MediaTek

Phase A — Understand the business

Lens 1 · Company Overview

MediaTek is the world's largest merchant supplier of smartphone system-on-chips by unit volume, and — more precisely — a fabless Arm-SoC platform company that sells the "good-enough-to-flagship" alternative to Qualcomm across every Arm-based consumer device. It designs; TSMC manufactures; MediaTek sells reference platforms to OEMs who want a turnkey chip-plus-software stack rather than to build silicon themselves.

The business runs in three reported revenue groups (renamed/re-cut over 2024–25 into this structure) [primary: transcripts/2025-Q4-prepared-remarks.txt]:

1. Mobile Phone — Dimensity smartphone SoCs (flagship 9-series, premium 8-series, mainstream/entry). ~59% of Q4 2025 revenue; >US$10B for FY2025, +8% YoY, a record [primary: 4q25-press-release.txt / prepared-remarks].
2. Smart Edge Platforms — everything non-phone at the edge: tablets, Chromebooks, smart TVs, Wi-Fi/broadband connectivity (Wi-Fi 7, 10G-PON), 5G modems, ASICs, automotive, and the new NVIDIA GB10 computing project. ~37% of Q4 2025 revenue; +21% YoY in FY2025 USD [primary: prepared-remarks].
3. Power IC — power-management chips. ~5% of revenue, roughly flat-to-down, a supporting attach product [primary: prepared-remarks].

FY2025 total revenue was NT$595,966M (US$19.1B), +12.3% YoY in NT$ / +15.6% in USD — a record year [primary: 4q25-press-release.txt]. Net income NT$106,118M, EPS NT$66.16 [primary: 4q25-press-release.txt].

Customers: the Chinese Android ecosystem is the demand base — Xiaomi, Oppo, Vivo, Honor, Transsion (Africa/emerging) — plus Samsung at the premium tier (Galaxy S24 FE, S25 FE design wins for Dimensity 9400). New non-phone customers now matter directionally: NVIDIA (GB10 co-developed for DGX Spark), Google (custom TPU inference ASIC), and DENSO (co-developed ADAS chips) [primary: prepared-remarks].

Contract structure: merchant silicon — per-unit, design-win-driven, no take-or-pay, no recurring/subscription revenue. Revenue is a function of (units shipped × ASP), so it is cyclical with handset demand and mix-sensitive. The ASIC business is different in character — large, multi-year, single-customer NRE-plus-volume programs — which is exactly why the market is treating it as a re-rating event (Lens 8, 12).

Lens 2 · Supply Chain

MediaTek sits in the middle of the Arm-mobile value chain, and its single most important structural fact is that it owns no fabs — it is entirely dependent on TSMC for leading-edge manufacturing.

Upstream (inputs MediaTek depends on):

• IP core / ISA: Arm Holdings — Arm CPU/GPU cores and the Armv9 ISA under license + per-chip royalty. MediaTek is one of Arm's largest royalty payers. (Arm's own move up the value chain into reference designs / potential first-party chips is a structural threat — Lens 13.)
• In-house GPU alternative: MediaTek also uses Arm Mali/Immortalis GPUs (unlike Qualcomm's owned Adreno) — a moat gap vs. Qualcomm, which owns its GPU IP.
• Foundry: TSMC — the binding chokepoint. Flagship Dimensity + the Google/NVIDIA ASICs are on TSMC N3/N2. MediaTek was "one of few companies completing 2nm tape-out at TSMC" [primary: prepared-remarks]. CoWoS/advanced-packaging capacity at TSMC is the AI-ASIC bottleneck: MediaTek reportedly requested a ~7× CoWoS increase for the Google TPU program.
• Memory: HBM (for ASICs), plus DRAM/NAND indirectly via its OEM customers' BOMs. Rising DRAM/NAND prices are a 2026 demand headwind on the phone side — management explicitly flagged memory/BOM cost inflation compressing smartphone end-demand [primary: prepared-remarks; web: wccftech/digitimes, 2026].
• Packaging/test (OSAT): ASE, Amkor and TSMC's own advanced packaging.

Midstream: MediaTek (design, platform integration, reference software, modem/connectivity IP).

Downstream (who buys):

• Handset OEMs: Xiaomi, Oppo, Vivo, Honor, Transsion, Samsung (premium tier).
• Non-phone: TV makers, broadband/telco equipment, Chromebook/tablet OEMs, NVIDIA (GB10), Google (TPU inference), DENSO/autos.

