United Microelectronics Corporation (NYSE: UMC; TWSE: 2303) — A Real Foundry Recovery Buried Under Circular Earnings

Independent research article. Report date: 2026-08-26. As-of ADS price: ~$19.03 (close 2026-08-26). Follow-up to the 2026-06-26 analysis. Financial figures are New Taiwan dollars unless stated; one NYSE ADS represents five Taiwan ordinary shares.

⚡ Claude’s Take

This block is the author’s own subjective opinion and is provided for general information, not investment advice. The analytical body that follows (Sections 1–15) is deliberately position-free and carries no price target; the single exception is this block.

Verdict: upgrade from AVOID to HOLD / WATCH — do not chase at ~$19; begin accumulating below ~$16 and become genuinely interested around ~$13–14. The change from the June report is real. The ADS fell 31.4%, from $27.73 to $19.03, while the operating business did what the prior bull test required: Q2 shipments rose 10.6% sequentially, blended ASP rose by a low-single-digit percentage, utilization climbed from 79% to 85%, gross margin reached 32.5%, 22nm became 17.5% of sales, and Q3 guidance points to utilization above 90% and gross margin in the mid-30s. UMC also delivered its first mass-produced 12-inch photonic IC. Calling that unchanged would be grading the thesis instead of the evidence.

The reason this is not a BUY is that the stock remains much more expensive than the operating business. At about $47.7 billion of market value, a simple H1 normalization produces roughly $0.30 per ADS of core earnings, versus headline H1 ADS EPS of $0.740. Annualizing the cleaner number, with credit for the guided H2 improvement, points to approximately $0.65–0.75 of 2026 operating earnings per ADS. The current price is therefore roughly 25–29x an improving but cyclical run rate, not the low-teens multiple implied by the headline earnings screen. UMC’s Q2 net income was NT$42.3 billion, but operating income was only NT$15.0 billion. The difference included NT$23.6 billion from associates and about NT$6.1 billion of fair-value gains. Several associates collectively own 904.5 million UMC shares, while UMC owns large stakes in those same entities. UMC’s own share-price rise therefore helped generate equity-method income and marked-up equity that flowed back into UMC’s accounts. That is legal accounting, but it is circular economics. If the ADS falls, part of the reported windfall can reverse.

There is a second catch. The near-term margin recovery is largely the attractive kind of semiconductor operating leverage: more wafers running through an installed cost base. The newly approved program is different. UMC raised 2026 capex to $2.0 billion, and the board approved $5.0 billion across 2026–2027 for Singapore cleanroom capacity and new Tainan facilities. Management expects depreciation to rise by a low-teens percentage annually for at least two years, while the advanced-packaging facilities do not inflect until 2028–2029 and Intel-linked 12nm production is not meaningful until 2028. Shareholders are being asked to capitalize today’s utilization recovery before seeing the returns on tomorrow’s plant.

My fair-value zone rises modestly from June’s $11–16 to $13–17 per ADS. The top end gives UMC credit for mid-30s near-term gross margin, firmer mature-node pricing, and credible photonics/advanced-packaging options. The range reflects roughly 20–23x normalized ADS earnings around $0.65–0.75, still a premium to UMC’s pre-mania history. At $19 the asymmetry is no longer absurd, but it still requires a durable margin step-up and useful returns from the new capacity. Conviction: medium. The evidence that turns me bullish is two reported quarters of at least mid-30s gross margin, positive ASP, and normalized free cash flow covering the higher capex while 22nm, photonics, and packaging revenue are disclosed with enough precision to verify management’s broad AI definition. The evidence that turns me bearish is falling ASP or utilization while depreciation rises, or a material reversal of associate/fair-value income after UMC’s share-price retreat. Tag: “The cycle improved; the accounting got stranger.”

Changes since 2026-06-26

• Confirmed: The mature-node recovery is operational, not merely promotional. Volume, utilization, ASP, gross margin, and 22nm mix all improved in Q2, and Q3 guidance extended the direction.
• Partly falsified: The prior report was too dismissive of near-term AI-adjacent execution. A mass-production 12-inch photonics delivery and a defined 2027 platform release are better evidence than an undated option, although the commercial scale remains undisclosed.
• Confirmed: Valuation risk mattered. The ADS fell 31.4% in two months, including a drop from a June intraday high of $28.96 to a July low of $16.54.
• New risk: Headline earnings and equity became dominated by securities marks and associate cross-holdings, making unadjusted P/E and P/B less informative than in the prior report.
• New capital-cycle commitment: UMC approved $5.0 billion of 2026–2027 spending and now expects low-teens annual depreciation growth for at least two years. The near-term cycle strengthened while the prospective supply response became larger.

📈 Stock Price Action — Five-Year Event Map

The UMC ADS has completed two separate boom-bust sequences in five years. The pandemic shortage carried it from roughly $6 in early 2021 to an adjusted peak near $11.8; the inventory correction then drove it below $5 in October 2022. It spent most of 2023–2025 between roughly $5 and $9 as a high-yield mature-node cyclical. The second boom was far sharper: from a year-end 2025 level near $8, it reached an unadjusted intraday high of $28.96 on 2026-06-25, then fell to $16.54 on 2026-07-28 before recovering to $19.03. The current price is below the 21-day and 50-day exponential averages of $19.32 and $19.77, but above the 200-day average of $15.16. Prices and moves are Facts; the attributed drivers are Interpretations.

Cycle narrative. The first two events were textbook foundry economics: shortages filled fixed-cost fabs and lifted pricing, then inventory and new capacity reversed the same leverage. The quiet 2023–2025 period reset expectations around a 29% gross-margin company paying a large dividend. The 2026 move was not proportional to reported earnings; it was a reclassification of UMC as an AI infrastructure beneficiary. The late-June collapse began before the Q2 result, so it cannot be explained by an adverse print. The Q2 response matters more: the ADS advanced the next day because management validated the near-term recovery, yet it did not reclaim the mania peak. The tape has moved from one-way momentum to a contested range in which operating improvement and multiple compression are happening simultaneously.

1. Executive Summary

United Microelectronics Corporation is a pure-play semiconductor foundry concentrated in mature and specialty processes. It manufactures chips designed by fabless companies and IDMs, with no leading-edge 7nm/5nm/3nm production. Its economic center is 22/28nm plus differentiated versions of older nodes used in communications, power management, display, embedded memory, sensors, automotive, and consumer devices. In Q2 2026, 22/28nm represented 37% of wafer sales; 40nm and 65nm contributed another 33%. Fabless customers were 85% of sales, Asia was 66% of geography, and communications plus consumer were 71% of applications.

The immediate cycle is better. Q2 revenue rose 17.0% year over year to NT$68.733 billion. Gross margin was 32.5%, operating margin 21.8%, and utilization 85%. Q3 guidance calls for high-single-digit shipment growth, firm dollar ASP, mid-30s gross margin, and above-90% utilization. Those figures represent genuine operating leverage and answer the most important question raised in June: pricing and loading are improving together. The 22nm mix reached 17.5% of sales, and 12-inch utilization is above the company average. UMC is not merely pushing low-priced legacy wafers through empty 8-inch fabs.

The moat remains narrow. In Greenwald’s taxonomy, UMC has limited demand-side captivity from qualified process design kits, customer tape-out costs, long product lives, and quality/reliability records in specialty processes. It lacks the combination of global scale, process leadership, and customer ecosystem that protects TSMC. It also lacks GlobalFoundries’ comparable U.S./European trusted-fab positioning. Customers incur switching costs after qualification, but new designs can be directed elsewhere; the advantage lives within specific process/customer niches rather than across the enterprise. UMC’s long-run market share and mid-cycle returns do not demonstrate a broad franchise moat.

The capital cycle is the central conflict. UMC spent US$12.0 billion from 2021 through 2025 and now plans another US$2.0 billion in 2026. The new board authorization totals US$5.0 billion across 2026–2027. Near-term returns benefit from filling already installed capacity, but the announced Singapore and Tainan facilities carry new depreciation before the associated revenue is fully visible. Management explicitly expects low-teens annual depreciation growth for at least two years. In Marathon’s framework, the industry’s stronger pricing signal is already eliciting a supply response—from UMC, Chinese foundries, and specialty competitors—before investors can establish whether AI-adjacent demand is durable.

Reported earnings require reconstruction. Q2 operating income was NT$14.950 billion and operating cash flow NT$33.698 billion. Yet parent net income was NT$42.260 billion because non-operating income reached NT$30.236 billion. Associate profit of NT$23.606 billion and fair-value gains near NT$6.054 billion were the dominant bridge. Several associates own UMC shares; UMC owns meaningful stakes in those associates. H1 headline ADS EPS was $0.740, while a rough normalization excluding associate and fair-value gains yields about $0.30. Reported book value is also inflated by market-value changes. The balance sheet remains liquid and conservatively levered, but the accounting optics are materially less clean than in 2025.

Capital allocation is mixed. UMC maintained a 25-year dividend record and shareholders approved NT$2.60 per ordinary share for 2025 earnings, equivalent to roughly $0.41 per ADS at current exchange rates. The balance sheet held NT$124.706 billion of cash at June 30. Offsetting those strengths, the company issued NT$12.12 billion of domestic unsecured convertible bonds in August and acquired NT$3.096 billion of treasury stock in Q2 while approving a much larger multiyear construction plan. The financial capacity exists; the unresolved issue is whether prospective projects clear the cost of capital after depreciation.

Valuation remains demanding even after the fall. At $19.03, roughly 2.505 billion ADS-equivalent shares imply market capitalization of about $47.7 billion. A 2026 operating-earnings range of $0.65–0.75 per ADS implies 25–29x, before giving full credit to investment assets. The broad AI definition—roughly $300 million of 2026 revenue, targeted to exceed $1 billion within three years—includes power management, connectivity, FPGA, packaging, and photonics, not direct accelerator compute. The embedded expectation is therefore more than an AI revenue ramp: it is that UMC preserves improved mature-node pricing, absorbs low-teens depreciation growth, and converts early packaging/photonics programs into high-return revenue.

Verdict: UMC’s operating trajectory is stronger than it was in June, but its enterprise economics remain those of a cyclical, capital-intensive, narrow-moat foundry. Headline net income overstates the recurring earnings base, and the capacity response increases the burden of proof. The institutional conclusion is scenario-dependent and position-free: the current market value requires a meaningful, durable improvement beyond a normal utilization rebound.

2. Business Overview

Foundry model and economic unit

UMC manufactures integrated circuits to customer designs. Revenue is driven by wafer shipments, price per wafer, node and specialty mix, exchange rates, and ancillary services. Gross profit is exceptionally sensitive to loading because depreciation, fab labor, utilities, and maintenance are substantially fixed over a useful operating range. A wafer added to an underutilized qualified line can carry a high incremental margin; a wafer requiring a new cleanroom and tool set carries a very different return. That distinction is essential to interpreting 2026.

