United Microelectronics

Phase A — Understand the business

Lens 1 · Company Overview

Business model. UMC is a dedicated ("pure-play") wafer foundry: it manufactures integrated circuits to customers' designs and never competes with them on end products. Revenue is "primarily from the manufacture and sales of wafers fabricated at our foundries," plus a small slice of wafer-probe, mask-tooling, and subcontracted assembly/test . Pricing is per-wafer or per-die, reviewed quarterly; because costs are largely fixed, "fluctuations in average selling price historically have had a substantial impact on our margins" .

Scale (FY2025). Operating revenue NT$237,553M (US$7,573M), +2.3% YoY ; **3,870 thousand 12-inch-equivalent wafers** shipped (+12.3% vs 3,446k in 2024) . So volume grew double digits while revenue barely moved — the tension that defines UMC right now (see Lens 5).

Customers. Two archetypes: fabless design houses (MediaTek, Realtek, Novatek) and integrated device manufacturers (Texas Instruments, Intel) . Wafer sales split ~81% fabless / ~19% IDM in 2025 (fabless 78.3→84.3→80.9% over 2023-25) . Top-10 customers = 57.0% of revenue in 2025 `` — concentrated but not single-customer-dependent (contrast TSMC's ~25% Apple).

Contract structure. Historically spot/short-cycle, but the strategic pivot is toward multi-year supply agreements underwriting capacity ("our expansion projects will be supported by customers who have signed multi-year supply agreements" ``) — a take-or-pay-like de-risking of the capex it does commit.

What it actually is: not a growth-tech name but a fixed-cost utility for legacy and specialty silicon — power-management ICs, RF, display drivers, MCUs, automotive, IoT — whose earnings are a leveraged bet on capacity utilization and node/product mix.

Lens 2 · Supply Chain

Map: equipment & materials vendors → UMC fabs → fabless/IDM customers → OEM end-products (phones, PCs, autos, IoT, AI servers).

• Upstream inputs. Silicon wafers, photomasks, process chemicals/gases, and the capital-equipment chain (litho, etch, deposition). UMC subcontracts mask-tooling and assembly/test out. It is a price-taker on WFE (wafer-fab equipment) but is now flexing buy-side power: it notified all suppliers of a ≥15% price cut effective Jan 1 2026 to defend margins ahead of the pricing war ``.
• The fabs (chokepoints named). 12-inch (300mm): Fab 12A (Tainan, Taiwan — 136,279 wpm, the crown jewel), Fab 12i (Singapore — 57,207 wpm), Fab 12X / USCXM (Xiamen, China — 31,500 wpm, now 100%-owned), Fab 12M / USJC (Kuwana, Mie, Japan — 39,767 wpm, ex-Fujitsu, power/auto with DENSO). 8-inch (200mm): six Hsinchu fabs (8A/8C/8D/8E/8F/8S) + Fab 8N (Suzhou, China). Plus 6-inch Wavetek ``.
• Chokepoint / single-source risk. The 12-inch base is overwhelmingly Taiwan-concentrated (Fab 12A is the single largest node) — the geopolitical single point of failure that the Singapore Fab 12i P3 greenfield (30k wpm, 28/22nm, production H2 2026) and the US initiatives are explicitly designed to diversify ``.
• Downstream customers. Fabless (MediaTek/Realtek/Novatek) and IDMs (TI/Intel) → their chips land in smartphones, comms infrastructure, PCs, autos, and — the 2026 swing factor — power-management ICs for AI servers ``.
• US/geographic diversification (new). Intel collaboration (Intel's Ocotillo, Arizona fabs make the joint 12nm node — capital-light US footprint for UMC) and a Polar Semiconductor MOU (Dec 2025) for US-based 8-inch .

Chain verdict: UMC sits in the middle of the value chain with genuine buy-side leverage over materials suppliers but real sell-side exposure to Chinese foundry overcapacity at the mature end (Lens 3/13).

Lens 3 · Competitive Advantages (moats)

Where the moat is real:

• Specialty process breadth + IP library. RF-CMOS/RF-SOI, embedded flash (eNVM), BCD (power), eHV (display), and now silicon photonics — the "high-value mature" strategy. 16,700+ patents (7,490+ US), cross-licenses with IBM and Broadcom, TIPS AAA IP-management certification ``. This is a switching-cost moat: once a customer's design is qualified on UMC's specialty PDK, re-porting to a rival is costly and slow.
• Scale & geographic diversity. Among the largest capacities of any foundry, spread across four jurisdictions — a resilience/sourcing-choice pitch that matters more every year of US-China friction.
• Cycle time / service. UMC leans on turnaround speed, yields, and the "MyUMC" online supply-chain layer as service differentiation ``.