Chokepoints & single-source dependencies:

• TSMC is a single point of failure on leading-edge and on CoWoS. MediaTek reaffirmed TSMC as its key long-term foundry partner even as Samsung Foundry courted it. This is a two-edged dependency: security of a best-in-class partner, but zero negotiating leverage on wafer/packaging price in a supply-constrained AI cycle. Management said outright it will "secure the capacity" and "adjust pricing to reflect rising supply-chain costs" [primary: prepared-remarks] — i.e., it is a price-taker upstream and is passing cost through downstream.
• Arm is a single-source for the ISA — no realistic RISC-V substitution at the flagship tier near-term.
• Concentration risk flips on the ASIC side: the Google TPU program is a single-customer dependency where Google is deliberately dual-sourcing across Broadcom + MediaTek (Lens 13) — the opposite of MediaTek's diversified phone book.

Names, not generics: Arm, TSMC, CoWoS/ASE/Amkor upstream; Xiaomi/Oppo/Vivo/Honor/Transsion/Samsung + NVIDIA/Google/DENSO downstream. The chain is real and specific.

Lens 3 · Competitive Advantages (moats)

MediaTek's moat is real but narrow, and it is a cost/scale/execution moat, not a pricing-power moat.

What protects it:

• Scale in Arm mobile SoCs. It powers >2 billion devices/year and is #1 in smartphone SoC units. That volume amortizes enormous R&D (FY2025 R&D NT$179.6B opex, ~30% of revenue, R&D alone ~26% of Q4 revenue [primary: 4q25-press-release.txt]) across more units than anyone except Qualcomm. A sub-scale entrant cannot fund a competitive flagship modem+SoC+connectivity roadmap.
• Integrated platform + software enablement. MediaTek sells a turnkey reference design (SoC + modem + Wi-Fi + BSP/software), which is exactly what cost-sensitive Chinese/emerging-market OEMs want. Switching from MediaTek to Qualcomm on a given SKU is a real re-engineering + re-qualification cost — moderate switching costs, not enormous.
• Connectivity IP. Wi-Fi 7 (revenue tripled in 2025), 5G modem (doubled), 10G-PON leadership, and now Wi-Fi 8 (first-mover, unveiled CES) [primary: prepared-remarks]. This is a genuine, defensible IP asset — connectivity revenue exceeded US$3B in 2025 [primary: prepared-remarks].
• Emerging-markets lock. MediaTek owns the entry/mid Android world (Transsion in Africa, mass-market India/SE-Asia). Qualcomm largely ceded this.

Where the moat is weak (the honest read):

• No first-party CPU or GPU. MediaTek rides Arm cores + Arm Mali GPU. Qualcomm owns Oryon (ex-Nuvia) CPU and Adreno GPU. At the very top of the flagship tier, MediaTek is IP-dependent in a way Qualcomm is not — its differentiation is integration/power-efficiency/price, not silicon-architecture ownership.
• Bargaining power is asymmetric — MediaTek needs its counterparties more than they need it. Upstream: TSMC and Arm hold the leverage. Downstream: it competes on price against Qualcomm, so OEMs play them off. The one place MediaTek gains leverage is the Arm↔Qualcomm dispute — MediaTek is the neutral beneficiary if Qualcomm's Arm license is ever destabilized (though Qualcomm won that suit in 2025, defusing the near-term upside).
• The ASIC "moat" does not exist yet. MediaTek won the Google TPU inference variant (Zebrafish/v8i) on cost/PPA and as a deliberate second source to pressure Broadcom's pricing. That is the opposite of a moat — Google structurally wants MediaTek to be replaceable. The design/IP capability MediaTek is building (400G SerDes, CPO, 3.5D packaging, custom HBM, IVR — [primary: prepared-remarks]) could become a moat over 2–3 program generations, but as of now it is a challenger renting a seat at Broadcom's table.

Verdict on moat: durable in mobile (scale + connectivity + emerging markets), contested at the flagship top-end (no owned CPU/GPU), and absent-but-building in data-center ASIC — which is precisely the segment the valuation is now pricing.

Lens 4 · Segments

No segments.csv on disk — all figures below are from the primary 4Q25 prepared remarks and press release, plus company FY color. Provenance is [primary] where the company stated it; `` where I compute a mix.