The company operates 12-inch and 8-inch fabs in Taiwan, Singapore, Japan, and China, with a small 6-inch legacy footprint. Q2 quarterly capacity was about 1.305 million 12-inch-equivalent wafers, guided to 1.325 million in Q3. Singapore Fab 12i was the visible increase, moving from 172,000 equivalent wafers of quarterly capacity in Q2 to an expected 192,000 in Q3. These are equivalent units, not literal wafer counts across all diameters.

One ADS represents five ordinary shares. Revenue and operating accounts are reported in Taiwan dollars, while ADS investors experience NT$/US$ translation. About 12.527 billion ordinary shares correspond to roughly 2.505 billion ADS equivalents. Aggregator market caps and per-share statistics are dangerous if the ADS ratio, ordinary count, and reporting currency are mixed. This report reconstructs those units directly.

Technology and revenue mix

UMC exited the leading-edge race years ago. Its commercial advantage is not smaller transistors; it is adapting stable process nodes for specific analog, power, RF, display, embedded-memory, image-sensor, and connectivity functions. Q2 2026 wafer-sales mix was:

Sub-40nm revenue was 52%, but that phrase can mislead: for UMC it means mostly 22/28nm, not EUV-based leading edge. The May 2026 release of a 14nm eHV FinFET platform and the Intel 12nm collaboration are development milestones, not current volume revenue. Management expects 12nm customer tape-outs in 2027, pilot activity first, and more meaningful production in 2028.

Application mix was communications 39%, consumer 32%, computer 13%, and other 16%. Geography was Asia 66%, North America 22%, Europe 8%, and Japan 4%. Fabless customers were 85%; IDMs 15%. The mix shows both opportunity and risk. Power, connectivity, and controllers can benefit from AI infrastructure and electrification, but consumer and communications remain the majority, exposing UMC to smartphones, PCs, displays, and inventory cycles.

Specialty platforms and customer captivity

Specialty processes add device structures, materials, libraries, or reliability regimes to a base node: high-voltage and BCD for power; RF SOI and RF CMOS for connectivity; embedded nonvolatile memory; CMOS image sensors; display drivers; and now silicon photonics, deep-trench capacitors, interposers, and wafer stacking. Once a chip is designed and qualified on a given process, moving it requires redesign, mask expense, requalification, yield learning, and sometimes end-customer approval. For automotive or industrial programs, the product life can last many years.

That creates genuine customer captivity at the design level. It does not make all UMC capacity differentiated. Commodity mature-node logic and standard display/consumer work remain exposed to price. Management says specialty products are roughly half of revenue, but the label is broad and does not equal monopoly economics. The correct test is whether the specialty mix produces stable market share, pricing, and returns through cycles. UMC’s gross margin fell from 45.1% in 2022 to 29.0% in 2025, demonstrating that captivity cushions the cycle without eliminating it.

Emerging AI-adjacent businesses

Management defines AI-related revenue broadly. It includes power-management ICs, connectivity, FPGA-related products, advanced packaging, silicon photonics, and other chips associated with AI end products. The company projects approximately US$300 million in 2026 and more than US$1 billion within three years. At the starting point, this is roughly 4% of a revenue base near US$8 billion; even the target would be meaningful but not dominant.

The most tangible milestone is the July mass-production delivery of a 12-inch photonic IC. UMC argues 12-inch manufacturing provides better process control, performance, yield, and propagation loss than prevailing 8-inch offerings. It plans a generally available platform in 2027. Advanced packaging offerings extend beyond interposers to deep-trench capacitors, wafer-to-wafer stacking, memory stacking, and customized memory solutions. Management cited more than 35 product tape-outs, many involving DTC. It did not disclose customer concentration, contracted volumes, pricing, or current packaging revenue, so the earnings significance remains an assumption.

Verdict: UMC is a diversified mature/specialty foundry with real design-level switching costs and improving 22nm mix. Its new AI-adjacent platforms are credible enough to monitor but too early and too broadly defined to transform the current business classification.

3. Industry Dynamics

Two foundry industries, not one

Leading-edge and mature-node foundry share a manufacturing vocabulary but have different economics. Leading-edge competition is restricted by extreme R&D, EUV tooling, yield knowledge, ecosystem depth, and annual capital budgets that few firms can fund. Mature and specialty nodes use older tools and processes, lower absolute capital per line, and serve fragmented end markets. Entry is still difficult, but government subsidies and patient strategic capital can support capacity that would not pass a private return hurdle.

UMC competes primarily in the latter market. TSMC remains present at mature nodes but allocates its best capital and engineering resources to advanced logic and advanced packaging. GlobalFoundries focuses on specialty platforms and geographic security in the United States, Europe, and Singapore. SMIC, Hua Hong, Nexchip, PSMC, Vanguard, Tower, Samsung, and various IDMs compete by node and application. The relevant competitive set changes with each qualified process; global foundry-share tables can overstate direct overlap.

Current utilization and pricing cycle

The durable semiconductor-primer framework is that pricing and capacity utilization drive foundry profit more than smooth end-demand growth. UMC illustrates it precisely. Utilization fell from shortage-era levels to the high-60s/70s during the correction, gross margin fell 16 points from its 2022 peak, and earnings declined. In 2026, shipments, ASP, and utilization began rising together. Q2 utilization was 85%; Q3 is guided above 90%. Twelve-inch lines are above corporate utilization while 8-inch is expected to recover toward 85% in Q3.

This is a favorable near-term supply-demand signal. It does not establish a structural shortage. Management describes a broader, more sustainable recovery led by AI spillover into connectivity, power, memory, and infrastructure, while acknowledging mixed non-AI end markets. Inventory has improved but has not disappeared as a risk. A few quarters of restocking can look identical to final-demand acceleration until customer inventories and sell-through diverge.

Capital cycle and prospective supply

UMC’s annual capex was approximately US$1.8 billion in 2021, $2.7 billion in 2022, $3.0 billion in 2023, $2.9 billion in 2024, and $1.6 billion in 2025. The company now plans $2.0 billion in 2026 and has authorized $5.0 billion across 2026–2027. Ninety percent of 2026 spending is directed to 12-inch assets. Singapore P4 prepares photonics cleanroom capacity; Tainan P7/P8 lays the foundation for advanced packaging; existing Singapore P3 and Taiwan 12A add BCD, photonics, memory stacking, and DTC tools.

In Capital Returns terms, the industry is moving from the repair phase into renewed investment before pricing power has been tested through a full cycle. Capacity announcements are not automatically bearish—customer-linked specialty tools can earn attractive returns—but cleanrooms and fabs create long-lived fixed supply. Management says deployment will be phased and demand-led, which is a useful discipline claim. The falsification test is whether customer commitments, utilization, ASP, and incremental returns appear before depreciation.

Chinese mature-node expansion is the largest external distortion. State-supported manufacturers can add 28/40/55/65nm capacity to achieve localization and security objectives rather than maximize shareholder returns. That weakens the normal high-price signal: excess returns attract capital, but the capital can keep arriving even after returns fall. Export controls may slow leading-edge progress while redirecting Chinese investment toward exactly the mature nodes where UMC competes. Conversely, restrictions and customer diversification can create demand for non-mainland supply, especially for sensitive customers. The result is segmentation, not a simple global shortage.

Demand vectors

AI infrastructure raises mature-node content through power conversion, connectivity, optical transceivers, FPGAs, controllers, sensors, and packaging. Automotive electrification increases power-management and microcontroller content. Robotics, satellites, and edge computing add further specialty demand. These are credible unit and content drivers, but most do not grant UMC direct exposure to accelerator economics. A dollar of power or connectivity revenue is still subject to alternative foundries, customer bargaining, and capital intensity.

The non-AI base remains mixed. Smartphones, consumer electronics, PCs, and displays account for much of UMC’s wafer mix and can swing with inventories. Management’s Q3 demand call is strongest in power-management ICs, sensors, and microcontrollers, with a rebound in 8-inch utilization. Sustainable industry tightening requires that breadth to persist after restocking.

Regulation and geopolitics

Taiwan concentration is economically material. A cross-strait conflict or blockade would overwhelm ordinary valuation analysis; lesser tensions can affect customer sourcing, insurance, freight, energy security, and capital allocation. UMC partly diversifies through Singapore, Japan, and China, but its core remains Taiwan. U.S. and allied export controls also shape which customers and tools can serve Chinese fabs. UMC’s older-node focus reduces direct exposure to the most advanced restrictions but not to entity-list compliance or customer end-use rules.

Verdict: The near-term mature-node cycle has tightened enough to improve UMC’s pricing and fixed-cost absorption. The longer capital cycle is less attractive: subsidized Chinese supply and UMC’s own new-build program increase the probability that today’s margin signal induces tomorrow’s depreciation and capacity pressure.

4. Competitive Position

Moat classification

Under Greenwald and Kahn’s framework, UMC has narrow demand-side captivity, not a broad cost or scale moat. The captivity comes from process qualification, design libraries, masks, yield learning, product-lifecycle continuity, and regulatory/customer approval. It is strongest in specialty and long-life products after tape-out. It is weakest before the next design is assigned, when customers can choose TSMC, GlobalFoundries, SMIC, Hua Hong, Samsung, Tower, or another qualified supplier.

UMC has scale relative to small specialty foundries, but scale alone is not an advantage when larger competitors operate similar or better nodes. TSMC’s larger R&D budget, utilization network, customer ecosystem, and process breadth are superior. UMC can earn good returns on a carefully reused mature-node platform; it cannot spread leading-edge development across the industry’s largest revenue base. Its Intel 12nm collaboration is a pragmatic way to share technology cost, and also evidence that UMC does not possess the technology independently.

Share and financial tests

Stable market share and sustainably superior returns are the empirical tests. UMC is generally the fourth-largest global pure-play foundry, with roughly 4–5% share depending on quarter and data definition. It has not gained enough share to demonstrate an expanding enterprise moat. Gross margin near 29% in 2025 was better than some peers but far below TSMC’s leading-edge economics. Through-cycle ROIC has been around or only modestly above the cost of capital, with shortage-year returns falling sharply as utilization normalized.

The Q2 improvement does not change the moat classification. Gross margin of 32.5% and operating margin of 21.8% show strong cost control and fixed-cost leverage, not necessarily pricing power. The more probative evidence is the low-single-digit ASP rise alongside higher shipments and management’s claim of value-based pricing. That claim becomes credible if ASP stays firm after utilization exceeds 90% and competitors add capacity. It fails if the next down-cycle forces broad price concessions.

Competitive comparisons

Process and ecosystem options

22nm is the clearest current differentiator. It lets customers migrate from 28nm for lower power and improved density without paying leading-edge cost. Its 17.5% Q2 mix indicates adoption beyond a slide-deck promise. The Intel 12nm project could extend this strategy, offering a FinFET-class node through shared development. But tape-outs in 2027 and meaningful production in 2028 leave execution, yield, customer adoption, and economics unproven.

