MediaTek
Fabless chip design: sells system-on-chip and connectivity silicon to handset/consumer/auto OEMs; outsources fabrication to TSMC; monetizes IP+design, increasingly via custom-silicon (ASIC) design-service engagements for hyperscalers.
Earnings, margins, COGS & capex
Two-speed business. Core mobile (~49% of Q1 FY2026 revenue) is mature and declining (-15% YoY) as low-end demand fades and DRAM cost inflation squeezes; Smart Edge Platforms (~46%, +13% YoY) -- connectivity, compute, auto and the emerging data-center ASIC line -- is the growth engine. FY2025 hit record revenue but net income slipped 1% and gross margin compressed ~2pts on mix/FX. The bull case is entirely the ASIC ramp; the base business is flattish-to-declining.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~53¢ is cost of goods and ~32¢ operating expense, leaving ~15¢ of operating profit (~16¢ net).
Revenue trend
Margins
compressing from 47.5% FY2025 / 49.6% FY2024 -- mix shift to lower-margin ASIC + FX
down from ~19% YoY on lower volume + sustained R&D
down from ~19% YoY; net income NT$24.38B, -17.4% YoY
COGS structure
Dominated by TSMC wafer/foundry costs (advanced nodes for flagship Dimensity) plus packaging/test and rising memory/DRAM component pass-through; wafer pricing and node mix are the primary gross-margin levers. Custom-ASIC revenue carries structurally lower gross margin than branded mobile SoC, diluting blended GM as it scales.
Capex
Minimal own-fab capex (fabless). Spend is R&D-weighted (advanced-node tape-outs, ASIC design teams, IP) rather than plant; opex ratio guided ~31% of revenue in Q2 FY2026.
Latest earnings
Revenue NT$149.2B at the high end of guidance; net income NT$24.38B, -17.4% YoY on margin compression -- mixed print (top-line at high end, earnings down double digits)
Q2 FY2026: revenue NT$140.2-149.2B (flat to -6% QoQ, -1% to -7% YoY); gross margin 46.0% +/-1.5pt; opex ratio 31% +/-2pt. FY2026 outlook held at mid-to-high single-digit USD revenue growth.
- Mobile % of revenue
- 49% (-15% YoY)
- Smart Edge % of revenue
- 46% (+13% YoY)
- EPS (Q1 FY2026, reported)
- NT$15.17 (vs NT$18.43 YoY)
- AI ASIC FY2026 target
- ~$2B (doubled from ~$1B; ~$2B Q4 run-rate)
Growth drivers
- Data-center AI ASIC — FY2026 revenue target doubled from ~$1B to ~$2B (management now sees a ~$2B run-rate as early as Q4 FY2026, ahead of schedule); does physical-implementation + SerDes IP for Google's TPU (v7e in production, gaining share on the v8t training chip, v9 orders reported) and co-designed Nvidia's GB10 superchip (DGX Spark)
- Flagship/premium Dimensity mobile SoC share gains at the high end
- NVLink Fusion adoption -- custom AI silicon with high-speed interconnect IP
- Automotive and edge-AI compute platforms
- Connectivity (Wi-Fi 7 and beyond) attach
Bull & bear
MediaTek is repricing from a cyclical mobile-SoC vendor into a top-3 merchant custom-AI-silicon house, with concrete hyperscaler design wins (physical-implementation/SerDes IP on Google's TPU, Nvidia GB10 co-design) that give it a credible path to a ~$2B FY2026 ASIC run-rate and 10-15% of a $70-80B TAM by 2027-28.
- Real, named ASIC wins -- not vaporware: physical-implementation + SerDes IP on Google's TPU (v7e/v8t/v9) and co-designed Nvidia's GB10 superchip, plus early NVLink Fusion adoption
- ASIC ramp visibility: FY2026 target doubled from ~$1B to ~$2B, with a ~$2B run-rate seen as early as Q4 -- ahead of schedule
- Diversification finally reduces smartphone-cycle dependence -- Smart Edge already ~46% of revenue and growing double digits
- Still #1 in smartphone SoC (~32-34% 2026 share) -- a durable cash engine funding the AI pivot
- Net-cash balance sheet and fabless model = high FCF conversion to sustain R&D and dividends through the transition
The stock has tripled off the ASIC narrative to a ~63x trailing multiple while the actual P&L is going the wrong way -- revenue and profit falling YoY, gross margin compressing toward 46%, and the ASIC business is lower-margin, customer-concentrated, and multi-sourced against Broadcom and Marvell.