[primary: transcripts/2025-Q4-prepared-remarks.txt; 4q25-press-release.txt]

Within the groups (the parts that matter):

• Flagship SoCs (Dimensity 9-series): US$3.0B in FY2025, the fastest-growing slice of Mobile — this is MediaTek finally monetizing the premium tier [primary: prepared-remarks].
• Connectivity (inside Smart Edge): >US$3.0B in FY2025, Wi-Fi 7 tripled, 5G modem doubled [primary: prepared-remarks].
• Computing (tablets + Chromebooks + NVIDIA GB10, inside Smart Edge): grew >80% to ~US$1.0B in FY2025 [primary: prepared-remarks].
• Data-center ASIC: essentially pre-revenue in 2025; guided >US$1B in 2026 at the Feb call, then doubled to ~US$2B (Q4-2026-weighted) by the April call [primary: prepared-remarks (Feb); web: Investing.com/TrendForce, 2026-04].

Trend and cause:

• The structural story is a mix shift — MediaTek is decelerating in commodity phones (entry/mid softening, Q1 2026 Mobile −15% YoY, Mobile fell to 49% of revenue in Q1 2026 ) while accelerating in premium mobile, connectivity, computing, auto, and now data-center ASIC. Smart Edge overtaking Mobile as the swing factor (46% vs 49% in Q1 2026) is the tell.
• Geography: not separately disclosed in the primary docs — MediaTek does not break out clean geographic revenue. Directionally, China + emerging Asia dominate the phone base; the new ASIC/NVIDIA/computing revenue is US-hyperscaler-facing. n/a — geographic segment split not disclosed in sourced filings.

---

Phase B — Measure performance

Lens 5 · Earnings Result (Q4 2025, reported 2026-02-04 — the latest fully-reported quarter with primary docs; Q1 2026 actuals summarized after)

The Q4 print (all TIFRS, [primary: 4q25-press-release.txt]):

• Revenue: NT$150,188M, +5.7% QoQ, +8.8% YoY — came in at the high end of guidance, helped by a favorable NT$/US$ FX (31.1 realized vs 30.6 guided) and stronger Smart Edge demand [primary: prepared-remarks].
• Gross margin: 46.1%, −0.4pp QoQ, −2.4pp YoY. FY2025 GM 47.5%, −2.1pp YoY — the multi-year margin story is a downtrend, driven by mix and rising supply-chain cost.
• Operating income: NT$21,850M, −1.5% QoQ, +2.0% YoY. Operating margin 14.5% (down from 15.6% QoQ and 15.5% YoY). FY2025 op margin 17.4%, −1.9pp YoY.
• Opex: NT$47,431M (31.6% of revenue), up on higher R&D (R&D NT$39,248M, 26.1% of revenue) — MediaTek is spending into the data-center pivot, and it shows in the opex line.
• Net income: NT$23,074M, −9.3% QoQ, −3.6% YoY. EPS NT$14.39 (vs NT$15.84 prior Q, NT$14.95 year-ago). Net margin 15.4%.
• Beat/miss: revenue at the high end of the guided range (a modest beat); earnings quality soft — the QoQ EPS decline was driven by opex, and YoY by lower gross margin. This was not a clean beat — it was a top-line beat with margin/EPS deterioration underneath.

Balance-sheet flags (all healthy [primary: 4q25-press-release.txt]):

• Cash & current financial assets: NT$247,709M (~US$8B), 33.3% of total assets — up sharply from NT$210.6B prior Q. Net-cash balance sheet, no leverage concern.
• Operating cash flow: NT$64,520M in the quarter (vs NT$39.2B prior Q) — strong cash conversion, well above net income.
• Receivables: 39 days (down from 44, up from 30 year-ago) — no red flag, seasonal.
• Inventory: 70 days (68 prior Q, 73 year-ago) — normal, no build-up.

Guidance (Q1 2026) [primary: prepared-remarks]:

• Revenue NT$141.2B–150.2B (flat to −6% QoQ, −2% to −8% YoY) — a soft, seasonal, memory-cost-pressured guide on the phone side, partially offset by Smart Edge recovery.
• Gross margin 46% ±1.5pp; opex ratio 31% ±2pp. Management is explicitly guiding to hold ~46% GM for 2026 via "disciplined pricing and strategic capacity allocation."

Q1 2026 actuals (reported 2026-04-30): Revenue NT$149.2B (−2.7% YoY, above ~NT$146.6B consensus); net income NT$24.4B; GM 46.3%. Mobile 49% of revenue (−15% YoY); Smart Edge 46% (+13% YoY). The headline was not the print — it was that MediaTek doubled its 2026 ASIC target to ~US$2B and pulled its US$70–80B 2027 ASIC-TAM forward from 2028, targeting 10–15% ASIC share (≈US$7–12B of 2027 revenue).