Silicon photonics and advanced packaging offer a different route: use UMC’s manufacturing competence in structures adjacent to advanced compute rather than compete in compute logic. The first mass-production photonics delivery is valuable evidence. UMC claims 12-inch process control and yield advantages over 8-inch offerings; competitors will dispute the uniqueness, and the company has not disclosed volume or customer economics. The same caution applies to DTC and stacking: more than 35 tape-outs show engagement, not revenue quality.

Customer bargaining and concentration

UMC does not disclose customers by name in quarterly materials, and no single customer is presented as the basis of the thesis. Fabless customers comprise 85% of revenue. Large chip designers can dual-source or shift new designs, while smaller customers value UMC’s libraries, support, and capacity continuity. Long product cycles can stabilize wafer demand, but customer inventory adjustments still propagate rapidly into utilization.

The durable advantage is therefore local: a qualified specialty flow with a customer that values continuity. The enterprise remains exposed to industry price and capacity. If UMC’s claimed specialty differentiation were a broad moat, gross margin and returns would not have compressed as sharply after 2022.

Verdict: UMC is a capable operator with real but narrow switching costs. It has no enterprise-wide scale or technology moat. The 22nm platform strengthens its position within mature foundry; photonics, packaging, and 12nm are options whose competitive economics have not yet been demonstrated.

5. Growth History and Forward Opportunities

A cyclical history, not a steady compounder

UMC’s revenue history reflects the semiconductor cycle more than a stable secular growth curve. Shortage conditions in 2021–2022 produced unusually high utilization, pricing, and profitability. Revenue and earnings then weakened across the inventory correction even as the company continued investing. FY2025 revenue recovered modestly to approximately NT$237 billion, but ordinary EPS of NT$3.34 was well below the NT$7.09 shortage-era peak. The pattern matters because a valuation based on peak incremental margins assumes the next downturn will be gentler than the last.

The first half of 2026 marks an operational acceleration. Revenue reached NT$129.77 billion, up 11.3% year over year. Q2 growth of 17.0% was supported by both volume and price, a healthier combination than volume bought through discounting. High-single-digit Q3 shipment guidance implies revenue can grow again even with firm rather than sharply rising ASP. If delivered, annualized revenue would move above NT$290 billion in Q3. That is a material cyclical recovery, though not yet a full-year structural growth rate.

22nm migration

The most visible organic driver is customer migration from 28nm to 22nm. UMC’s 22nm share reached 17.5% in Q2, a record. The combined 22/28nm pool was 37%, down from 40% a year earlier; applying the mix to revenue indicates combined dollars still rose about 8%. The nuance prevents two opposite mistakes. It is wrong to say the platform is not growing merely because combined mix fell, and equally wrong to imply the entire 22/28 pool is gaining share. Separate 22nm disclosure suggests migration within the pool while older nodes also rebounded.

22nm can offer lower power and smaller die than 28nm without the design cost and wafer price of leading-edge processes. Connectivity, display, controllers, and edge applications can benefit. The commercial test is sustained dollar growth, ASP, and yield—not just mix—because 22nm can cannibalize 28nm. If it replaces the same customer function at a better price and margin, the migration creates value; if it only prevents attrition, it protects the base.

Silicon photonics

The July 2026 shipment to SILITH moved photonics from development to production. The companies said the 1.6T platform reached production readiness in 18 months and was qualified by a leading cloud-infrastructure customer. UMC plans general availability in 2027. The 12-inch format may improve process control, performance, propagation loss, and yield while offering more dies per wafer than 8-inch manufacturing.

The competitive benchmark is demanding. Tower reported a silicon-photonics annual run rate near US$680 million in Q2, targets more than US$1 billion in Q4, has disclosed US$1.3 billion of 2027 contracts and US$290 million of customer prepayments, and has shipped millions of coherent photonic ICs with Marvell. UMC disclosed none of those commercial proof points. Its first shipment validates manufacturing capability, not leadership or scale.

Investors should watch four measures: named or countable production customers, annualized wafer revenue, customer funding or take-or-pay commitments, and segment-level incremental return on the Singapore P4 tools. Without them, a general platform release is a product roadmap rather than an investable earnings bridge.

Advanced packaging and DTC

UMC’s packaging strategy avoids direct competition with TSMC’s CoWoS system. It focuses on 3D wafer-to-wafer hybrid bonding, bridge dies, discrete deep-trench capacitors, customized memory stacking, and memory-to-memory stacking. Management says more than ten customers are active and more than 35 products are in discussion or tape-out through early 2027. Existing fabs can support early production, while Tainan P7/P8 provide the shell for later scale.

This is commercially more advanced than the June report recognized. It is also unquantified. UMC does not report packaging revenue, margin, backlog, capacity, or customer concentration separately. Management expects the facility-driven inflection only in 2028–2029 because construction lead time exceeds 20 months. The correct model treats early revenue as optional upside and new-building depreciation as a visible cost, not the reverse.

Intel 12nm collaboration and 14nm eHV

The Intel collaboration is strategically attractive because it gives UMC access to a 12nm platform and Arizona production capacity without owning the entire leading-edge development burden. The process design kit was ready in May 2026. Customer design work and tape-outs begin in 2027, which management characterizes as a pilot or early-ramp year; meaningful production is expected in 2028. The timeline is later than a casual reading of the earlier “production in 2027” description.

The collaboration could serve communications infrastructure, networking, and other products needing FinFET density without the expense of advanced nodes. It could also provide geographic diversification. Risks include execution across two companies, yield, customer adoption, economics shared with Intel, and the possibility that demand moves to newer nodes before volume scales. Management will prove 12nm before considering anything beyond it; there is no 7nm plan.

Separately, UMC released a 14nm embedded-high-voltage FinFET platform for display-driver designs in May. It claims up to 40% lower power and 35% smaller die than its 22nm process. No production customer, volume date, or revenue was disclosed. The platform extends UMC’s established eHV specialty, but 22nm remains the finest node contributing material revenue.

AI revenue target: useful and definition-heavy

The roughly US$300 million 2026 AI-revenue forecast and more-than-US$1-billion three-year objective are the first quantitative anchors for the theme. The definition includes connectivity, power, FPGA, photonics, and packaging associated with an AI end product. That breadth is economically defensible—AI systems require much more than accelerators—but it makes comparisons difficult. The number should not receive an accelerator or advanced-packaging multiple without evidence on product mix and margin.

Exceeding US$1 billion would add more than US$700 million over three years, a roughly 49% compound rate from the stated 2026 base. Against an approximately US$8–9 billion company, that is a significant but not transformative contribution unless the rest of the business also grows. It could lift mix and asset utilization; it cannot alone support a several-fold enterprise re-rating if mature-node pricing and returns normalize.

Market size and realistic share

TrendForce reported Q1 2026 top-ten foundry revenue of US$47.95 billion, a simple US$191.8 billion annualized proxy. UMC’s US$1.93 billion represented 3.9%. The full figure is not UMC’s addressable market because leading-edge AI logic dominates growth and UMC cannot manufacture it. No reliable public primary source separates a current mature/specialty-node serviceable market by process and application.

The defensible growth case therefore starts from UMC’s own revenue rather than a promotional TAM: mid-single-digit through-cycle wafer growth, periodic ASP/mix improvement, 22nm migration, and several hundred million dollars of AI-adjacent revenue. A bull case adds successful 12nm and packaging ramps after 2028. A bear case assumes existing-node price competition offsets the new platforms and the asset base grows faster than revenue.

Growth scorecard

Verdict: Growth has moved from a mostly cyclical recovery plus distant options to a cyclical recovery plus several early-commercial options. The evidence supports higher near-term revenue and margin expectations, but not a transformation into a leading-edge or advanced-packaging franchise. The decisive variable is return on the new assets, not the size of the announced opportunity.

6. Financial Quality

Five-year operating arc

The income statement shows a shortage peak, three years of normalization, and a 2026 recovery. ADS-equivalent EPS below means five times ordinary-share EPS and remains denominated in Taiwan dollars; it is not U.S.-dollar ADS EPS.

The H1 reported row is the trap. Revenue and operating profit did recover, but reported net income exceeded full-year 2025 net income in six months because investment-related gains exploded. Gross and operating margins are the cleaner measures of foundry performance. Q2 alone improved to 32.5% gross margin and 21.8% operating margin from Q1’s 29.2% and 18.5%.

The Q2 earnings-quality break

Q2 parent net income was NT$42.260 billion, or NT$3.39 per ordinary share and US$0.537 per ADS. Operating income was only NT$14.950 billion. The bridge was:

The associate line is not ordinary operating diversification. At June 30, associates collectively held 904.533 million UMC ordinary shares, about 7.2% of the issued base. UMC simultaneously owned 36.49% of Hsun Chieh, 26.78% of Yann Yuan, 18.08% of Silicon Integrated Systems, 13.8% of Faraday, and 12.85% of Unimicron, among other interests. As UMC’s share price rose, entities holding UMC recognized investment gains; UMC then recognized its share of those entities’ results. The same market move also increased other financial-asset marks and OCI.

This is reflexive rather than fabricated. The associates are legal entities with other assets and activities, and equity-method accounting is prescribed. Economically, however, capitalizing the gain as if it were foundry earnings double-counts a share-price move. It also creates reversal risk after the ADS and Taiwan ordinary fell from their June peaks.

Quoted associate investments had carrying value of only NT$30.031 billion and fair value of NT$232.714 billion at June 30, compared with fair value of NT$54.202 billion at year-end 2025. That difference is relevant asset value, but not all of it is freely realizable without tax, governance, liquidity, and circular-ownership consequences. It should not be added dollar-for-dollar to enterprise value while also capitalizing associate income.

Normalization

A transparent rough normalization subtracts NT$23.606 billion of associate income and NT$6.054 billion of fair-value gains from Q2 pre-tax income. Applying FY2025’s 16.44% effective tax rate yields approximately NT$12.973 billion of after-tax earnings, around NT$1.04 per ordinary share or US$0.165 per ADS at the period conversion rate. Using operating income alone after the same tax gives about NT$1.00 per ordinary share. The defensible Q2 range is therefore NT$1.00–1.04 per ordinary share or US$0.16–0.17 per ADS. For H1, a simpler normalization using the reported period tax and excluding NT$26.421 billion of associate income and approximately NT$8.169 billion of fair-value/other gains produces about NT$23.751 billion after tax, NT$1.90 per ordinary share, or roughly US$0.30 per ADS.

This is intentionally simple and conservative. It may exclude some recurring dividends or investment economics and may retain minor non-core items. Its purpose is not to create false precision; it demonstrates that headline H1 EPS was more than twice the operating run rate. A valuation should separately assess (1) foundry earnings, (2) net financial assets, and (3) cross-holding value, with appropriate discounts and no double count.

Cash flow and capital intensity

Cash generation is stronger than accounting operating income because depreciation is noncash, but free cash flow is volatile with the build cycle.