Where the moat is thin:

• No leading edge. 14nm and under = 0.0% of sales ``. UMC cannot serve the highest-value logic (AI accelerators, flagship SoCs) — that is TSMC's monopoly. UMC competes where processes are commoditizing.
• Bargaining power is asymmetric. Over materials suppliers: strong (the 15% cut). Over customers: weak-to-moderate — ASP fell 5.0% (2024) and 5.4% (2025) `` even as volumes rose, i.e. UMC could not hold price. That is the signature of a business without pricing power at the commodity end.
• The dangerous rival is structural, not a company. Chinese foundries (SMIC, Hua Hong) are flooding 28–90nm capacity, subsidized, and discounted Taiwan foundries up to 40% on mature nodes in late 2024 ``. UMC's countermove — abandon the low-value commodity fight, climb to specialty/high-value mature — is the correct strategy but a concession that the low end has no defensible moat.

Moat grade: narrow and specialty-dependent. Durable in RF/eNVM/BCD/photonics niches; absent in commodity logic.

Lens 4 · Segments

UMC does not report P&L by product segment; it discloses wafer-sales mix by process node and revenue by customer geography ``. Both from the 20-F:

Wafer sales by process technology (% of wafer sales):

``. 28nm-and-below = 36.8% of foundry revenue in 2025 (up from 30.7% in 2023) — this is the mix-up story and the single most important segment trend. 28nm is where UMC has scale, where TSMC may cede share, and where the specialty growth sits.

Revenue by customer HQ geography (%):

``. US-headquartered demand is shrinking as a share (26.6→22.0%) while Taiwan rises — worth watching against the "UMC as US-diversification play" narrative.

Trend cause: the mix shift to 28nm + the 2025 volume recovery (utilization 68.7→75.2%) lifted revenue only 2.3% because ASP fell 5.4% and NTD appreciated 2.9% ``. Volume and mix up; price and FX down — the four forces that net to near-flat 2025 revenue but a clear inflection into 2026.

---

Phase B — Measure performance

Lens 5 · Earnings Result

FY2025 (the 20-F print) ``:

• Revenue NT$237,553M (US$7,573M), +2.3% YoY.
• Gross profit NT$68,906M (US$2,197M); gross margin 29.0% — down from 32.6% (2024) and 34.9% (2023). Three straight years of compression.
• Operating income NT$43,949M (US$1,401M); operating margin 18.5% (was 22.2% / 26.0%).
• Net income to parent NT$40,360M (US$1,287M), −17.3% YoY; net margin 16.9%.
• Basic EPS NT$3.30 (US$0.53/ADS); diluted NT$3.27 ``.
• Utilization 75.2% (up from 68.7%) `` — the demand recovery is visible in the fab, but margin still fell because ASP −5.4% and depreciation surged.
• The margin killer is depreciation: NT$56,427M in 2025 vs NT$45,472M (2024) vs NT$37,758M (2023) — the P&L is now absorbing the cost of the 2022-24 capex wave (see Lens 9). Fixed costs are **70.8% of manufacturing cost**, so margin lives and dies on utilization × ASP.

The 2026 inflection (post-20-F, all ``):

• Q1 2026 (reported Apr 29): revenue NT$61.04B (+5.5% YoY, −1.2% QoQ), GM 29.2% (+2.5pp YoY), utilization 79%, net income NT$16.17B, EPS NT$1.29 — vs NT$0.85 consensus, a +48% beat and more than double the prior-year quarter . UMC also **launched a new share buyback** in Q1 — a capital-allocation shift (it bought back zero shares 2023-25 ``).
• Q2 2026: revenue NT$68.73B (+12.6% QoQ, +17.0% YoY) — the highest since Q4 2022, ending a 15-quarter drought ``. June alone NT$23.1B (+22.85% YoY), a near-4-year monthly high. H1 2026 revenue NT$129.77B (+11.28% YoY).
• Drivers: communications recovery + power-management ICs for AI applications + stable computing/consumer/industrial ``.
• Market reaction: the stock is now sentiment-driven — it fell ~6% on the strong Q2 revenue (sell-the-news) and −8.78% on July 7 in a sector-wide AI/CPI selloff , yet is **+216% YTD** .