Market reaction — this is the whole story. The stock did not react to the earnings print; it reacted to the ASIC guidance. MediaTek went limit-up for consecutive sessions, from a year-end-2025 close of ~NT$1,430 to ~NT$3,155 (+120%) in early 2026, then on to ~NT$4,195 by 2026-07-06. What the tape reveals: the market is pricing MediaTek as an AI-ASIC name, not a smartphone name — the phone business is now the floor, the ASIC ramp is the story.

Lens 6 · Earnings Calls (sentiment trend)

Only one full transcript is on the shelf (4Q25 prepared remarks); the tone-shift below is reconstructed from that primary doc + web summaries of the 1Q–3Q 2025 calls and the 1Q26 call.

What management is focused on (4Q25 → 1Q26): the pivot narrative is total. Rick Tsai's 4Q25 remarks lead with "ubiquitous AI," then spend the bulk of the script on data-center ASIC, advanced packaging (3.5D, CoWoS, custom HBM, CPO, 400G SerDes, IVR), NVIDIA GB10, DENSO ADAS, and 2nm tape-out — the phone business is discussed almost defensively ("we expect overall smartphone end demand to be negatively impacted" by memory/BOM cost) [primary: prepared-remarks].

Tone shift over the last ~4 calls:

• 1Q–2Q 2025: "strong revenue growth amid margin pressures" — the recurring frame was defending gross margin against competition and cost. Cautious.
• 3Q 2025: first real emphasis on "data center expansion" alongside AI-smartphone / Smart Edge strength — the ASIC option starts appearing.
• 4Q 2025 (Feb 2026): confident, expansionary — "very confident in >US$1B ASIC in 2026, multiple-billion in 2027, follow-on project revenue in 2028." Records everywhere ("another year of record revenue"). [primary: prepared-remarks]
• 1Q 2026 (Apr 2026): the pivot goes vertical — ASIC target doubled to US$2B, TAM pulled forward, executives "urge investors to take a long view". That last phrase is telling: management is now managing expectations downward on a stock that has tripled — a subtle tone shift from selling growth to cooling euphoria.

Recurring phrases: "ubiquitous AI," "edge AI and cloud AI," "record revenue," "disciplined pricing / strategic capacity allocation," "trusted partner for data-center customers."

What they stopped saying: the pure "we are the smartphone SoC leader" framing has receded — smartphone leadership is now stated as a foundation ("solid foundation") for the AI story, not the story itself. They also went quiet on aggressive margin-defense language relative to early-2025 — replaced by "adjust pricing to reflect rising supply-chain costs" (cost pass-through, not margin expansion).

Lens 7 · Comps

Peer set: the merchant-SoC + custom-ASIC complex. Multiples are `` with source/date; where a specific multiple was not sourced, it is n/a rather than fabricated.

The comps read (the sharp point): MediaTek at ~63x P/E is priced above Broadcom (<30x), Marvell (~28x), and dramatically above its nearest structural twin Qualcomm (~14x) — despite Qualcomm owning more of its own IP (Oryon CPU, Adreno GPU) and having a larger, more diversified non-handset franchise (auto, IoT, PC). The entire ~4–5x P/E premium over Qualcomm, and the premium over the actual ASIC incumbents, is the market capitalizing an ASIC business that is ~US$2B of 2026 revenue on a ~US$21B revenue base. For the multiple to be right, you have to believe Goldman's US$12.3B 2027 ASIC / 39%-of-revenue path — i.e., you are paying an ASIC-pure-play multiple for a company that is 90%+ still a cyclical phone/edge chipmaker. That is the crux of the bear case (Lens 12/13).

Lens 8 · Stock-Price Catalysts (moves >5%, last ~5 years)

Mostly ``. The pattern is unusually clean for this name.

• 2026-Q1/Q2 — the ASIC re-rating (the defining move): consecutive limit-up (+10%/day) sessions, ~NT$1,430 → ~NT$3,155 (+120%), then to ~NT$4,195 by July. Trigger: confirmation of the Google TPU inference win + doubling the 2026 ASIC target to US$2B. Foreign brokers escalated targets — Goldman NT$5,000, Macquarie NT$10,000.
• 2025-late — NVIDIA halo: MediaTek surged with NVIDIA earnings and Taiwan-market AI rallies; the GB10/DGX Spark co-development gave it direct NVIDIA-ecosystem exposure.
• 2024-Q4 — Arm↔Qualcomm dispute: MediaTek rallied on the read-through that it benefits if Qualcomm's Arm license is threatened.
• Recurring — smartphone-cycle prints: historically the stock moved on China-Android demand, flagship Dimensity ramps (9400/9500), and gross-margin surprises — the "old" MediaTek reaction function.

