Foxconn / Hon Hai Precision
High-volume, thin-margin contract manufacturing and systems assembly at massive scale: builds hardware (smartphones, servers, networking, components) to OEM/hyperscaler spec. Revenue is enormous but net margin is ~2.3%; economics run on turns, scale, vertical integration and capacity utilization rather than pricing power. Increasingly moving up-stack into AI-server rack integration, cooling/power, and its own data-center/'AI factory' offerings, plus a nascent EV/MIH platform play.
Earnings, margins, COGS & capex
FY2025 set records on every top-line measure: revenue NT$8.1T (+18%), net profit attributable to parent NT$189.3B (+24%), EPS NT$13.61 (highest since the 1991 listing), ROE 11.25%, dividend NT$7.2/share. The structural story is mix: the Cloud & Networking division reached ~41% of revenue and, for the first time in company history, overtook Smart Consumer Electronics — driven by Nvidia GB200/GB300 AI-server racks. Margins remain classically thin (gross 6.15%, net 2.34%) but operating margin improved YoY (2.92%->3.20%) as higher-value systems integration grew. Q1 2026 continued the surge (revenue +29%, net profit +19%). The caution: Q4 2025 net profit fell ~2% YoY to NT$45.2B on record revenue and missed consensus (Bloomberg ~NT$59.9B), as high-priced AI product diluted gross margin and consumer FX weighed.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~94¢ is cost of goods and ~3¢ operating expense, leaving ~3¢ of operating profit (~2¢ net).
Revenue trend
Margins
down slightly from 6.25% (mix toward lower-GM server hardware)
up from 2.92% — improving on scale/opex leverage
up from 2.23%
healthy for an EMS at this scale
down from 2.17% — Q4 net profit fell ~2% YoY and missed consensus despite record revenue (server-mix GM dilution + FX)
COGS structure
COGS is ~94% of revenue (gross margin ~6%), overwhelmingly components (Nvidia GPUs, memory, TSMC-fabbed silicon, connectors, PCBs, mechanicals, batteries) plus direct labor and manufacturing overhead. AI-server bills-of-material are GPU-dominated, so a single large program moves both revenue and absolute COGS enormously while adding only thin incremental margin. Component cost/availability (HBM, GPUs, substrates) and FX (NT$/USD) are the main COGS swing factors.
Capex
Rising materially to add AI-server rack, liquid-cooling and power capacity and to expand US, Mexico, India and Taiwan footprint, plus data-center/'AI factory' buildouts (the US AI-server hub targets ~2,000 racks/week by end-2026, up from ~1,000/week). Company does not headline a clean capex-intensity ratio; direction is up and is a key watch item against thin margins and FCF.
Latest earnings
Mixed. Q4 2025 net profit NT$45.2B fell ~2% YoY and MISSED consensus (Bloomberg ~NT$59.9B) as server mix diluted gross margin and consumer FX weighed — a clear caution flag on bottom-line conversion. Q1 2026 rebounded to a record-high first-quarter net profit (+19% YoY) on operating-margin recovery (op profit +63%). Full-year FY2025 grew profit faster than revenue (+24% vs +18%).
Management guides strong 2026 growth led by AI servers (rack shipments ~doubling YoY), maintaining ~40% AI-server rack share; launched the next phase of a five-year AI transformation plan. No precise full-year revenue guidance number headlined.
- FY2025 revenue
- NT$8.1T (~$255B), +18%
- FY2025 net profit (parent)
- NT$189.3B, +24%
- FY2025 EPS
- NT$13.61 (record)
- FY2025 dividend
- NT$7.2/share (record; 52.9% payout)
- Cloud & networking mix
- ~41% of FY2025 revenue (48% in Q1 2026; now > consumer electronics)
- AI-server rack share
- ~40% global
Growth drivers
- Nvidia GB200/GB300 AI-server rack assembly — the dominant near-term driver; chairman Young Liu guided AI-server rack shipments to roughly double in 2026 vs 2025
- Cloud & networking now ~41% of FY2025 revenue and the #1 segment, overtaking consumer electronics for the first time (48% of Q1 2026 revenue)
- ~40% global share in AI-server rack assembly (corroborated by Digitimes, not company-only)
- Move up-stack into liquid cooling, power, and full data-center / 'AI factory' turnkey systems for hyperscalers
- Five-year AI-centered transformation plan; own-brand data-center and services ambitions
- EV / MIH open platform (contract-manufacturing model applied to cars) — long-dated optionality
- Geographic diversification (US/Mexico/India/Vietnam) reducing single-country manufacturing risk
Bull & bear
Foxconn is the indispensable, highest-share builder of the physical AI buildout — it turns Nvidia silicon into deployed racks at a scale nobody can match — and the market still prices it well below AI-infrastructure peers even as cloud/networking becomes its largest, fastest-growing segment.