Read: the operational trough is behind UMC. 2026 is a real recovery year (utilization ~79%, revenue breaking multi-year highs, margins stabilizing near 30%). The question the tape is asking is not "is it recovering" (yes) but "how much recovery is already in the price" (a lot — Lens 7/8/12).

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the shelf (transcripts/ empty); this is ``-derived and thinner than a transcript-grounded read would be.

• Tone shift 2025 → 2026: From cautious/defensive (guiding to price discipline, flagging Chinese overcapacity, cutting supplier costs) to constructive. On the Q1 2026 call management guided high-single-digit % QoQ wafer-shipment growth for Q2, GM near 30%, continued pricing discipline, and "robust demand in communication, consumer, and AI-related segments" ``.
• Recurring phrases (2026): "specialty," "high-value mature," "22/28nm," "power management for AI," "pricing discipline," "geographic diversification," "12nm with Intel," "silicon photonics."
• What they stopped saying: the 2023-24 language of "inventory correction," "sluggish demand," and "utilization headwinds" has receded.
• New in 2026: explicit AI-adjacency framing (PMIC for AI servers, silicon photonics/CPO for AI datacenter I/O) — management is actively repositioning the equity story from "cyclical value foundry" to "AI-infrastructure participant," which is precisely what the multiple is now discounting.

Sentiment trend: improving and increasingly AI-narrative-forward. Caveat: without transcripts on the shelf this is the lens with the weakest grounding — a refresh should ingest the Q1/Q2 2026 calls.

Lens 7 · Comps

The whole mature/China-foundry complex has re-rated to extraordinary multiples in the 2026 AI cycle. Multiples are `` (dates/sources noted) or n/a; no multiple below is fabricated.

**UMC-specific derived multiples :** roughly net cash (~NT$45B / ~US$1.4B: cash NT$110,660M vs bonds NT$50,228M + LT bank loans ~NT$15.3B ), so EV ≈ $61.1B. On FY2025 revenue $7.573B and operating income $1.401B: EV/Sales ≈ 8.1x, EV/EBIT ≈ 43.6x . ROE FY2025 ≈ **~10–11%** — roughly a third of TSMC's historical ~30%.

The one line that matters: UMC — a mature-node foundry with no leading edge, GM 29%, ROE ~10% — trades at ~38.5x trailing, higher than TSMC's ~29x (the leading-edge monopoly, 72% share, GM ~59%, ROE ~30%). You are paying more per dollar of UMC's shrinking-then-recovering earnings than per dollar of TSMC's compounding earnings. On forward 2026 numbers the gap narrows (UMC ~26x `` vs TSMC low-20s) but the anomaly stands: UMC is priced at least as richly as the best asset in the industry.

Lens 8 · Stock-Price Catalysts (what moves >5%)

52-week range $6.56 → $28.96, currently $24.86, +216% YTD 2026 ``. The 2026 melt-up is the dominant fact. Pattern of >5% moves:

• The 2026 re-rating (multi-hundred-%): chip-cycle recovery + AI narrative. Specific legs: Q1 2026 +48% EPS beat; the Wedbush "2026's biggest overlooked winner" call including the thesis that if TSMC reallocates 28nm wafers, UMC is the largest direct beneficiary ; the imec silicon-photonics/CPO license (Dec 2025); Intel-12nm "on schedule for 2027"; reports of **planned H2 2026 wafer price hikes** .
• Jun 20–21 2026: +10.66% (2-day) ``.
• July 7 2026: −8.78% — sector semiconductor selloff on CPI/AI concerns ``.
• Q2 revenue print: ~−6% despite record revenue (sell-the-news) ``.
• Historical (pre-2026): foundry-cycle inflections, utilization/ASP data points, Chinese-foundry pricing headlines, and the 2018-20 Micron/DOJ overhang (Lens 10) were the swing factors — plus the dividend, when UMC was a yield stock.

What the market reacts to for this name has changed: it used to trade on utilization, ASP, and dividend (a value/cyclical). In 2026 it trades on AI-narrative flow and momentum — which cuts both ways and explains the violence of the 52-week range.