What the pattern reveals: for most of its history the market reacted to smartphone-cycle demand and margin. In 2026 the reaction function flipped entirely to a single variable: data-center ASIC guidance and the Google/NVIDIA relationships. The stock is now a high-beta AI-infrastructure derivative that happens to have a large phone business attached — it trades on ASIC headlines, not handset units. This makes it catalyst-fragile in both directions: any Google multi-sourcing headline (which already surfaced — Digitimes, 2026-06) or ASIC-timeline slip is now a first-order risk to the whole quote.

---

Phase C — Judge people & books

Lens 9 · Management

CEO — Dr. Rick Tsai (Vice Chairman & CEO since June 2017). This is an unusually strong operator with a directly relevant pedigree.

• Track record: ran TSMC as President & CEO 2005–2009 — including steering TSMC through the 2008–09 financial crisis — after building fabs on record timelines and running operations with a "zero-defect" culture. He is a manufacturing-and-execution CEO, not a promoter.
• What he built at MediaTek: hired in 2017 explicitly to stop gross-margin bleed from Qualcomm/Spreadtrum price competition and to fix the company's failure to win at 4G. On his watch MediaTek won the 5G cycle, took #1 smartphone-SoC share, broke into the premium/flagship tier (Dimensity 9000→9500), built the Wi-Fi/connectivity franchise, and now the data-center ASIC pivot. That is a genuine multi-year turnaround-and-expand record.
• Tenure & skin in the game: ~9 years as CEO. Specific insider-ownership %: n/a — insider-transactions.csv not on disk and not separately sourced. MediaTek is a widely-held Taiwanese public company (founder-adjacent — chairman Ming-Kai Tsai is the founder; Rick Tsai is a professional-manager hire, not the founder).
• Capital-allocation history: disciplined and shareholder-friendly. Dividend payout ~81% of earnings (~64% of cash flow); dividend CAGR ~14%/yr over 10 years. MediaTek is a high-payout, net-cash, low-M&A-risk capital allocator — it returns cash rather than empire-builds. ROE has historically been strong (a high-payout fabless model), though a precise 5-yr average ROE is n/a.
• Red flags: minimal. No related-party-deal pattern, no promotional-CEO behavior — if anything the 1Q26 "urge investors to take a long view" is a counter-promotional signal (management cooling its own stock). The one governance nuance: founder-chairman + professional-CEO split, which has worked well here.
• Archetype: professional manager of the highest caliber, operating a founder-chaired company. For a company making a capital-intensive, execution-heavy leap into data-center silicon, a TSMC-manufacturing-veteran CEO is close to the ideal profile — this is a strength of the thesis, not a risk.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst on the primary 4Q25 statements + FY color. Bottom line: the accounting looks clean and conservative; the risks here are business/cyclical, not forensic.

• Revenue recognition: merchant per-unit hardware sales — point-in-time on shipment, low aggressive-recognition risk. The one area to watch going forward is ASIC NRE + milestone revenue (large, lumpy, single-customer) — how MediaTek books TPU program revenue in 2026–27 will be worth scrutinizing, but there is no evidence of a problem yet. ``
• Cash flow vs earnings: cash-positive quality — Q4 operating cash flow NT$64.5B exceeded net income NT$23.1B; FY cash generation strong [primary: 4q25-press-release.txt]. Earnings are not being flattered by accruals — the opposite.
• Receivables/inventory vs revenue: receivables 39 days, inventory 70 days — both in normal seasonal ranges, neither outrunning revenue [primary: 4q25-press-release.txt]. No channel-stuffing or inventory-build signal.
• Margins / non-GAAP: MediaTek discloses TIFRS and Non-TIFRS. The Non-TIFRS adjustments (SBC, acquisition-amortization, tax effect) are modest — e.g. Q4 TIFRS net NT$23.07B vs Non-TIFRS NT$23.6B; FY EPS NT$66.16 TIFRS vs NT$67.46 Non-TIFRS [primary: prepared-remarks]. The GAAP-to-non-GAAP gap is small (~2%) — this is NOT a company hiding dilution behind adjusted metrics (contrast with US ASIC peers where SBC is enormous). Genuinely clean.
• Balance sheet: net cash (NT$247.7B cash/financial assets), no leverage stress, no goodwill-impairment overhang flagged [primary: 4q25-press-release.txt].
• The real forensic watch-items are forward-looking, not historical: (1) gross-margin trajectory (down 2.1pp in 2025; ASIC is structurally lower-margin than flagship mobile — a >US$2B ASIC ramp could dilute blended GM even as revenue grows); (2) opex/R&D inflation funding the data-center build; (3) ASIC revenue-recognition transparency as it scales.

