Compare moats
Up to three covered companies, band by band. Every call is a curated editorial judgment, never a disclosed figure — and every band carries its cited basis.
| United Microelectronics Corporation | AppLovin | Broadcom | |
|---|---|---|---|
| Moat rating | narrow The FY2025 20-F Competition section cuts both ways: UMC calls itself "one of the world's largest independent semiconductor foundries" with production capacity "currently among the largest of all semiconductor foundries in the world" and says it believes it competes favorably with competitors on each principal element of foundry competition, yet in the same section concedes that "[s]ome of our competitors have substantially greater production, financial, research and development and marketing resources than we have" and that "several new dedicated foundries have commenced operations and compete directly with us," whose entry is "likely to initiate a trend of competitive pricing and create potential overcapacity in legacy technology" that "may erode our profit margins." A real but bounded advantage, not a durable wide one. | narrow The filing's own words support a narrow moat. Auctions run 'at vast scale and at microsecond-level speeds,' revenue is earned only 'when our advertisers achieve their return on advertising spend targets,' and FY2025 adjusted EBITDA margin was 82.3%. What limits the width: clients 'are not required to enter into long-term agreements' and 'Substantially all of the Company's contracts with customers are cancelable at any time'; the solutions 'primarily operate in the mobile app ecosystem and specifically mobile gaming'; and distribution runs through 'the Apple App Store, the Google Play Store, and Meta' while AppLovin relies 'in part on IDFA.' | wide The FY2025 10-K describes category-leading franchises across a broad semiconductor portfolio plus acquired infrastructure software (VMware, CA, Symantec), with top-five end customers at approximately 40% of net revenue — diversified concentration by AI-supply-chain standards — and the Q3 FY2026 release shows non-GAAP operating margin guided at approximately 66% of revenue while revenue grew 86%. |
| Moat type | intangibles ip The 20-F states plainly that "Our success depends on our ability to obtain patents, licenses and other intellectual property (IP) rights covering our production processes and activities," and reports more than 16,700 patents worldwide as of December 31, 2025, of which more than 7,490 are in the U.S., 623 domestic and foreign patents granted in 2025 alone, NT$17,725 million of R&D (7.5% of operating revenues) and 1,591 R&D professionals. The Competition section names where UMC believes it wins: it competes favorably "particularly [on] our technical competence and research and development capabilities." The proprietary specialty-process library, not scale or a network, is the cited source of the advantage. | network effects Network effects, with a caveat. The 10-K states the loop: as more advertisers use the solutions 'we gain access to more data regarding users and user engagement,' 'further strengthening our scaled distribution,' and 'As our distribution grows, we gain better insights for Axon AI.' The mechanism is two-sided matching of advertiser demand to publisher supply, though much of the force comes from data-and-learning scale. It is not switching costs, since contracts are 'cancelable at any time,' nor intangibles or IP, since 'there can be no assurance that we will adequately protect our intellectual property or that it will provide any competitive advantage.' | switching costs The 10-K describes custom accelerators (XPUs) co-designed for hyperscalers and AI frontier-model companies — multi-year, customer-specific silicon programs — alongside enterprise infrastructure software; both are positions a customer leaves only at high cost. |
| Leadership | fast follower UMC describes a deliberate follow strategy in the 20-F: it intends "to avoid investments in technologies that do not present commercial potential for volume production" and believes "it is beneficial to defer investment in premature equipment needed to claim the earliest advanced technology and instead to purchase more advanced and less expensive versions of such equipment from vendors who design this equipment based on pre-production lessons learned from the earliest technology." The node ladder bears this out: the most advanced process in production in 2025 is Fab 12A's 0.014 micron, wafers at 14 nanometers and under rounded to 0.0% of wafer sales in each of 2023, 2024 and 2025, and the next rung — UMC's own 12nm FinFET process, which Intel replicated in 2025 "based on our electrical design rules and UMC-verified process" for manufacture at its Arizona fab — is only "expected to begin in 2027." | at parity The filing discloses no share of any market, names 'Meta, Google, Amazon, and Unity Software' as competitors, and concedes rivals 'may have greater resources' and 'better technological or data analytics capabilities, or stronger brands or competitive positions in certain product segments.' Its one positional statement is parity language: 'We believe we compete favorably with respect to these factors.' Two unquantified points cut upward: Meta and Google are also clients, and MAX is called 'the preferred in-app bidding solution for many publishers worldwide.' | clear leader The 10-K claims 'category-leading semiconductor and infrastructure software solutions' — the company's own attribution — and the Q3 FY2026 release reports AI semiconductor revenue of $16.7 billion in one quarter, up 221% year over year, a scale no other custom-accelerator design house reports. |