---

Phase C — Judge people & books

Lens 9 · Management

• Leadership (just changed). Jason Wang became sole CEO on Feb 25, 2026, ending the Wang + SC Chien co-president model in place since 2017; Ming Hsu became President & COO; SC Chien exited to chair Unimicron (a UMC-invested substrate company) . Chairman **Stan Hung** (34 yrs, ex-CFO, also Chief Strategic Officer); CFO **Chitung Liu** (25 yrs, ex-UBS MD) . The streamlining is framed as faster decision-making/execution — a governance positive, though it concentrates power as the equity story turns aggressive.
• Track record. The Wang/Chien duo drove UMC's decade-long turnaround: the 2018 decision to stop chasing the leading edge (walking away from 14nm→10nm) and reallocate to specialty/mature is the reason UMC has ~20% operating margins and net cash today rather than the balance-sheet stress of a sub-scale advanced-node also-ran. That is a genuinely good capital-discipline call, quantified by three years of ~18–26% operating margin ``.
• Skin in the game. Directors & officers own just 6.62% of shares — professional managers, not founder-owners. UMC is **widely held** (largest holders are Taiwanese dividend ETFs: Capital Tip 5.17%, Cathay 4.31%, Yuanta 3.34%) and explicitly **not controlled by any person, corporation, or government**.
• Capital allocation. Historically dividends over buybacks — but the dividend is declining with earnings: NT$3.60 (2023) → 3.00 (2024) → 2.85 (2025) → 2.60 proposed (2026) ; FY2025 payout ~79% of EPS . Capex was cut hard: NT$94.1B (2023) → 91.1B (2024) → 50.6B (2025) — the build cycle is over, which is *why* 2025 FCF was strong (~NT$52.1B / ~US$1.66B). And in Q1 2026 UMC launched a share buyback `` — a new lever, sensible while the cash pile (US$3.5B) is idle, ironic while the stock is +216%.
• Red flags (governance). Notable circular cross-holdings typical of Taiwan Inc: UMC owns 17.99% of Silicon Integrated Systems (SIS), whose representative is CEO Jason Wang, and SIS owns 2.12% of UMC; UMC owns 36.49% of Hsun Chieh Investment, which owns 3.51% of UMC and is represented on the board by COO Ming Hsu . Related-party foundry sales to SIS + Faraday were NT$4,387M (US$140M) in 2025 . Common in Taiwan, arms-length per the filing, but a web an outside investor should see.

Management grade: capable, disciplined, professional (not owner-operators). The turnaround thesis is real; the incentive alignment is modest.

Lens 10 · Forensic Red Flags

Grounded in the 20-F `` + regulatory/regulatory-findings.md + web.

• Earnings quality: clean. Operating cash flow (NT$99,864M) exceeds net income handily, driven by the large non-cash depreciation add-back (NT$59,259M D&A) — the opposite of the classic red flag (earnings running ahead of cash). No sign of receivables/inventory outrunning revenue in the disclosed data; expected-credit-loss provisions are negligible (NT$2M).
• The real accounting risk is depreciation timing, not manipulation. Margins are hostage to a depreciation wave from the 2022-24 capex; non-GAAP flattery is not the issue (UMC reports IFRS), but reported GM understates cash margins because ~71% of COGS is fixed/depreciation ``. Watch for any switch in useful-life assumptions that would flatter margins as the AI story is sold.
• Related-party web (Lens 9) — SIS/Hsun Chieh/Faraday cross-holdings and NT$4.39B related-party foundry sales ``. Disclosed, arms-length per management, immaterial to the P&L, but a standing governance item.
• Bargain-purchase gain on gaining significant influence over SIS flowed through non-operating income `` — a non-recurring, non-cash item to normalize out of any "earnings quality" read.

Regulatory findings (required):

• SEC: regulatory-findings.md reports 0 Litigation Releases and 0 AAERs naming UMC in the 2021-2026 window (EDGAR EFTS) ``.
• 10-K/20-F Item 8 (Legal Proceedings), the company's own disclosure: "these disputes have not had any material adverse impact… nor are there any other material litigation or non-litigation matters required to be disclosed" `` — i.e. no material pending litigation disclosed.
• Non-SEC enforcement (web search): the one material historical item — UMC pleaded guilty (Oct 2020) to one count of receiving/possessing a stolen trade secret (Micron DRAM technology, to benefit Chinese SOE Fujian Jinhua) and paid a US$60M fine; DOJ dropped the economic-espionage and broader conspiracy counts and the $400M–$8.75B damages claims, and UMC assisted DOJ against Fujian Jinhua . **Resolved and now historical** (not in current Item 3), but a forensic analyst must weight it: it is a real IP-integrity/governance blemish with a China nexus, directly relevant to the "UMC as trusted US-diversification foundry" narrative it is now selling. The **DDoS attack on UMC's website (Oct 2024)** caused no data loss .