Regulatory findings (required sub-section).

• SEC (EDGAR LR/AAER): regulatory/regulatory-findings.md confirms MediaTek has no CIK — not an SEC filer, no EDGAR enforcement search possible. Zero SEC findings by construction.
• Non-SEC enforcement (web search): No material antitrust/consumer-protection enforcement found (no China SAMR, EU, DOJ, FTC action surfaced) as of the search date. The one material live litigation is a US patent case: ParkerVision v. MediaTek (W.D. Texas, Waco / Judge Albright). The jury trial scheduled for 2026-03-20 was postponed pending revisions to ParkerVision's damages expert reports; it is the first of a planned three trials. This is an NPE ("patent-troll") damages case — a contingent liability, not an operational threat, but worth tracking for a potential damages award. There is also the TSMC↔Longhorn IP / Mago Barca patent-transfer matter involving MediaTek patents, resolved via a 2024 collaboration agreement — not an enforcement action against MediaTek.
• 10-K Item 3 equivalent: MediaTek files under Taiwan-IFRS, not a US 10-K — no Item 3 to quote; legal-proceedings disclosure lives in its TWSE annual report, not sourced here.
• Net: No material regulatory or accounting findings. Verified via SEC EDGAR EFTS (no CIK → n/a), web search (no material enforcement), and primary 4Q25 financials as of 2026-07-06. The only open legal item is the ParkerVision NPE patent litigation (postponed, damages-stage).

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Base actuals: FY2025 EPS NT$66.16 (TIFRS) / NT$67.46 (Non-TIFRS); FY2025 revenue NT$596.0B; net margin ~17.8% [primary: 4q25-press-release.txt]. Currency NT$; ~NT$31.2 = US$1.

Driver assumptions (each labeled):

• Smartphone: low-single-digit decline-to-flat units in 2026 on memory/BOM-cost demand pressure, offset by premium mix + price pass-through; low-single-digit recovery 2027–28 [primary: prepared-remarks guidance; web: Counterpoint, 2026].
• Smart Edge (ex-ASIC): management says it grows "healthily" in 2026 even excluding ASIC — assume ~+10% (connectivity Wi-Fi 8, computing, auto) [primary: prepared-remarks; estimate].
• Data-center ASIC: ~US$2B in 2026 (mostly Q4-weighted), scaling to ~US$5–7B in 2027 (, anchored between management's "multiple-billion" and Goldman's US$12.3B — I deliberately model below the Goldman bull case). Lower gross margin than flagship mobile (~30–40% vs ~50%+), so revenue-accretive but blended-margin-dilutive.
• Blended gross margin: ~46% in 2026 (company guide), drifting toward 43–45% by 2028 as ASIC mix grows.
• Opex/R&D: grows high-single/low-double-digit as the data-center build continues [primary: opex trend].
• Share count: roughly flat (buybacks minimal, dilution ~0).

Three-year EPS paths:

FY2026 base EPS ~NT$68; the FY2027 fork (NT$68 bear vs NT$115 bull) is entirely an ASIC-ramp function.

Forecast to log (base case, for Brier tracking — not logged in --watchlist unattended mode per SKILL rules): "2454.TW FY2027 non-TIFRS EPS ≥ NT$88, p=0.45, resolves 2027-12-31." Conviction on the base is genuinely low because the FY2027 number is dominated by one exogenous variable (Google's sourcing decision) — hence p<0.5.