- Valuation prices in near-flawless ASIC execution: ~63x trailing earnings (NT$6.73T mkt cap on FY2025 net income NT$106.1B) on a business with -2.7% YoY revenue and -17.4% YoY profit in Q1 FY2026
- Google explicitly splits TPU work across Broadcom, Marvell AND MediaTek (dual-sourcing reportedly through TPU v10) -- share and pricing power are not secured; Broadcom dominates (~70%) custom-AI ASIC
- Structural gross-margin dilution: ASIC design-service revenue is materially lower-margin than branded mobile SoC
- Core mobile is shrinking (-15% YoY) with DRAM cost inflation killing low-end unit demand
- Concentration + geopolitics: dependent on TSMC, Arm IP, and a Taiwan manufacturing base exposed to export-control and cross-strait risk
What it is worth
Peer-relative multiple + reverse read on the ASIC ramp. Trades at ~NT$6.73T (~$208B) market cap -- roughly 63x trailing earnings (FY2025 net income NT$106.1B) after a ~3.7x run off the 52-week low. Peer set: QCOM (~mid-teens P/E), AVGO (premium AI-ASIC multiple), MRVL, NVDA. The multiple sits far above the mobile-SoC peer group and only makes sense if the custom-AI-silicon ramp compounds.
ASIC share is capped by Broadcom/Marvell and slips timing; mobile keeps declining on DRAM inflation and GM erodes below 46%. A ~63x multiple compresses toward the mobile-SoC peer group (teens-to-20s P/E), implying material downside.
ASIC delivers ~$2B FY2026 with credible but multi-sourced share; mobile stays flattish and GM holds ~46%. Mid-to-high single-digit USD revenue growth roughly supports the current elevated multiple with limited further re-rating.
ASIC hits the ~$2B FY2026 target (~$2B Q4 run-rate, ~$8B annualized) and scales to 10-15% of a $70-80B (2027) TAM, i.e. $7-12B incremental higher-visibility revenue, sustaining the premium multiple and re-rating alongside AVGO/MRVL.
The current price implies the market is valuing MediaTek as an emerging top-3 merchant custom-AI-silicon franchise, not a mature mobile-SoC vendor -- despite core revenue/profit currently declining YoY. The gap between a ~63x multiple and a shrinking base P&L is the entire debate.
SWOT
Strengths
- #1 global smartphone SoC vendor by shipments (~32-34% share, 2026) with scale in mobile
- Deep advanced-node design capability and priority TSMC relationship
- Proven custom-silicon design-service credibility — Google TPU physical-implementation/SerDes-IP role and Nvidia GB10 co-design
- Broad connectivity IP (Wi-Fi, 5G modem) and diversified end markets (mobile, TV, auto, IoT)
- Net-cash balance sheet funds R&D through cycles
Weaknesses
- Core mobile revenue declining (-15% YoY Q1) and structurally mature
- Gross margin compressing (~46% vs ~50% two years ago) on mix + FX
- Weaker in premium modem/flagship-halo vs Qualcomm; historically over-indexed to a shrinking value tier
- ASIC business is lower-margin and dilutes blended profitability as it scales
- Heavy customer concentration risk in nascent ASIC line (Google, Nvidia)
Opportunities
- Capturing 10-15% of a $70-80B (2027) custom AI-silicon TAM as hyperscalers diversify away from merchant GPUs
- NVLink Fusion ecosystem -- becoming a preferred custom-ASIC partner for Nvidia-connected systems
- Automotive SoC and edge-AI inference at the device
- Premium-mobile share gains as flagship Dimensity closes the gap on Snapdragon
Threats
- DRAM/memory cost inflation squeezing handset BOMs and unit demand at the low end
- Qualcomm re-taking premium share and pushing into PCs/auto/ASIC
- Google multi-sourcing TPU across Broadcom + Marvell + MediaTek (dual-sourcing reportedly extending through TPU v10) -- share not guaranteed
- Geopolitical/Taiwan-concentration and export-control risk
- Smartphone volume stagnation; 'cheap phones dying' compressing the value-SoC pool
Moats, dependencies & bottlenecks
Moats
#1 shipments (~32-34%) give volume, IP amortization and OEM relationships, but the pool is mature and margin-eroding.
Ability to tape out leading-edge SoCs and secure TSMC capacity is a real barrier; shared with Apple/Qualcomm/Nvidia, not exclusive.