Forensic verdict: books are clean and cash-backed; the single historical black mark (Micron/Jinhua) is resolved but reputationally live given the US pivot.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

Bottom-up from FY2025 actuals + 2026 prints + guidance. Output ``; no forecast.ts logged (watchlist/unattended rule). Fiscal years = calendar years.

Anchor actuals: FY2025 basic EPS NT$3.30 ; Q1 2026 EPS **NT$1.29** ; H1 2026 revenue +11.3% YoY, Q2 record NT$68.73B ; guidance GM ~30%, high-SD% Q2 shipment growth .

• Base (most likely). 2026 is a clean recovery: utilization ~78-80%, GM ~29-30%, ASP flattening as H2 price hikes offset China pressure. FY2026 revenue ~NT$270-280B (+14-18%). Quarterly EPS ~NT$1.29 → ~1.55 → ~1.6 → ~1.6 → FY2026 EPS ≈ NT$6.0/sh (~US$0.95/ADS) ``. FY2027 ≈ NT$6.5 (12nm/Intel begins to contribute, specialty grows, China caps ASP). FY2028 ≈ NT$7.0-7.5 (silicon photonics + 12nm ramp). → forward P/E on 2026E ≈ 26x at $24.86.
• Bull. AI-specialty super-cycle + TSMC 28nm spillover + pricing power returns + photonics revenue by 2028: FY2026 ≈ NT$6.5, FY2028 ≈ NT$9+ ``.
• Bear. 2026 is a restock peak; China overcapacity re-crushes mature ASP in 2027; AI-PMIC demand normalizes: FY2026 ≈ NT$5.5 then flat-to-down NT$5.0-5.5 FY2027-28 ``.

Even the bull FY2028 EPS (~NT$9/sh = ~US$1.43/ADS) at today's $24.86 is ~17x — so today's price already prices the bull case three years early. The base case (~26x forward 2026) is rich for a mature-node cyclical whose 10-year-normal multiple is ~12x.

Lens 12 · Bull vs Bear

Bull case (narrative). UMC is the mispriced pick-and-shovel of the AI-and-recovery cycle. The chip cycle has turned (Q2 broke a 15-quarter revenue drought); utilization is back to ~79% and rising; the mix is climbing to high-value 28nm/specialty (36.8% and growing). Three optionality legs the market is starting to pay for: (1) silicon photonics / CPO — the imec iSiPP300 license puts UMC into 12-inch PICs for 800G/1.6T optics into AI datacenters, a genuinely new TAM as copper interconnect hits its limits; (2) Intel 12nm — a capital-light way to add a 12nm node and a US footprint, on schedule for 2027; (3) TSMC 28nm reallocation — if TSMC vacates 28nm for leading edge, UMC is the #1 beneficiary (Wedbush). Net cash, ~US$1.7B FCF, and a fresh buyback. If AI keeps pulling mature/specialty capacity tight, 2026 earnings re-rate the whole story.

Bear case (2-3 permanent-impairment risks). (1) Chinese overcapacity is structural, not cyclical — SMIC + Hua Hong are adding subsidized 28-90nm capacity faster than demand; UMC's own ≥15% supplier price cut and three years of ASP erosion (−5.0%, −5.4%) say it has no pricing power at the commodity end, and "high-value mature" is a smaller pond than the equity now assumes. (2) The AI legs are optionality, not earnings — photonics is at risk production (2026-27), 12nm is 2027 production, and neither moves 2026-27 EPS; the market is capitalizing them as if they already ship. (3) The multiple itself — +216% YTD to ~38.5x trailing, above every sell-side target, on a business with ~10% ROE. Pre-mortem (18 months out, thesis broke): 2026 was a restock peak; H2-2026/2027 mature ASP rolls over under Chinese supply; the AI-PMIC pop normalizes; photonics/12nm slip or come in low-margin; the stock mean-reverts toward the $11-16 analyst targets — a 35-55% drawdown with earnings still up YoY (a pure multiple de-rate). Contrarian view the market is refusing to see: UMC's 2026 is a real cyclical recovery, but the price has decoupled from a mature-node foundry's normalized earnings power — the tape is trading a TSMC-grade multiple onto a business that deliberately chose not to be TSMC.