Lens 12 · Bull vs Bear

Bull case. MediaTek is a proven #1 Arm-SoC franchise with a net-cash balance sheet and an 81%-payout dividend, and it has just bought a credible option on the second-largest structural growth market in tech — custom AI silicon. It powers 2B+ devices/year, owns the connectivity IP stack (Wi-Fi 7→8, tripling revenue), broke into the flagship tier (US$3B), and grew computing 80% via a direct NVIDIA relationship (GB10/DGX Spark). Now it has won the Google TPU inference variant and doubled its 2026 ASIC target to US$2B, with a self-cited US$70–80B ASIC TAM by 2027 and a 10–15% share ambition (US$7–12B). If even the base ASIC path lands, EPS roughly doubles by 2028 and MediaTek stops being a phone cyclical and becomes a diversified edge+cloud-AI silicon platform — with a CEO (ex-TSMC crisis-era CEO) uniquely suited to execute a manufacturing-heavy data-center leap. The secular tailwind (inference compute at hyperscalers, edge AI on every device) is real, durable, and multi-year.

Bear case (the risks that could permanently impair the thesis or the multiple):

1. The ASIC business is a rented seat, not a moat. Google chose MediaTek as a deliberate second source to pressure Broadcom's pricing, and Google is already multi-sourcing/diversifying its ASIC partners. MediaTek does not own the customer, the volume, or the pricing power. A single Google roadmap change (in-house more, or hand the inference variant to Marvell) removes the entire reason the stock is at 63x. This is a single-customer, single-program dependency masquerading as a growth franchise.
2. The core (90%+ of revenue) is a decelerating, margin-compressing cyclical. Smartphone SoC share fell 38%→32% YoY in Q1 2026, Mobile revenue −15% YoY, entry/mid (MediaTek's stronghold) is where demand is softening most, and memory/BOM cost inflation is a 2026 headwind. Gross margin is in a multi-year downtrend (−2.1pp in 2025) and ASIC mix will dilute it further. The base business does not justify a growth multiple.
3. Expectations are extreme. At ~63x P/E (vs Qualcomm ~14x, Broadcom <30x, its own industry ~48.7x), the market has already priced the bull ASIC path. Macquarie's NT$10,000 target implies you must believe the Goldman US$12.3B-2027 ASIC line. Any miss on ASIC timing, margin, or share re-rates the stock hard toward its Qualcomm-comparable ~14–20x.

Pre-mortem (it's 18 months out, mid-2027, the thesis broke — what happened?): The 2027 ASIC ramp slipped or shrank — Google pulled more TPU volume in-house or split the inference variant to Marvell, so MediaTek's US$5B+ 2027 ASIC number came in at ~US$2–3B. Simultaneously the smartphone cycle stayed soft (memory-cost-driven), blended GM slid to ~43%, and EPS landed near the bear ~NT$68 instead of ~NT$88+. The 63x multiple collapsed to ~25x on a de-rate-plus-miss, and the stock round-tripped a large chunk of the +200% 2026 move. Nothing was fraudulent or operationally broken — the price simply capitalized a bull case that only partly materialized.

Are multiples too high? For the current earnings base, unambiguously yes — 63x for a company where 90%+ of revenue is a low-teens-margin cyclical growing mid-single-digits is only defensible on the ASIC option. The multiple is a bet, not a valuation.

Contrarian view (what the market is refusing to see): The consensus is fixated on the ASIC upside and treating MediaTek as an AVGO/MRVL-style ASIC pure-play. What it's under-weighting: MediaTek has structurally less pricing power in ASIC than Broadcom or Marvell (it won on being the cheap second source), so even if the revenue ramp is real, the margin on it will disappoint relative to the ASIC-pure-play multiple being applied. The bear surprise isn't "ASIC fails" — it's "ASIC succeeds on revenue but at 30% gross margin, diluting the blend and never earning the AVGO multiple."

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

Dismantling the bull case.