Google TPU (v7e/v8t/v9) physical-implementation/SerDes work and the GB10 co-design prove capability, but this is a contestable RFP market vs Broadcom/Marvell -- reputational, not structural, lock-in.
High-speed interconnect (SerDes) and Wi-Fi/5G IP add attach value and differentiate ASIC bids.
Dependencies
Foundry / manufacturing Fabless -- all advanced-node fabrication outsourced to TSMC; wafer pricing and capacity allocation directly set COGS and supply.
CPU/GPU IP licensing Core SoC architectures license Arm ISA and cores; licensing terms and roadmap shape competitiveness.
TPU physical-implementation/SerDes-IP win anchors the data-center ASIC story but is multi-sourced with Broadcom + Marvell.
Partner + customer GB10 co-design and NVLink Fusion partnership are a growth vector but tie a key franchise to Nvidia's roadmap.
Advanced tape-outs depend on leading EDA + IP; export-control exposure on toolchain.
DRAM cost inflation is pressuring handset BOMs and MediaTek's addressable unit demand.
Advantages
- Largest smartphone-SoC install base and OEM design-win engine
- Fabless model -> low capex, high FCF conversion, net-cash balance sheet
- Full-stack IP: SoC + connectivity + high-speed interconnect (SerDes) usable in ASIC bids
- First-mover credibility in merchant custom-AI silicon (Google TPU, GB10, NVLink Fusion)
Weaknesses
- Declining, margin-eroding core mobile business
- Blended gross-margin dilution as lower-margin ASIC scales
- Concentrated, contested ASIC customer base (Google, Nvidia)
- Geographic/geopolitical concentration in Taiwan + TSMC dependence
- Weaker premium-flagship and standalone-modem position than Qualcomm
Bottlenecks
- TSMC advanced-node wafer allocation and pricing
- Data-center ASIC design-team capacity and hyperscaler qualification cycles
- DRAM/memory cost inflation gating low-end handset volumes
- Winning/retaining share against Broadcom + Marvell in each hyperscaler ASIC socket
Top signals & trends
Top signals
Management says the ramp is ahead of schedule; the whole re-rating hinges on this landing.
Mix + FX compression; watch whether ASIC scale erodes blended GM further.
Core cash engine shrinking; low-end demand hit by DRAM cost inflation.
Expectations are elevated; execution bar is high, downside on any ASIC slip.
Validates MediaTek's capability but caps share and pricing power.
Trends
$70-80B TAM by 2027; MediaTek targeting 10-15% share -- the core bull driver.
'Cheap phones dying' compresses MediaTek's historical value-SoC pool.
Raises handset BOMs, pressures unit demand and MediaTek's mobile revenue.
Drives premium SoC content and Smart Edge platform growth.
Opens custom-ASIC sockets adjacent to Nvidia systems for interconnect-capable designers.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Sole leading-edge foundry partner -- fabricates MediaTek's advanced-node SoCs and ASICs.
Licenses CPU/GPU IP underpinning MediaTek SoC architectures.
EDA tools + IP for advanced tape-outs.
EDA / design-flow toolchain.
Memory (DRAM) -- a BOM component and a cost-inflation pressure point.
Major Chinese handset OEMs -- largest Dimensity/Helio SoC buyers (context only -- not investment calls).
Buys MediaTek SoCs for parts of its handset/TV lineup alongside captive Exynos.
Data-center ASIC customer -- MediaTek does physical-implementation + SerDes IP for Google's TPU (v7e/v8t/v9).
Customer/partner for GB10 (DGX Spark) co-design.
Primary smartphone-SoC rival (Snapdragon); stronger in premium/flagship + standalone modem, also pushing into PC/auto/ASIC.
Dominant custom-AI ASIC vendor (~70% share) and lead Google TPU partner -- the benchmark MediaTek must win share from.
Custom-AI ASIC + interconnect rival, also part of Google's multi-vendor TPU supply chain.
Merchant-GPU incumbent that custom ASICs target; simultaneously a MediaTek partner (GB10, NVLink Fusion) -- competitor and customer.
Captive + merchant mobile SoC rival; Exynos share rising (~7% Q1 2026).
Vertically integrates its own A-series/M-series silicon -- removes premium volume from the merchant SoC pool.
China-based entry/value SoC vendor gaining low-end 5G/LTE share (context only -- not a US-investable or recommended name).