| Pricing power | weak The 20-F claims no premium — "our current level of pricing is comparable to that of other leading foundries in each respective geometry" — and reports that 2025 average selling price declined 5.4%, driving gross margin down from 32.6% in 2024 to 29.0% in 2025 even as 12-inch-equivalent wafer shipments rose from 3,446 thousand to 3,870 thousand and average capacity utilization improved from 68.7% to 75.2%. It also notes that "[p]rices for wafers of a given level of technology generally decline over the processing technology life cycle." | strong Strong, with the mechanism shown. The 10-K states that 'Advertisers are charged dynamically based on their campaign goals , rather than a simple fixed price per impression or per action.' Growth composition is the evidence: FY2025 revenue +70% with 'the volume of installations increased 3% and net revenue per installation increased 72%'; H1 2026 +56% with per-installation revenue +75% against 'a decrease in the volume of installations of 10%.' Adjusted EBITDA margin was 82.3% in FY2025 and 84.2% in H1 2026, though payment is performance-contingent and contracts cancel at will. | strong Q4 FY2026 non-GAAP operating income is guided at approximately 66% of revenue, flat from a year ago, while consolidated revenue is guided up 93% — margin held through a near-doubling of revenue. |
| Summary | UMC's durable asset is a proprietary mature- and specialty-node process library — 55/40/28/22nm embedded memory for MCU and automotive, 55/40/28/22/14nm embedded high-voltage for display drivers, 110/55nm BCD for power management, 55/28/22nm BSI-CSI for image sensors and 110/90/55/40nm RF-SOI — backed by more than 16,700 patents worldwide at end-2025 and 7.5% of revenue spent on R&D. That library keeps a concentrated customer base qualified into UMC's fabs (top ten customers were 57.0% of 2025 operating revenues; Texas Instruments, Intel, MediaTek, Realtek and Novatek are named as primary customers). What it does not buy is price: 2025 average selling price fell 5.4% and gross margin fell from 32.6% to 29.0%, while the filing warns that new dedicated foundries are likely to initiate competitive pricing and overcapacity in legacy technology. | One reportable segment: an AI-priced auction matching advertiser demand to publisher supply, where 'Revenue from Axon Ads Manager comprises substantially all of our revenue.' The filed flywheel is real - more advertiser spend yields more user-engagement data, 'further strengthening our scaled distribution,' which feeds Axon AI - and it converts: FY2025 revenue $5.48B, up 70%, at an 82.3% adjusted EBITDA margin, carried by price (net revenue per installation +72%) not volume (installations +3%). What the filing does not evidence is a lock. Contracts are cancelable at any time, revenue concentrates in mobile gaming, and the distribution and identifier layers belong to Apple, Google and Meta - two of whom are simultaneously named competitors and disclosed clients. Wide margins with no disclosed structural lock is narrow. | Broadcom holds two moats under one roof. In AI silicon it is the design partner for custom accelerators and the merchant supplier of the Ethernet switching, NIC, PHY and optical parts that connect them — 'connectivity at scale, within and across AI server racks and across AI data center sites' in the filing's words. In software it owns acquired enterprise franchises. The 10-K's own competition section flags the counterweight: competition is increasing 'as new companies, including some of our customers, enter the markets.' |
| Chain position | Mature- and specialty-node wafer supplier: 2025 wafer sales were 41.4% communication, 30.6% consumer and 11.6% computer, with the AI tie indirect — 2.5D silicon interposers with deep-trench capacitors shipped to major customers for HPC, cloud computing and LLM applications. | Demand-side AI ad platform: it monetizes third-party mobile-game publishers' inventory and sells outcomes to advertisers, so it sits downstream of Apple/Google app distribution and Apple's IDFA identifier layer and upstream of publisher monetization. Meta and Google are simultaneously named competitors and disclosed clients. Single reportable segment after the June 2025 Apps divestiture. | Layer-4 chip supplier twice over: custom accelerators for hyperscalers and the Ethernet switching/optical silicon between every AI rack. |
| Products (share / barrier) |
|
|
|
| Long-horizon vote | -0.01 at weight 0.20 · swarm neutral Editorial prior, not backtested. | +0.11 at weight 0.20 · swarm neutral Editorial prior, not backtested. | +0.42 at weight 0.20 · swarm bullish Editorial prior, not backtested. |