Q2 free cash flow benefited from stronger loading before most newly authorized construction spending. The 2026 capex increase to US$2.0 billion and the US$5.0 billion 2026–2027 program mean FY2025’s NT$52.1 billion should not be treated as a no-growth perpetuity. UMC’s presentation reports Q2 free cash flow of NT$23.970 billion under a different capex convention; the table consistently uses cash-flow-statement PP&E acquisition. Timing differences between cash capex, incurred capex, tool delivery, construction in progress, and depreciation will make cash and accounting earnings diverge.

Depreciation rose from NT$37.8 billion in 2023 to NT$45.5 billion in 2024 and NT$56.4 billion in 2025 as the prior investment wave entered service. Management now expects low-teens annual growth for at least the next two years. EBITDA can therefore improve while gross margin stalls. CFO Chitung Liu explicitly emphasized confidence in EBITDA growth but said gross margin will depend on how new-fab depreciation is booked. That is a warning against extrapolating Q2 incremental margin.

Balance sheet and liquidity

At June 30, cash was NT$124.706 billion, total assets NT$665.973 billion, liabilities NT$222.049 billion, and equity NT$443.924 billion. Funds and investments were approximately NT$146.853 billion and PP&E NT$259.212 billion. The company has substantial liquidity, access to domestic debt markets, and no near-term solvency issue.

The word “fortress” needs qualification. Balance-sheet safety is high, but reported equity jumped partly because securities and associate marks flowed through profit and OCI. Q2 total comprehensive income was about NT$72.7 billion versus net income of NT$42.2 billion. P/B based on June equity therefore embeds the same market reflexivity that distorts earnings. Tangible operating assets and realizable net financial assets are safer anchors than unadjusted total equity.

Returns and incremental economics

FY2025 ROE was about 13% and estimated ROIC about 8.4%, around a reasonable cost of capital rather than evidence of a strong franchise. Shortage-year returns were far higher; that is precisely why mid-cycle normalization matters. Q2 operating improvement will raise near-term returns, but the denominator is also growing through construction and tool purchases.

The most important future calculation is incremental ROIC on P4/P7/P8, not consolidated EBITDA growth. Filling a depreciated line can generate exceptional incremental cash margin. A newly built line must cover operating cost, depreciation, maintenance capex, working capital, and the cost of capital. Management has not disclosed project revenue, customer prepayments, or returns sufficient to perform that calculation.

Verdict: Liquidity and cash conversion are strong, while through-cycle returns are ordinary. Q2 foundry earnings improved materially, but headline net income and book value are unusually low quality because of fair-value and cross-holding effects. Financial safety is high; earnings comparability and prospective capital efficiency are not.

7. Capital Allocation

Dividend policy and shareholder yield

UMC has paid a dividend for 25 consecutive years and remains oriented toward annual cash distributions. The May 27, 2026 AGM approved NT$32.704 billion for FY2025 earnings. The initial NT$2.60 per ordinary share was adjusted to NT$2.60808262 after restricted-share cancellation. Five ordinary shares per ADS imply gross entitlement near NT$13.04, roughly US$0.41 before withholding at an exchange rate around NT$31.6/US$. At $19.03, the gross yield is about 2.2%.

The distribution fell while the payout ratio rose. That is not automatically imprudent—earnings were depressed and the balance sheet was liquid—but it shows the dividend cannot indefinitely absorb declining operating earnings and rising capex. H1 2026 standard free cash flow of NT$34.330 billion only slightly exceeded the dividend payable, before NT$3.096 billion of treasury-stock purchases.

Treasury stock: compensation, not retirement

The prior report said UMC did not buy back shares. That is now stale. Q2 financing cash flow includes NT$3.096 billion to acquire 30.551 million ordinary shares at an average NT$101.33. The shares are intended for transfer to employees rather than cancellation. The program can align employees and offset restricted awards, but it is not equivalent to retiring shares below intrinsic value. Investors should treat it primarily as compensation funding unless the company ultimately cancels shares.

Following restricted-share cancellation and the treasury purchase, a working market-cap basis is approximately 12.540 billion ordinary shares and 2.508 billion ADS equivalents. Exact issued, outstanding, and diluted counts differ by valuation purpose; the important point is to preserve the 5:1 ratio.

Investment program and realized returns

Capital allocation into fabs is more consequential than the dividend. Standard cash PP&E capex totaled NT$227.762 billion from 2023 through 2025, while FY2025 operating income remained NT$13.942 billion below FY2023. Cash capex/depreciation was 2.42x in 2023, 1.95x in 2024, and 0.85x in 2025. This does not prove each project failed—depreciation and cycle timing lag spending—but consolidated realized incremental returns have been weak.

The new NT$148.680 billion / roughly US$5.0 billion authorization restarts the build cycle before those returns are fully visible. UMC says spending is phased, customer-driven, and aimed at higher-value photonics and packaging. That is a sensible intent. The missing evidence is customer funding, take-or-pay volume, project margin, and incremental ROIC. Capital discipline should be judged by disclosed economics after the plants enter service, not by the strategic attractiveness of AI end markets.

Intel 12nm is the most capital-efficient growth option because Intel provides Arizona manufacturing capacity and UMC contributes process/customer expertise. Singapore P4 and Tainan P7/P8 are UMC-owned asset commitments and should not be described as capital-light. The mix of partnership and owned capacity is reasonable; the timing raises risk because meaningful revenue is later than depreciation and construction cash outlays.

Investments, cross-holdings, and exchangeable debt

UMC has historically held stakes in technology affiliates including Hsun Chieh, Yann Yuan, Silicon Integrated Systems, Unimicron, and Faraday. Some are customers or ecosystem participants; others are financial investments. The Q2 cross-holding windfall demonstrates the governance complexity. Management should explain whether the stakes are strategic, available for sale, or effectively permanent, and how it prevents circular ownership from obscuring performance.

The Q2 bond-related cash event did not involve UMC-share conversion. It involved UMC’s 2021 US$400 million exchangeable bonds backed by Novatek shares. During Q2, holders exchanged US$192 million of principal for about 11 million Novatek shares. The transaction monetizes an investment and reduces the exchangeable liability; it does not directly dilute UMC ordinary shareholders.

After quarter-end, UMC separately issued domestic zero-coupon convertible bonds to fund machinery. The first authorization was up to NT$12 billion of par, priced for conversion at NT$146.0 per ordinary share, and produced NT$12.12 billion of proceeds. The second authorization was up to NT$4 billion, with a NT$130.7 conversion price. Full conversion at the authorized par amounts would imply roughly 113 million new ordinary shares, around 0.9% of the post-cancellation base before anti-dilution adjustments. This is modest prospective dilution and low-cost financing, but it also confirms that the new equipment program is not funded solely from retained cash flow.

Incentives and governance

Six of nine directors are independent, and all six sit on the audit, remuneration, and nominating committees. Directors were elected in May 2024 for three-year terms. FY2025 aggregate director compensation was NT$48 million; executive compensation was NT$923.4 million, including NT$286 million of bonus. Restricted-stock expense was NT$483 million, about 0.20% of revenue.

Earlier restricted plans included ROE, operating-margin, and EPS hurdles; the 2024 plan adds relative total shareholder return and other performance/market conditions. The filed disclosure is insufficient to establish project-level ROIC alignment. That gap matters because management can achieve EPS or EBITDA growth through capital expansion that fails to earn the cost of capital.

UMC began filing U.S. Section 16 ownership reports in 2026 after a change in law; the available history is therefore incomplete and should not be treated as a long-term behavioral record. CFO Chitung Liu sold 1.9 million ordinary shares at NT$154.89 on July 13, a material transaction but small relative to the company. A vice president sold 10,000 shares on June 30. No open-market insider purchase signal appeared in the reviewed filings. Foreign-private-issuer history and Taiwan reporting conventions limit comparability.

The prior report’s description of 6.62% director/officer ownership overstated direct personal alignment because the disclosed category includes holdings and structures that are not all equivalent to discretionary executive capital. The safer conclusion is moderate governance alignment, not founder-like ownership.

Verdict: UMC is financially conservative and historically shareholder-oriented, but the capital-allocation burden has increased. Dividends remain credible, the employee treasury program is not a true buyback, and affiliate holdings complicate earnings. The US$5.0 billion build requires project-level returns that management has not yet disclosed.

8. Changes and Headwinds — Last Two Years

Operating and strategic timeline

What improved

The strongest positive change is the breadth of the operating recovery. Higher shipments were accompanied by higher blended price, rising utilization, and gross-margin expansion. Chinese peers also reported very high utilization—SMIC 93.7% and Hua Hong 102.8%—which refutes the idea that a mature-node glut is already depressing the whole market. UMC’s near-term bear trigger is not occurring.

Specialty execution also improved. 22nm reached a record 17.5% of sales. Silicon photonics crossed into production. Advanced packaging has more than ten active customers and more than 35 products in discussion or tape-out. These facts do not prove moat or return, but they are more substantial than a collection of roadmaps.

What worsened

The growth became more capital intensive. The June report treated much of the AI option as capital-light. That remains fair for Intel 12nm, where Intel provides capacity, but not for Singapore photonics and Tainan packaging. The US$5.0 billion authorization, low-teens depreciation growth, and domestic convertible financing change the downside arithmetic.

Earnings quality also deteriorated sharply. FY2025 non-operating income was small enough that the operating story dominated. In Q2 2026, associate and fair-value gains exceeded operating income by almost two times. The cross-holding structure makes the gain partly reflexive to UMC’s own share price. A falling share price can reduce future associate results and OCI even if wafers perform well.

The Intel timeline moved right or was clarified: 2027 is tape-out/pilot, with meaningful production in 2028. Advanced-packaging facility inflection is 2028–2029. Investors therefore face spending and depreciation before commercial scale.

Unresolved headwinds

• China supply: Current capacity is full, but SMIC and Hua Hong are spending aggressively and adding capacity. Subsidies can prolong the response after market returns weaken.
• Inventory breadth: AI-related infrastructure demand is strong, while smartphone/consumer and automotive/industrial inventory signals remain mixed or above history.
• Currency: UMC reports in NT dollars and prices much business in U.S. dollars. NT-dollar appreciation hurt 2025 gross margin; ADS returns add translation risk.
• Taiwan concentration: Most critical production and management remain exposed to cross-strait, energy, water, logistics, and earthquake risks.
• Depreciation: Low-teens annual growth for at least two years can cap gross margin even if EBITDA grows.
• Disclosure: AI, photonics, and packaging revenue, backlog, margins, customer funding, and project capital are not separated.

Verdict: The last two years contain a genuine cycle and execution improvement, offset by a more aggressive capital response and a new accounting-quality problem. Near-term headwinds are lighter than in June; medium-term capital and valuation risks are heavier.

9. Risk Analysis

Risk interactions

The most dangerous risks are correlated. A mature-node supply response can reduce ASP just as new-fab depreciation rises. That combination compresses gross margin from both sides and weakens free cash flow while the dividend payout is already elevated. It would also reduce the probability that the broad AI target earns attractive returns.