• What structurally breaks the money machine: MediaTek monetizes (units × ASP). The unit engine (entry/mid smartphones) is in secular share-decline (38%→32% YoY) and cyclical demand-decline (memory-cost squeeze); the ASP engine (flagship + ASIC) is real but flagship is a share-taker in a shrinking premium pool and ASIC is single-customer. The whole re-rating rests on ASIC — remove it and you have a 12x-15x cyclical de-rating from 63x.
• Where revenue is concentrated / what happens if it shifts: The stock's value is concentrated in one Google program. Google is explicitly a multi-sourcer and has Broadcom (training, the high-margin part), Marvell, and its own team. If Google shifts inference volume — which it has every structural incentive to keep contestable — MediaTek's US$7–12B 2027 dream becomes US$2–3B. Meanwhile the phone revenue is concentrated in cost-sensitive Chinese OEMs playing MediaTek against Qualcomm on price.
• Why the moat is weaker than bulls think: No owned CPU (Arm-dependent) or GPU (Arm Mali, vs Qualcomm's Adreno). In ASIC, no IP incumbency vs Broadcom's decade of custom-silicon relationships. The "moat" is cost + TSMC access + integration — all rentable, all contestable.
• Most dangerous competitor bulls underestimate: not Qualcomm — Broadcom. Broadcom holds the training TPU (higher margin, deeper Google relationship) and is the reference for what a real ASIC moat looks like; Google is using MediaTek to discipline Broadcom's price, which caps MediaTek's ASIC margin structurally. Second: Google's own in-house team — the ultimate multi-sourcing threat.
• Worst capital-allocation / governance items: honestly, few — this is a clean, high-payout, net-cash, low-M&A allocator with a credible CEO. The short case is valuation + customer-concentration, not management malfeasance. (An honest short admits the quality is high; the price is the trade.)
• Assumptions that must hold for today's price: (1) Google TPU inference volume ramps to multiple-US$B for MediaTek and stays with MediaTek through 2027–28; (2) ASIC gross margin doesn't crater the blend; (3) the phone business stabilizes rather than continuing to lose share; (4) the ~US$70–80B 2027 ASIC TAM is real and MediaTek captures 10–15%.
• If growth disappoints 20–30%: an ASIC ramp of ~US$3B (vs ~US$5–6B base / US$12B bull) in 2027, with soft phones, drops EPS toward the bear ~NT$68 and would justify a de-rate to ~20–25x — implying materially lower than the current ~NT$4,195 quote. The asymmetry at 63x is unfavorable: limited upside if bull plays out (largely priced), large downside if it doesn't.
• Single scenario that permanently impairs the business (not just the multiple): there isn't a clean one — MediaTek's phone franchise is durable enough that the business survives an ASIC disappointment. What gets permanently impaired is the narrative and the multiple, not the enterprise. That distinction matters: this is an over-valuation short, not a zero.

Lens 14 · Management Questions (ordered by information value)

1. On the Google TPU program: what share of that ASIC volume is contractually committed vs. re-competed each generation, and what protects your position when Google's explicit strategy is to multi-source to pressure pricing? (This single answer moves the thesis more than any other.)
2. What is the gross margin on data-center ASIC revenue vs. your corporate ~46%, and what does the blended-margin path look like as ASIC scales to US$2B→US$5B+?
3. You doubled the 2026 ASIC target to US$2B and pulled the US$70–80B TAM forward from 2028. What specifically changed between February and April 2026 to justify doubling — new programs, higher volume on the existing one, or ASP?
4. Beyond the current Google inference variant, how many distinct ASIC customers/programs are in the pipeline, and when does customer concentration in this segment fall below 50%?
5. How do you defend flagship smartphone share against Qualcomm's owned Oryon CPU + Adreno GPU without first-party CPU/GPU IP — is licensing Arm's top cores sustainable, or do you need to own more of the stack?
6. On the smartphone base: with entry/mid share falling (38%→32%) and memory/BOM costs pressuring demand, is unit share loss cyclical or structural, and where does mobile revenue stabilize?
7. What is your CoWoS/advanced-packaging capacity allocation from TSMC for 2026–27, and is packaging (not wafers) the binding constraint on the ASIC ramp?
8. How much of the US$70–80B 2027 ASIC TAM is inference (your entry point) vs. training (Broadcom's stronghold), and can you credibly move up into training-class programs?
9. Capital allocation: with the data-center build requiring heavier R&D and capacity commitments, do you sustain the ~81% dividend payout, or does the ASIC investment cycle change the return policy?
10. On the NVIDIA GB10/DGX Spark relationship — is that a one-off computing program or the start of a broader NVIDIA custom-silicon partnership, and how large can it get?
11. What is the revenue-recognition treatment for ASIC NRE and volume — how should investors model the lumpiness and avoid misreading milestone timing as demand?
12. Wi-Fi 8 and connectivity have been a quiet compounder (>US$3B, Wi-Fi 7 tripled). What is the multi-year connectivity revenue trajectory, and is it defensible against Qualcomm/Broadcom?
13. On the DENSO ADAS partnership and automotive: what is the realistic 3-year revenue contribution, and is auto a needle-mover or a strategic option?
14. What is the risk that the ParkerVision patent litigation (or the follow-on trials) results in a material damages award or an ongoing royalty, and how are you reserved?
15. If the ASIC ramp underperforms your base case by 30%, what is the downside protection in the model — how much of today's growth investment is committed/sunk vs. flexible?

---