

11. Insider activities

UMC is dual-listed (TWSE 2303 / NYSE ADR). As a foreign private issuer it historically did not file SEC Form 4; instead it discloses monthly insider holding changes via Taiwan's MOPS, which it mirrors into SEC Form 6-K filings. A new wrinkle: under the US Holding Foreign Insiders Accountable Act (enacted Dec 18 2025), UMC's directors and officers must, from March 18 2026, report holdings and transactions under Section 16(a) - so US-style insider data is now becoming available. The data below comes from UMC's monthly 6-K disclosures (the official MOPS mirror).

Recent transactions (most recent first):

Buys - read the signal. The one open-market accumulation is independent director Su Lin Wang, who went from zero to 101,000 shares in January 2026. A board member building a position from scratch is a constructive signal, though as an independent director (not the CEO or CFO) and a single, modest-sized purchase, it is a mild positive rather than a thunderclap. There was no cluster of insider buying - this is the only buy in the window.

Sells - work out the why. The selling is concentrated among officers in February-March 2026, led by CFO Chitung Liu trimming 649,000 shares across two months (~14% of his stake). The timing coincides with the period right after the strong Q4 2025/Q1 2026 results and the share-price rally that the market staged in early 2026 (the stock jumped ~6% on Q1 results), and with the new Section 16(a) reporting taking effect. None of the 6-K filings disclose a specific reason for the sales, and there is no indication of a 10b5-1-style plan. The most likely explanations are routine diversification and tax management against a rallied share price by executives who hold multi-million-share positions accumulated over long tenures - not a thesis-driven exit. But it should be stated plainly: reason not disclosed in the filings, and the sells are broad across six officers, which is worth noting even if individually small relative to their holdings.

Net assessment. Insiders were net sellers in the most recent window, with the activity concentrated in officers (led by the CFO) selling into strength, against a single offsetting buy by an independent director. The officer selling is modest relative to total holdings and best read as diversification after a rally rather than a red flag, but the absence of any executive open-market buying - the strongest bullish tell - means the insider signal lands at neutral to mild concern. Nothing here suggests insiders see trouble; nothing here shows the conviction of executives putting fresh personal capital into the stock.

12. Scenarios

Bull case. The Singapore Phase 3 fab ramps on schedule through H2 2026 and fills with "China+1" diversification demand, letting UMC charge a geographic premium that holds ASPs firm even as Chinese commodity capacity floods the low end. The 22nm migration compounds - the 50-plus qualified tape-outs turn into volume, 22/28nm pushes past 40% of revenue and keeps climbing, and the H2 2026 price increase sticks because customers have nowhere cheaper to qualify in time. Specialty stays at half of revenue and grows, insulating margins. Then the optionality starts paying: the Intel 12nm partnership delivers first tape-outs in 2027 on schedule, giving UMC a credible FinFET offering without the capex, and advanced packaging plus silicon photonics make UMC a real participant in the AI-adjacent buildout. The depreciation hump rolls off, operating leverage kicks in as utilization climbs back toward the 80s, and the market stops pricing UMC as a sleepy mature-node also-ran and starts pricing it as a specialty grower with a re-entry option into advanced logic.

Base case. UMC delivers roughly what it has guided. Utilization stays in the high-70s to low-80s, wafer shipments grow through 2026 with H2 stronger than H1, and 22/28nm posts the double-digit growth management promised. Singapore ramps but its depreciation keeps gross margin under pressure near-term, so earnings grow modestly rather than dramatically. Specialty defends the franchise against Chinese commodity pressure, but commodity logic stays a slow leak. The Intel partnership progresses on its 2027 timeline without yet moving the revenue needle. Capital returns continue - a large but gently declining dividend plus small buybacks. UMC remains the world's third-to-fifth foundry, a disciplined, cash-generative, cyclical business that grows in line with or slightly ahead of its addressable market, neither re-rating sharply nor breaking.

Bear case. Chinese overcapacity gets worse faster than UMC can pivot. SMIC, Nexchip, and HuaHong keep adding subsidized 28-65nm capacity, commodity wafer prices fall through UMC's cost floor, and the planned H2 2026 price increase gets walked back as customers dual-source to China. A demand air-pocket - payback for the 2025 tariff-driven pull-forward - drops utilization back toward the low 70s or below, and operating leverage works in reverse just as Singapore's depreciation peaks, squeezing margins from both sides. A stronger New Taiwan dollar compounds the hit, each percentage point of appreciation shaving margin. The Intel 12nm program slips or disappoints, stranding UMC at 14nm with no path back to advanced logic and no growth story beyond a contested, commoditizing mature-node base. Capital returns shrink further, and UMC drifts down the foundry rankings as the gap to TSMC widens and the floor rises beneath it from China.

13. Further reading

• Foundry Industry Model - SemiAnalysis [paid]

Sources

• UMC Q1 2026 earnings call transcript - Investing.com
• UMC Q4 2025 earnings call transcript - Investing.com
• UMC Q3 2025 earnings call transcript - Insider Monkey
• UMC Q2 2025 earnings call transcript - Insider Monkey
• UMC Q1 2026 results 6-K - SEC EDGAR
• UMC investor relations / quarterly results
• UMC company overview - Wikipedia
• Foundry market share 2025 - Taipei Times
• Global foundry market record 2025 - Tom's Hardware
• Mature-node foundries face overcapacity (cites SemiAnalysis) - EE Times
• SemiAnalysis Foundry Industry Model
• UMC dividend history - StockAnalysis
• UMC NT$2.85 FY2024 dividend approved
• UMC NT$2.60 FY2025 dividend approved
• UMC Q1 2026 results & buyback authorization 6-K - StockTitan
• UMC March 2026 insider share changes 6-K - StockTitan
• UMC Section 16(a) new insider disclosure rules 6-K - StockTitan
• UMC capital reduction & buyback progress May 2026 - Globe and Mail
• UMC customers / market position - wccftech

A note on scope per the report's rules: I deliberately kept revenue, margin, and valuation figures out of the narrative (only revenue-mix %, utilization %, market-share %, and the Section 10 dividend/buyback figures appear, as permitted). Section 13 carries a single observed paid SemiAnalysis entry; no qualifying Stratechery or MBI Deep Dives coverage of UMC was found.