Are multiples too high? Yes, on any through-cycle or peer-normalized basis — though within the (insane) 2026 mature/China-foundry peer set (SMIC 84x, Tower ~100x, Hua Hong 34-92x, GF 48x) UMC's 38.5x is not the most extreme.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull:

• Where revenue concentrates and what breaks it. ~81% fabless, top-10 = 57% ``, heavy in communications/consumer — exactly the end-markets most exposed to a 2027 inventory correction after a 2026 restock. A single large fabless customer trimming orders swings utilization, and utilization is the margin.
• The moat is weaker than bulls think. UMC's differentiation is "specialty + service + geography." But SMIC/Hua Hong are climbing the same specialty ladder (BCD, RF, PMIC) with state backing and are raising prices in 2026 precisely because they, not UMC, increasingly set the marginal mature-node price. UMC took a −5.4% ASP hit while volumes rose — that is not a moat, that is a price-taker.
• Most dangerous competitor bulls underestimate: not TSMC (different league) but Hua Hong / SMIC — subsidized capacity that can run at cash-cost through a downturn and structurally cap UMC's "high-value mature" pricing.
• Worst capital-allocation / governance: the Taiwan cross-holding web (UMC↔SIS with the CEO as SIS's rep; UMC↔Hsun Chieh with the COO as its rep) , and — reputationally — the **2020 Micron/Fujian-Jinhua guilty plea** , which undercuts the "trusted Western-diversification foundry" pitch the Intel/Polar deals depend on.
• What must hold for today's price: a durable AI-driven mature/specialty super-cycle and the AI optionality (photonics, 12nm) converting to real margin by 2027-28 and Chinese overcapacity not re-crushing ASP. If growth disappoints ~20-30%, EPS lands near the bear NT$5.0-5.5 and a 38.5x multiple is indefensible — ~50%+ downside on a de-rate alone, with sell-side already at $11-16.
• Single scenario that permanently impairs: a Taiwan Strait geopolitical shock — the 12-inch base (Fab 12A) is Taiwan-concentrated; the Singapore/US diversification is years and single-digit-% of capacity away.

Short verdict: the company is not broken — the price is the trade. This is a valuation/expectations short, not a fraud or a structural-collapse short.

Lens 14 · Management Questions (ordered by information value)

1. Of your 2026 revenue upside, how much is restock/pull-forward vs. structural — and what is your evidence it is not a 2027 inventory-correction setup?
2. On 28nm ASP: with SMIC/Hua Hong adding capacity, what is your realistic 2026-28 pricing path, and where do you actually have pricing power vs. where are you a price-taker?
3. Silicon photonics (imec iSiPP300): what revenue and gross margin do you underwrite for 2027 and 2028, and how many customers are past risk-production to committed volume?
4. Intel 12nm: what economics accrue to UMC (vs. Intel) — is this a royalty/licensing stream, a capacity call, or a go-to-market fee — and what does it add to EPS in 2027-28?
5. If TSMC reallocates 28nm, quantify the share you can actually win given your capacity and qualified customers — is the Wedbush thesis real or hopeful?
6. Your stock is +216% YTD and above every analyst target — how does that change your buyback logic; would you rather return cash or hoard it for the next down-cycle?
7. The dividend has fallen four years running; is UMC still a yield story, or are you formally reallocating to buybacks/growth?
8. Capex fell from ~NT$94B to ~NT$50B; what is the normalized 2026-28 capex, and does the depreciation wave peak in 2026 (i.e. does reported GM inflect up)?
9. What share of 2026 "AI-related" PMIC demand is durable design-win vs. spot allocation that reverses when GPU supply loosens?
10. On the Micron/Fujian-Jinhua history: how do you satisfy Intel, Polar, and US customers on IP integrity and export-control compliance?
11. Fab 12i P3 Singapore (H2 2026): who are the anchor multi-year-supply customers underwriting it, and at what utilization does it earn its cost of capital?
12. Your China fabs (Xiamen, Suzhou): how do you manage US export-control and customer-diversification risk as the US footprint grows?
13. What is the normalized through-cycle ROE you are managing to, and how do you get from ~10% back toward the high-teens?
14. What ended the co-president model, and what decisions will now be made faster or differently under a single CEO?
15. Which single business line would you exit or divest in the next three years, and why?

---