Accounting marks add a separate procyclical layer. If the share price falls because operating expectations weaken, associate holdings of UMC can lose value and feed lower equity-method income or OCI back into UMC’s statements. Reported earnings and book value can therefore deteriorate more than operating income, confusing screen-based investors.

The offsetting interaction is also real. If AI-connected power, optical, and packaging demand stays strong, existing fabs fill before the new facilities arrive, ASP remains firm, and cash generation partly funds construction. Successful specialty ramps can then raise the quality of mix at the same time as depreciation grows. That is the operating path required to avoid a capital-cycle squeeze.

Tail risk versus underwriting risk

Taiwan conflict is the largest severity but is not the most useful day-to-day underwriting variable; it is difficult to price with ordinary scenarios. The recurring underwriting risks are ASP, utilization, depreciation, and capital productivity. A rigorous thesis should not use the geopolitical tail to excuse paying any price, nor ignore it because the event probability is hard to estimate.

Verdict: UMC has low balance-sheet risk but high cycle, capital-intensity, and regional concentration risk. The central downside is not insolvency; it is paying a growth multiple for earnings that normalize while new assets depress returns.

10. Valuation Discussion (Embedded Expectations)

Rebuilt current valuation

Valuation starts with unit discipline. One ADS represents five ordinary shares. Approximately 12.540 billion economically outstanding ordinary shares imply 2.508 billion ADS equivalents. At $19.03, market capitalization is $47.73 billion. June cash, current time deposits, and current FVTPL assets less funded debt provide analytical net liquidity of about $2.84 billion, producing enterprise value of approximately $44.88 billion. Strategic investments are excluded from this liquidity calculation because they are not cash equivalents and include cross-holding complexity.

Trailing normalized operating metrics through Q2 are approximately:

The normalized TTM net-income estimate combines reported H2 2025 parent profit with normalized H1 2026 profit. It is an analytical denominator, not company guidance. It retains the benefit of current operating recovery and excludes the dominant H1 associate/FVTPL gains.

Why headline P/E and P/B mislead

Reported TTM profit would imply roughly 18.4x P/E, and June equity would imply about 3.4x P/B. Those figures are not independent evidence of value. Associate and FVTPL marks inflated H1 earnings, while the same investment and cross-holding marks increased book equity. Using both as confirmation counts the same share-price-driven gain twice.

Nor can the gross fair value of associates simply be added to an earnings valuation. Some affiliates own UMC shares, UMC owns affiliates, the stakes may not be readily monetizable, and selling could create tax, control, market-impact, or ecosystem consequences. Historical-cost fab book also understates replacement cost for depreciated lines. Reported book value is therefore neither worthless nor a clean liquidation/reproduction value.

Historical and peer context

The 31% drawdown removed much of the June blow-off premium, but current normalized multiples remain above UMC’s FY2020–2025 reported ranges: approximately 5.5–25x P/E, 1.5–3.8x P/B, 1.8–3.7x sales, and 2.4–7.3x EV/EBITDA. The comparison is imperfect because current earnings are normalized and historical earnings were reported/cyclical. It is nevertheless difficult to describe 31.8x normalized earnings and 12.6x EBITDA as a conventional mature-foundry valuation.

Directional peer readings sharpen the issue. UMC’s normalized P/E is near TSMC’s trailing P/E, and its EV/EBIT is slightly higher, despite TSMC’s wide scale/technology moat and much higher margins. UMC’s EBITDA multiple is already near GlobalFoundries, while UMC currently earns better margins but lacks GFS’s geographic-security position. Tower trades at far higher multiples, but has disclosed silicon-photonics scale, contracts, and customer prepayments that UMC has not. Accounting, geography, and mix prevent mechanical comp pricing; the useful conclusion is that UMC already receives substantial credit for improvement.

The prior report’s exact claim that every valuation measure sat at the 99.94th percentile is not reproducible. Current backfilled data support 99.94th for P/B and P/S at the June date, about 99.26th for P/E, and 99.72nd for the composite. The correction does not alter the qualitative conclusion that June was extreme.

Earnings-power and reverse-DCF tests

Under a Greenwald earnings-power lens, a 9–10% cost of equity applied to $47.73 billion of market capitalization requires roughly $4.30–4.77 billion of perpetual no-growth equity earnings, versus normalized current earnings/FCF near $1.5–1.65 billion. Most market value therefore represents franchise growth rather than current earnings power. That can be rational for a wide-moat compounder; it is a higher burden for a roughly WACC-return foundry.

A simplified enterprise reverse DCF produces the same message. At 9.5% WACC and 2.5% terminal growth, $44.88 billion of EV requires about $3.14 billion of steady-state annual free cash flow. A sensitivity range is:

At 15%, 20%, and 25% steady-state FCF margins, the midpoint requires revenue of approximately $20.9 billion, $15.7 billion, and $12.6 billion, respectively, versus $7.93 billion TTM. The market must be underwriting a combination of revenue scale and durable margin, not merely Q3 utilization.

FY2029 operating scenarios

These are analytical operating scenarios and do not estimate the share price. All use about 2.53 billion diluted ADS equivalents, including maximum current convertible dilution.

The bear case assumes modest revenue growth, supply-driven pricing pressure, and new depreciation without sufficient mix. The base case assumes 22nm, AI-adjacent platforms, and firm mature-node pricing lift revenue and preserve mid-30s gross margin. The bull case requires roughly 21% annual revenue growth from TTM, 41% gross margin, and 22% FCF margin—an outcome close to the reverse-DCF requirement. It is materially better than UMC’s current through-cycle economics.

Depreciation sensitivity

Q2 annualized D&A was about NT$65 billion. Low-teens growth for two years implies roughly NT$81–84 billion by 2028, a NT$16–19 billion / US$0.52–0.60 billion annual pre-tax increase. At a 16.44% tax rate and 2.53 billion diluted ADS equivalents, that is approximately $0.17–0.20 per ADS of annual earnings headwind absent incremental gross profit. On $10 billion of revenue it equals roughly five to six operating-margin points.

This calculation is the bridge between business quality and valuation. Q3 utilization above 90% proves legacy-fab absorption. It says little about whether P4/P7/P8 can generate enough gross profit to overcome the new D&A. Intel 12nm becomes meaningful in 2028, and the new packaging/photonics facilities in 2028–2029. The valuation test occurs after depreciation, not before.

What is embedded

The present enterprise value appears to require most of the following:

1. Legacy utilization and ASP remain strong long enough to finance construction.
2. Revenue reaches at least the base-case range rather than merely relabeling existing power/connectivity revenue as AI.
3. New-build gross profit exceeds $0.5–0.6 billion of additional annual D&A.
4. More than $1 billion of AI-related revenue is incremental and earns at least corporate margins.
5. Capex normalizes after the build, allowing accounting profit to convert into free cash flow.
6. Dilution stays close to the roughly 1% currently authorized.

Verdict: The share-price decline improved the setup but did not produce a mature-foundry valuation. Current EV requires a structural earnings and cash-flow improvement near the upper end of plausible scenarios. The load-bearing evidence is post-depreciation cash ROIC on the new program in 2028–2029.

11. Variant Perception

Consensus-like view

The constructive narrative is coherent: mature-node supply has tightened, TSMC is directing capital elsewhere, UMC’s utilization is heading above 90%, 22nm is a record share, and AI infrastructure creates demand for power, connectivity, photonics, and packaging. The balance sheet can fund growth, and Q2 plus Q3 guidance show that earnings estimates were too low. Under this view, UMC is no longer a stagnant dividend stock but a specialty infrastructure platform.

Bull case

The strongest bull argument is operating leverage plus credible options. Q2 showed volume and price rising together, Q3 guidance extends margins, and Chinese peers’ high utilization suggests demand can absorb new supply. The 12-inch photonics delivery validates manufacturing; more than 35 packaging products and Intel 12nm create multiple routes to 2028 growth. If the company achieves $14 billion of revenue with low-40s gross margin and 20%+ FCF margin, today’s valuation can be supported by future cash flow rather than multiple expansion.

Bear case

The bear case is not that Q2 was weak. It is that investors capitalize old-fab scarcity rents as permanent just when management starts a new asset cycle. UMC has roughly 4% share, no scale moat, and a history of ASP/margin reversal. Chinese and peer capex can cap pricing; low-teens D&A arrives before 12nm and new packaging scale; broad AI labeling hides product economics. Headline earnings and equity are further inflated by circular marks. The base scenario still leaves the current enterprise at 29x FY2029 FCF.

The clean variant

The cycle can be genuinely strong while the security still over-capitalizes temporary scarcity and unproven expansion returns. That differs from dismissing every AI milestone as hype. The variant accepts the Q2 evidence, corrects the prior report where needed, and moves the debate to incremental return on new capital.

Positioning supports this two-sided read. The ADS is down 31% from June and below its 21/50-day averages, but remains 25.5% above its 200-day average and up 190.8% over 252 sessions. Factor exposure is strongest to semiconductors (+0.775), Taiwan (+0.696), and the market (+0.651), with positive momentum (+0.291) and quality (+0.196). Taiwan and technology regimes remain statistically strong while style momentum is softer. With about 49.7% annualized idiosyncratic volatility and only 39.6% factor-model R-squared, UMC is a post-blow-off unwind inside a favorable regional/sector regime—not a simple momentum winner or falling knife.

Verdict: The market is likely right about the current utilization cycle and could still be wrong about duration, accounting quality, and new-build returns. The differentiated question is not “Is AI real?” but “How much incremental post-depreciation cash will UMC retain from its AI-adjacent build?”

12. Fact vs. Interpretation

Verdict: The factual case is stronger operationally and weaker in accounting quality than it appeared in June. The major interpretations—durability, incremental AI economics, and capital returns—remain unproven.

13. Open Questions

1. How much of the US$300 million 2026 AI-associated revenue is truly incremental rather than existing power/connectivity revenue reclassified by end use?
2. What are current revenue, gross margin, backlog, capacity, and customer prepayments for silicon photonics and advanced packaging separately?
3. Which associates generated the NT$23.606 billion Q2 equity-method income, and how much came from their UMC holdings?
4. Will UMC monetize, unwind, or retain the circular cross-holdings, and what tax/governance discount is appropriate?
5. What customer commitments support P4/P7/P8, and what project-level cash ROIC does management require?
6. How much of Q3’s expected margin is price, mix, utilization, currency, and temporary cost benefit?
7. Can 22nm dollar revenue grow without combined 22/28nm mix declining, and does 22nm carry above-corporate margin?
8. What yield and customer milestones must Intel 12nm reach before meaningful 2028 production?
9. When will the second domestic convertible close, and how will conversion/employee-share transfers affect diluted ADS count?
10. Does CFO selling continue, and will UMC maintain Section 16-style disclosure long enough to create a usable insider history?
11. How quickly do SMIC and Hua Hong qualify current expansions, and where do their marginal prices settle?
12. After the current build, what maintenance capex is required to sustain the larger asset base?

Verdict: Disclosure is sufficient to verify the current cycle but insufficient to value the new platforms as standalone franchises. The most important missing data are customer-backed revenue and post-depreciation project returns.

14. What Must Be True

Scoring the prior report’s tests

Prior bull test: two to three quarters of gross margin moving toward the mid-30s with rising ASP and 22nm/advanced-packaging traction. Status: tracking, not passed. Q2 delivered 32.5% margin, positive ASP, record 22nm, and a photonics shipment; Q3 guidance points to mid-30s. Only one quarter is reported, and packaging economics remain undisclosed.

Prior bear test: Chinese mature-node capacity causes ASP and utilization to roll over while depreciation compresses margin. Status: not occurring now. UMC, SMIC, and Hua Hong are highly utilized and pricing is constructive. The capacity and depreciation mechanism remains live for 2027–2029.

For the operating bull case to be right

• Reported gross margin stays at or above the mid-30s for at least two quarters after Q2, with positive/firmer ASP.
• 22nm dollars continue growing and separate disclosure shows economics better than 28nm cannibalization alone.
• AI-associated revenue exceeds the 2026 base toward the $1 billion objective without relying mainly on relabeling.
• Photonics and packaging disclose contracted demand or customer funding before major P4/P7/P8 tools enter service.
• New gross profit offsets low-teens D&A growth, preserving at least mid-20s operating margin through the ramp.
• Post-depreciation incremental cash ROIC on the US$5 billion program clears a reasonable cost of capital.

Bull falsification test: two reported quarters below roughly 32% gross margin while utilization remains high, or project D&A rises without measurable photonics/packaging revenue and customer commitments. Either would show that volume does not translate into durable economics.

For the operating bear case to be right

• Chinese and peer capacity enters faster than specialty demand, pushing ASP lower.
• Q3’s utilization peak proves to be inventory restocking rather than end demand.
• 22nm growth mostly cannibalizes 28nm and lacks pricing/margin benefit.
• Intel 12nm or photonics/packaging milestones move beyond 2028–2029.
• Associate/FVTPL gains reverse as the share price normalizes, reducing headline earnings and book value.
• Capex remains elevated after 2027, preventing normalized earnings from converting into free cash flow.

Bear falsification test: sustained positive ASP and mid/high-30s gross margin through new-fab depreciation, accompanied by disclosed project ROIC above the cost of capital and free cash flow near the reverse-DCF requirement. That would show the expansion created specialty economics rather than commodity supply.

Decision dashboard

Verdict: Both prior tests remain useful but the key time horizon has shifted. Near-term operating recovery is substantially verified; the decisive 2028–2029 test is whether the new assets earn above their depreciation and cost of capital.

15. Source Appendix

Appendix B prioritizes UMC’s 2025 Form 20-F, Q1/Q2 2026 SEC-furnished results and reviewed statements, UMC IR releases, Taiwan Stock Exchange and Federal Reserve data, peer primary disclosures, and public factor/price datasets. Material corrections and unsupported claims are recorded rather than silently carried forward.

Verdict: The thesis is supported primarily by current company filings and peer primary disclosures. Vendor valuation fields are used only after ADS, currency, timing, and accounting reconciliation.

APPENDIX A — Standard Diligence Questionnaire

UMC — Standard Diligence Questionnaire Appendix

Supplement to the 2026-08-26 United Microelectronics Corporation research memorandum. Facts are distinguished from interpretations and assumptions where material.

General

What thoughtful questions have other investors asked?

The Q2 call concentrated on the questions that matter most: how UMC defines AI revenue; whether the more-than-US$1-billion three-year goal is photonics/packaging or also conventional power and connectivity; what prompted US$5 billion of new capacity; when P4/P7/P8 begin production; whether advanced packaging is CoWoS-like; how 12-inch and 8-inch utilization differ; whether pricing is cyclical or value-based; when Intel 12nm becomes meaningful; how depreciation affects gross margin; and whether customer inventory is rebuilding ahead of end demand.

The best additional questions are financial: how much Q2 associate income came from affiliates marking UMC shares; whether management will unwind circular holdings; what customer deposits or take-or-pay contracts support new facilities; and what post-depreciation return hurdle management uses. Those questions distinguish an operating recovery from a value-creating expansion.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low?

Interpretation: Operating earnings are recovering from a cyclical low but are not demonstrably at a new structural peak. Q2 gross margin of 32.5% is above 2025’s 29.0% but well below 2022’s 45.1%. Utilization of 85%, guided above 90%, is strong; ASP has only recently turned positive. Reported net income is at an artificial high because associate and fair-value gains dominate it.

Are results driven by the external environment or company-controlled actions?

Both. External supply-demand tightening drives shipments, ASP, and utilization. Company-controlled actions—22nm migration, process qualification, fab execution, photonics, and packaging—affect mix and capture. The Q2 margin change is predominantly fixed-cost absorption plus price/mix, not proof of a new moat. The $5 billion build is a company capital-allocation choice that will shape later returns.

How stable are revenues?

Revenue is less stable than long product qualifications imply. It rose from NT$213.0 billion in 2021 to NT$278.7 billion in 2022, fell to NT$222.5 billion in 2023, then recovered to NT$237.6 billion in 2025. Customer switching costs stabilize individual designs, but inventory and wafer pricing move the consolidated result. Communications and consumer represent 71% of Q2 sales, limiting end-market stability.

Outlook for products and services?

Near term, Q3 shipments are guided up high single digits, ASP firm, utilization above 90%, and gross margin mid-30s. 22nm is the most visible growth platform. Silicon photonics has entered mass production at undisclosed scale; advanced packaging is early-commercial; Intel 12nm is pilot/tape-out in 2027 and meaningful in 2028. The outlook is positive through the current cycle, with execution and depreciation risk rising after 2027.

How big is the market; growing or shrinking; domestic or international?

TrendForce’s Q1 2026 top-ten foundry revenue of US$47.95 billion implies a simple US$191.8 billion annualized global proxy. It is not UMC’s serviceable market because leading-edge logic dominates. UMC has about 3.9% overall share and roughly US$8–9 billion of current revenue. Its business is international: 66% of Q2 revenue came from Asia, 22% North America, 8% Europe, and 4% Japan. No reliable primary source isolates a mature/specialty-node SAM; a precise TAM would be false precision.

Business Quality & Competitive Moat

Is the industry getting more or less competitive?

Near-term utilization and price are improving, but medium-term competition is intensifying through capacity. SMIC and Hua Hong are highly utilized today while spending aggressively; UMC is adding its own facilities. State-supported localization weakens normal return discipline. Specialty niches can remain attractive, but undifferentiated mature capacity is structurally competitive.

How profitable is UMC?

FY2025 gross margin was 29.0%, operating margin 18.5%, ROE about 13%, and estimated ROIC about 8.4%. Q2 2026 improved to 32.5% gross and 21.8% operating margin. Profitability is better than many mature-node peers in the current quarter, but through-cycle ROIC is around the cost of capital, not a wide-moat return.

How profitable is the industry; how many competitors; barriers to entry?

Profitability is bifurcated. TSMC’s Q2 gross/operating margins were 67.7%/60.3%, reflecting leading-edge scale and ecosystem advantage. UMC was 32.5%/21.8%; GFS 28.3%/9.7%; Tower about 30%/19.6%; SMIC 25.3%/about 17.8%; Hua Hong 16.5% gross. Direct comparability is limited, but the gap is unmistakable.

Barriers include billions of dollars of plant/tools, yield learning, process IP, PDK/EDA ecosystems, customer qualifications, reliability, and secure supply. Subsidies lower the financial barrier, especially at mature nodes. Competitors include TSMC, GFS, SMIC, Hua Hong, Samsung, Tower, Nexchip, Vanguard, PSMC, and IDMs by niche.

Can the business be easily understood?

The operating model is understandable: price times wafer shipments less a fixed-heavy cost base. The difficult parts are node-equivalent capacity, capital/depreciation timing, affiliate cross-holdings, and project-specific returns. A reader who does not reconcile ordinary shares to ADSs or operating income to associate gains can reach a materially wrong valuation.

Can it be undermined by foreign low-cost labor?

Labor arbitrage is not the main threat. Semiconductor fabs depend on tools, yield, IP, utilities, engineering, and scale. The foreign threat is subsidized capital and strategic overcapacity, particularly from China, not simply low wages. UMC’s Taiwan cost base is already competitive.

Do brands matter?

Consumer brand does not. Reputation for yield, quality, delivery, IP protection, and long-term supply does. In automotive and industrial markets, site qualification and reliability records function like a B2B trust asset. TSMC’s ecosystem reputation is stronger; UMC’s is credible in mature/specialty processes.

What is the nature of competition?

Competition occurs before tape-out on process performance, price, design support, capacity, geography, and qualification. After a design is qualified, switching becomes costly. Capacity utilization then determines how aggressively foundries price the next designs. The market combines local captivity with industry-wide cyclicality.

What are customers’ switching costs?

Customers must port designs and IP, buy masks, requalify yield/reliability, and sometimes repeat automotive or end-customer site approval. Costs are meaningful for a live design and can protect years of wafer demand. They are weaker for new designs, where customers can select another foundry or dual-source. The moat is design/process/site-specific, not customer-wide.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet?

Process know-how, PDK libraries, yield learning, customer qualifications, and a trained workforce are not fully capitalized. Depreciated fabs may have replacement value above book if they remain productive. Conversely, quoted associate investments had fair value far above carrying value at June 30, but part of that value is circular because affiliates own UMC shares. A clean NAV requires look-through statements and discounts.

Off-balance-sheet liabilities?

Ordinary purchase commitments, environmental obligations, leases, guarantees, and construction commitments matter, but no hidden liability threatens solvency in the reviewed filings. The economically important future burden is authorized capex and associated depreciation. Customer/capacity guarantee deposits of about NT$40 billion are liabilities, not debt, and should not be mistaken for excess cash.

How conservative is the accounting?

Operating statements are conventional IFRS, but Q2 headline earnings are not conservative as an economic run rate. NT$23.606 billion of associate income and NT$6.054 billion of fair-value gains dominated net income. Low H1 tax was helped by a deferred-tax benefit. The disclosures permit normalization; screen-level presentation does not.

How capex-hungry is the business?

Extremely. Standard cash PP&E capex was NT$48.0 billion in 2021, NT$80.1 billion in 2022, NT$91.5 billion in 2023, NT$88.5 billion in 2024, and NT$47.7 billion in 2025. UMC plans US$2.0 billion in 2026 and authorized roughly US$5.0 billion across 2026–2027. Low-teens D&A growth for two years shows the continuing economic cost.

Capital Allocation & Management

How much FCF is generated and how is it used?

Standard OCF less cash PP&E produced NT$42.3 billion in 2021, NT$65.7 billion in 2022, negative NT$5.5 billion in 2023, NT$5.3 billion in 2024, and NT$52.1 billion in 2025. H1 2026 generated NT$34.3 billion before back-loaded capex. Uses are fabs/tools, dividends, strategic investments, debt management, and employee-share funding. The cycle makes one-year FCF a poor steady-state measure.

What is management’s philosophy?

Management emphasizes phased, customer-linked investment, value-based pricing, balance-sheet liquidity, and annual dividends. The philosophy is rational; the empirical test is whether new project ROIC clears the cost of capital. The current disclosure does not provide enough project economics to verify that claim.

Significant acquisitions recently?

No large recent acquisition drives the thesis. UMC historically consolidated its Xiamen and Japan operations and holds ecosystem investments, but the current program is organic capacity and partnerships. M&A risk is lower than capex risk.

Is UMC buying back shares?

It bought 30.551 million ordinary shares for NT$3.096 billion in H1 2026. The shares are for transfer to employees, not retirement, and therefore are compensation inventory rather than a conventional value buyback.

Is it issuing large amounts of shares to insiders?

Restricted-stock expense is modest at NT$483 million, about 0.2% of 2025 revenue. Domestic convertibles could add roughly 0.9% to the ordinary base at maximum authorized par. Treasury shares may later be transferred to employees. Dilution is not large today, but the economic share count should include these programs.

Compensation policy?

The 2024 restricted plan includes service, performance, and market conditions, with relative TSR among the metrics. Earlier plans used ROE, operating margin, and EPS. Project-level ROIC is not clearly disclosed, a weakness for a capital-intensive company. Six of nine directors are independent; board and executive compensation are modest relative to enterprise size.

Motivations and alignment?

Management appears motivated by operating scale, technology relevance, customer continuity, dividends, and share performance. Personal ownership is less concentrated than the prior report suggested; Chairman Stan Hung’s personal stake was around 0.47%, while the broader 6.62% group figure includes represented legal entities. CFO Chitung Liu sold 1.9 million shares in July. Alignment is moderate, not owner-operator level.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer?

UMC is a NYSE ADS of a Taiwan foreign private issuer, not an MLP and not a K-1 issuer. One ADS represents five ordinary shares. Taiwan-dollar accounts, U.S.-dollar ADS price, withholding tax, and the ADS ratio must be reconciled.

Dividend policy?

Annual cash dividend approved by shareholders, without a fixed payout promise. FY2025 distribution was adjusted to NT$2.60808262 per ordinary share, about US$0.41 gross per ADS at current FX and roughly 2.2% yield at $19.03. Payout rose to about 81% as earnings declined.

How profitable is the business at the current valuation?

Normalized TTM net income is about US$1.50 billion / US$0.599 per ADS, yielding roughly 31.8x normalized P/E. EV is about US$44.88 billion, or 5.66x revenue and 12.6x EBITDA. These are demanding multiples for roughly WACC through-cycle ROIC.

Is net income diverging from operating cash flow?

Yes, for two reasons. Depreciation makes operating cash flow exceed operating earnings, while Q2 noncash investment marks make reported net income exceed foundry operating income. Q2 OCF was NT$33.697 billion versus parent net income NT$42.260 billion, but normalized net income was only about NT$13 billion. Neither reported net income nor near-term FCF should be used alone.

Risks & Downside

What factors could cause the stock to decline?

Lower ASP/utilization, Chinese capacity, smartphone/consumer inventory weakness, D&A rising faster than gross profit, delayed 12nm/SiPho/packaging, unproductive capex, cross-holding gain reversal, NT-dollar appreciation, dividend pressure, dilution, insider selling, export controls, or broader Taiwan/technology factor reversal.

Risk of catastrophic loss?

A cross-strait conflict, blockade, severe earthquake, extended power/water disruption, cyberattack, or major export-control breach could cause catastrophic operating loss. Geographic diversification mitigates but does not remove Taiwan concentration. Ordinary cycle downside is much more probable than total physical loss.

Chance of total loss?

Low under normal business conditions because UMC has ample liquidity, diversified customers, productive assets, and no solvency stress. It is not zero because geopolitical disruption could make assets inaccessible and ADS claims impaired. The more realistic security risk is severe multiple and earnings compression, not bankruptcy.

Recent News & Events

Has the environment changed recently?

Yes. Q2 demand, utilization, ASP, and margins improved; Q3 guidance strengthened. Silicon photonics entered mass production. At the same time, UMC approved US$5 billion of new facilities, raised 2026 capex, issued domestic convertibles, and reported unusually large cross-holding/investment gains. Current Chinese utilization is high, delaying the glut thesis but encouraging more supply.

Significant acquisitions?

None material to the update. The strategic change is owned capacity expansion rather than acquisition.

Change in accounting policies?

No material policy change is required to explain Q2. The change is the magnitude of fair-value and equity-method results under existing policies, plus a tax benefit linked to a renewed Taiwan–Singapore agreement. Comparability changed even though policy did not.

New markets, facilities, or management changes?

New markets include 12-inch silicon photonics, advanced packaging/DTC/stacking, 14nm eHV, and future Intel-linked 12nm. Singapore P4 and Tainan P7/P8 are new facility programs; P3 continues ramping. No CEO/CFO change anchors the update. CFO share selling is a governance data point, not a management transition.

APPENDIX B — Source Appendix

UMC — Source Appendix

Target: United Microelectronics Corporation (NYSE ADS: UMC; TWSE ordinary: 2303) Research date: 2026-08-26 Status: Final evidence register for the 2026-08-26 report. It records source provenance, claim tests, contradictions, and unsupported items.

1. Price, ADS-ratio, FX, and valuation-claim audit

Current price context

• FACT: The ADS closed at $19.03 on 2026-08-26, down 31.4% from the prior report’s $27.73 reference price. The adjusted-price total return over the same interval was approximately -30.2%, reflecting the subsequent dividend.
• FACT: The trailing five-year low/high were $5.36 / $28.96; the trailing 52-week low/high were $6.56 / $28.96. Current price is 34.3% below the intraday high.
• FACT: Current price is 3.8% below the 50-day EMA ($19.77) and 25.5% above the 200-day EMA ($15.16). The 21-day EMA is $19.32.
• DATE-CORRECTION: The prior report’s discussion blurred the January 2025 trough with the current 52-week low. The January 2025 intraday low was $5.61 on 2025-01-28; the current 52-week low is $6.56 on 2025-09-02.

2. Five-year event map evidence

Price moves below are facts from AZI unadjusted OHLC. Event attribution is separately labeled because a contemporaneous disclosure can corroborate operating conditions without proving why the market moved.

3. Factor-model evidence and handling rules

UMC snapshot and trend

FactorsToday’s leaderboard annualizes every return horizon. Its headline m3/m6 readings of -35.3%/+256.7% annualized de-annualize to approximately -10.3%/+88.9%, broadly but not exactly reconciling with the AZI session-based figures above. Use AZI raw returns in reader-facing work and identify any FactorsToday short-horizon number as annualized or de-annualized.

All Factors loadings

The All Factors calibration is dated 2026-07-31 (pull accessed 2026-08-26), covers 756 sessions, and has R-squared 39.586% / adjusted R-squared 36.912%. Betas are volatility-scaled; correlated sector/industry/country exposures must not be added as if independent.

Quality × Momentum cross: both loadings are positive in this model. UMC is therefore not the classic positive-momentum/negative-quality combination. However, positive long-horizon momentum coexists with a negative three-month return and a one-third drawdown from the June high.

Risk-adjusted track record and regime

FactorsToday annualized readings: 1-year return 199.2%, volatility 61.9%, max drawdown -37.9%, Sharpe 3.19; 3-year return 48.3%, volatility 44.1%, max drawdown -37.9%, Sharpe 1.05; 5-year return 20.0%, volatility 42.6%, max drawdown -54.5%, Sharpe 0.42; lifetime annualized return 12.6%, volatility 45.3%, max drawdown -69.0%, Sharpe 0.23. These are historical/model facts, not forecasts.

As of the pull, the style Momentum factor was soft/neutral: 21-day return -1.30% (z -0.64), 63-day approximately flat (z -0.39), 126-day +3.45% (z -0.09), 252-day +12.82% (z +0.42). UMC’s dominant Taiwan factor was extreme over 126/252 days (+19.93%/+22.98%; z +2.69/+2.68), as was Technology (+19.94%/+22.50%; z +2.99/+2.37). The Semiconductors industry factor was strong over 252 days (+18.65%, z +1.72) but did not cross the |z| >= 2 extreme threshold. Intraday factor z-scores on 2026-08-26 were non-extreme.

Synthesis (interpretation): UMC is best described as a high-volatility post-blowoff momentum unwind inside a still-favorable Taiwan/technology regime, not a clean one-way momentum trade and not a simple falling knife. Approximately 40% of return variation is factor-explained and annualized specific volatility is ~50%, so a substantial part of the move remains stock-specific. This is timing/positioning context only.

4. Prior-thesis test scoring and evidence flags

• Prior price/valuation warning — tracked: $27.73 on 2026-06-25 to $19.03 on 2026-08-26 is -31.4%. This validates the direction of the warning, not a valuation estimate.
• Operational bull test — tracking, not yet passed: Q2 ASP improved low-single digits, gross margin reached 32.5%, and 22nm mix reached 17.5%; Q3 guidance calls for mid-30s gross margin and >90% utilization. The prior test required two to three reported quarters, so one reported quarter plus guidance is insufficient.
• Valuation evidence flag: do not use the current AZI P/E/P/B or FactorsToday market cap as clean valuation inputs without reconciliation. AZI fundamentals remain on lagged inputs; Q2 headline EPS and book value contain circular/cross-holding effects; FactorsToday market cap of $49.16 billion is about 2.7% above the $47.87 billion implied by UMC’s 2026-06-30 issued-share count divided by the 5:1 ADS ratio at $19.03.
• Unsupported baseline flag: the prior report’s precise “99.94th percentile across P/E, P/B, P/S and composite” statement cannot be reproduced. Current backfill only supports 99.94 for P/B and P/S.
• Attribution flag: historical price drivers are interpretations unless a discrete same-day company release is identified. Do not present broad semiconductor-cycle explanations as proven causes.
• Positioning gap: no independently validated current short-interest/days-to-cover series was available in this workstream. Omit rather than infer.
• Date-integrity flag: the 2021-2025 quarterly reports below are retained solely for historical event-map context. They are not current operating evidence.

5. Public source register

1. “Download Historical Data — UMC.” AZI Trading. Dataset through 2026-08-26; accessed 2026-08-26. Vendor CSV / historical market data. https://azitrading.com/controls/download-data.php?t=UMC.
2. “UMC Fundamentals History.” AZI Trading. Dataset through 2026-08-26; accessed 2026-08-26. Authenticated vendor API / backfilled valuation history. https://azitrading.com/controls/get-fundamentals.php?ticker=UMC&history=1. Historical percentiles may change with vendor backfill; this is not a primary source.
3. “Individual Stock Daily Trading Information — 2303, June 2026.” Taiwan Stock Exchange Corporation. Published/updated 2026-06; accessed 2026-08-26. Primary exchange JSON. https://www.twse.com.tw/exchangeReport/STOCK_DAY?response=json&date=20260625&stockNo=2303.
4. “Individual Stock Daily Trading Information — 2303, August 2026.” Taiwan Stock Exchange Corporation. Published/updated 2026-08; accessed 2026-08-26. Primary exchange JSON. https://www.twse.com.tw/exchangeReport/STOCK_DAY?response=json&date=20260826&stockNo=2303.
5. “Taiwan / U.S. Foreign Exchange Rate (DEXTAUS).” Board of Governors of the Federal Reserve System, distributed by Federal Reserve Bank of St. Louis FRED. Updated through 2026-08; accessed 2026-08-26. Official statistical series / CSV. https://fred.stlouisfed.org/series/DEXTAUS.
6. “Frequently Asked Questions.” United Microelectronics Corporation. Current page; accessed 2026-08-26. Company IR webpage / primary source. https://www.umc.com/en/Html/faqs. Used for issued ordinary shares and outstanding ADSs at 2026-06-30.
7. “United Microelectronics Corporation Announces Initial Public Offering of ADSs.” United Microelectronics Corporation, 2000-09-18; accessed 2026-08-26. Company press release / primary historical source. https://www.umc.com/en/News/press_release/Content/investor/20000918. Used to confirm the 5:1 ordinary-share/ADS ratio historically; current ratio is corroborated by UMC’s FAQ and filings.
8. “UMC Reports Second Quarter 2026 Results.” United Microelectronics Corporation, furnished to the U.S. Securities and Exchange Commission as Exhibit 99 on 2026-07-29; accessed 2026-08-26. 6-K exhibit / primary source. https://www.sec.gov/Archives/edgar/data/1033767/000119312526322027/umc-ex99.htm. Used for Q2 operations, Q3 guidance, capex, and per-ADS information.
9. “Unaudited Consolidated Financial Statements for the Six Months Ended June 30, 2026.” United Microelectronics Corporation, furnished to the U.S. Securities and Exchange Commission as Exhibit 99.1 on 2026-07-29; accessed 2026-08-26. Interim financial statements / primary source. https://www.sec.gov/Archives/edgar/data/1033767/000119312526322030/umc-ex99_1.htm. Used for non-operating income, associate stakes/holdings, fair-value marks, equity and book-value quality.
10. “2Q26 Results Report.” United Microelectronics Corporation, 2026-07-29; accessed 2026-08-26. Quarterly results report / primary source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2026/Q2_2026/UMC26Q2_report.pdf.
11. “1Q26 Results Report.” United Microelectronics Corporation, 2026-04-29; accessed 2026-08-26. Quarterly results report / primary source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2026/Q1_2026/UMC26Q1_report.pdf.
12. “4Q25 Results Report.” United Microelectronics Corporation, 2026-01-28; accessed 2026-08-26. Quarterly results report / primary source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2025/Q4_2025/UMC25Q4_report.pdf.
13. “1Q25 Results Report.” United Microelectronics Corporation, 2025-04-30; accessed 2026-08-26. Quarterly results report / primary historical source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2025/Q1_2025/UMC25Q1_report.pdf.
14. “2Q24 Results Report.” United Microelectronics Corporation, 2024-07-31; accessed 2026-08-26. Quarterly results report / primary historical source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2024/Q2_2024/UMC24Q2_report.pdf.
15. “3Q22 Results Report.” United Microelectronics Corporation, 2022-10-26; accessed 2026-08-26. Quarterly results report / primary historical source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2022/Q3_2022/UMC22Q3_report.pdf.
16. “3Q21 Results Report.” United Microelectronics Corporation, 2021-10-27; accessed 2026-08-26. Quarterly results report / primary historical source. https://www.umc.com/upload/media/08_Investors/Financials/Quarterly_Results/Quarterly_2020-2029_English_pdf/2021/Q3_2021/UMC21Q3_report.pdf.
17. “FactorsToday API: UMC stock info.” FactorsToday, latest data 2026-08-26; accessed 2026-08-26. Third-party factor/market-data API. https://www.factorstoday.com/api/stock-info/UMC.
18. “FactorsToday API: UMC stock loadings.” FactorsToday, All Factors calibration dated 2026-07-31; accessed 2026-08-26. Third-party statistical model API. https://www.factorstoday.com/api/stock-loadings/UMC.
19. “FactorsToday API: UMC leaderboard.” FactorsToday, data dated 2026-08-26; accessed 2026-08-26. Third-party risk/return API. https://www.factorstoday.com/api/leaderboard/UMC.
20. “FactorsToday API: UMC stock-specific volatility.” FactorsToday, 252-session window ending 2026-07-31; accessed 2026-08-26. Third-party factor-model API. https://www.factorstoday.com/api/stock-specific-vol/UMC.
21. “FactorsToday API: UMC related stocks.” FactorsToday, accessed 2026-08-26. Third-party factor-profile similarity API. https://www.factorstoday.com/api/related-stocks/UMC.
22. “FactorsToday API: Historic Factor Returns.” FactorsToday, accessed 2026-08-26. Third-party statistical factor-regime API. https://www.factorstoday.com/api/factor-returns/historic.
23. “FactorsToday API: Intraday Factor Returns.” FactorsToday, accessed 2026-08-26. Third-party statistical factor-regime API. https://www.factorstoday.com/api/factor-returns/intraday.
24. “About FactorsToday.” FactorsToday. Current methodology page; accessed 2026-08-26. Vendor methodology. https://www.factorstoday.com/about. Used for factor construction, 10% volatility scaling, ElasticNet loadings, nested-model comparability, R-squared interpretation, and annualization rules.

6. Additional primary and industry sources

25. “Annual Report on Form 20-F for the year ended December 31, 2025.” United Microelectronics Corporation, filed 2026-04-30. Audited financials, customers, process roadmap, qualification, governance, compensation, risks, and ADS basis. SEC filing.
26. “Annual Report on Form 20-F for the year ended December 31, 2023.” United Microelectronics Corporation, filed 2024-04-25. FY2021–2023 comparative financials. SEC filing.
27. “UMC Reports First Quarter 2026 Results.” United Microelectronics Corporation, filed 2026-04-29. Q1 operations, node/application mix, and EPS. SEC exhibit.
28. “Reviewed Consolidated Financial Statements for the Three Months Ended March 31, 2026.” United Microelectronics Corporation, filed 2026-04-29. SEC exhibit.
29. “UMC Q2 2026 Earnings Call Transcript.” Call dated 2026-07-29, published by The Motley Fool 2026-08-07. Management commentary on pricing, utilization, AI revenue, capex, depreciation, Intel 12nm, packaging, and inventory; transcript may contain errors and was checked to company materials. Transcript.
30. “UMC and SILITH Achieve Mass Production Milestone for Silicon Photonics.” United Microelectronics Corporation, 2026-07-14. First 12-inch production delivery and 2027 platform timetable. Company release.
31. “UMC Releases 14nm eHV FinFET Platform.” United Microelectronics Corporation, 2026-05-14. Platform claims and development status. Company release.
32. “Strong AI Demand Drives Top 10 Foundries in 1Q26.” TrendForce, 2026-06-12. Foundry revenue and market-share estimates. Industry release.
33. “Global Pure Foundry Market Share: Quarterly.” Counterpoint Research, updated 2026-06-04. Q4 2024–Q1 2026 share history. Industry data.
34. “TSMC Second Quarter 2026 Results.” Taiwan Semiconductor Manufacturing Company, 2026-07-16. Peer revenue and margin benchmark. SEC-furnished results.
35. “GlobalFoundries Reports Second Quarter 2026 Financial Results.” GlobalFoundries, 2026-08-05. Specialty-foundry peer benchmark. Company release.
36. “Tower Semiconductor Reports Second Quarter 2026 Results.” Tower Semiconductor, 2026-08-04. Specialty-foundry and silicon-photonics peer benchmark. Company release.
37. “Tower Semiconductor Signs Customer Contracts for $1.3 Billion Silicon Photonics Revenue for 2027.” Tower Semiconductor, 2026-05-13. Contract/prepayment comparison. Company release.
38. “Tower Semiconductor and Marvell Ship Over Five Million Coherent Photonic ICs.” Tower Semiconductor, 2026-06-18. Commercial-scale comparison. Company release.
39. “Unaudited Results for the Three Months Ended June 30, 2026.” Semiconductor Manufacturing International Corporation, 2026-08-13. Capacity, utilization, shipments, pricing, margin, and capex. HKEX filing.
40. “Hua Hong Grace Reports 2026 Second Quarter Results.” Hua Hong Semiconductor, 2026-08-13. Capacity, utilization, shipments, margin, and capex. HKEX filing.
41. “USTR Initiates Section 301 Investigation on China’s Targeting of the Semiconductor Industry for Dominance.” Office of the U.S. Trade Representative, 2024-12-23. Primary government description of non-market policies and self-sufficiency goals. USTR release.

7. Capital, ownership, and regulatory sources

42. Domestic convertible-bond board resolutions. United Microelectronics Corporation, filed 2026-06-03. Up to NT$12 billion first and NT$4 billion second bonds for machinery. SEC exhibit.
43. First domestic convertible-bond pricing. United Microelectronics Corporation, filed 2026-07-30. NT$146 conversion price. SEC exhibit.
44. First domestic convertible-bond proceeds. United Microelectronics Corporation, filed 2026-08-05. NT$12.12 billion proceeds. SEC exhibit.
45. Second domestic convertible-bond pricing and July ownership report. United Microelectronics Corporation, filed 2026-08-14. NT$130.7 conversion price, restricted-share cancellation, and CFO share-count change. Public filing copy.
46. Form 4 — Chitung Liu. Filed 2026-07-13. CFO/SVP sale of 1.9 million ordinary shares. SEC ownership XML.
47. Form 4 — Che-Jen Hu. Filed 2026-06-30. Vice-president sale of 10,000 ordinary shares. SEC ownership XML.
48. 17 C.F.R. Section 240.3a12-3. Current e-CFR. Foreign-private-issuer exemptions relevant to interpreting the short U.S. insider-filing history. Regulation.

8. Methodology and coverage notes

• The filing corpus covers 60 months and contains 218 primary documents, but UMC 6-K cover documents do not always include separate Exhibit 99 attachments. Q1/Q2 2026 exhibits were therefore retrieved separately and read directly.
• ADS market capitalization uses ordinary shares divided by five, not actual deposited ADS count and not ordinary shares multiplied by the ADS price.
• Standard free cash flow means operating cash flow less cash PP&E acquisition. Company-defined FCF using a different capex convention is labeled separately.
• Q2 normalized earnings exclude associate profit and FVTPL/other gains and apply a normalized tax rate; this is analysis, not company guidance.
• No validated current short-interest series was found. It is omitted.