- AI-server rack assembly (~40% share) is a picks-and-shovels position on the entire AI-capex supercycle, with rack shipments guided to ~double in 2026
- Segment mix inflection is real and durable: cloud & networking (~41% FY2025, 48% in Q1 2026) has passed consumer electronics — the story is no longer 'iPhone assembler'
- Record FY2025 (EPS NT$13.61, ROE 11.25%) with operating margin actually improving, plus a record dividend — growth and cash return together; Q1 2026 op profit +63% shows margin can recover
- Move up-stack into cooling/power/full data-center systems raises content and margin per rack over time (air-to-liquid + GB300)
- Trades below AI-infra multiples for a company compounding double-digit revenue with a genuine AI re-rating catalyst; optional EV/MIH upside is largely free
It is still a ~2.3%-net-margin contract manufacturer whose two mega-customers (Apple, Nvidia) hold the pricing whip, whose fastest-growing revenue can dilute the bottom line, and whose earnings ride a single AI-capex cycle with heavy China exposure.
- Thin, fragile margins: Q4 2025 net profit fell ~2% YoY on record revenue and missed consensus — proof that server mix can grow the top line without the bottom line
- Extreme customer concentration — Apple and Nvidia effectively set terms; either dual-sourcing or in-sourcing racks would gut share
- AI-server assembly is contestable: Quanta, Wiwynn, Wistron, Jabil, Celestica, Super Micro all chase the same Nvidia programs, and Nvidia can spread the work
- Rising capex into a possibly cyclical AI buildout pressures FCF; a hyperscaler digestion phase hits the growth engine directly
- Geopolitical/tariff and Taiwan-Strait risk over a still-large China manufacturing footprint; FX (NT$/USD) swings the reported result
What it is worth
Earnings-multiple / mix re-rating. FY2025 EPS NT$13.61 at ~NT$240 implies a ~17-18x trailing P/E — mid-to-high teens, already above Foxconn's historical ~10-12x, so a partial AI re-rating is embedded but it still trades below pure AI-infrastructure names. The debate is whether the AI-server mix shift earns a further re-rating.
AI-capex digestion and/or customer dual-sourcing stall server growth while thin margins and rising capex compress FCF; the multiple de-rates back toward low-teens and earnings flatten — limited or negative return.
Mid-to-high-teens P/E holds; EPS compounds with AI-server growth partly offset by mix dilution and capex — total return roughly tracks double-digit earnings growth plus the ~3% dividend yield.
Sustained AI-server rack doubling + margin lift from systems/cooling/power drives EPS growth and a re-rating toward low-to-mid-20s P/E — meaningful upside on both earnings and multiple.
At ~$106B market cap the stock is priced closer to a scaled EMS than to an AI-infrastructure enabler, but no longer at trough-cycle assembler multiples. The bull case is further re-rating as cloud/networking (~41% and rising) and up-stack systems lift blended margins; the bear case caps the multiple because net margin stays ~2.3% and two customers hold pricing power. Dividend yield ~3% (NT$7.2 on ~NT$240). Watch bottom-line conversion (Q4 2025 profit fell YoY and missed consensus on record revenue) as the swing factor. Not financial advice.
SWOT
Strengths
- Unmatched manufacturing scale and speed — can stand up gigawatt-class AI-server capacity faster than any rival
- Deep vertical integration (components, connectors via FIT, mechanicals, cooling, assembly) captures more of the BOM
- First-mover, ~40% share position in Nvidia GB200/GB300 rack integration — the fastest-growing hardware category
- Anchor relationships with both Apple and Nvidia plus every major hyperscaler
- Record FY2025 profitability and improving operating margin despite thin gross margin
Weaknesses
- Structurally thin margins (gross ~6%, net ~2.3%) — little pricing power; a customer/component squeeze hits hard
- Heavy customer concentration (Apple historically ~40-50%+ of revenue; Nvidia/AI-server now similarly critical) — pricing dictated by a few giants
- Capital-intensive, low-return-on-sales model; rising capex pressures free cash flow
- Q4 2025 net profit fell YoY and missed consensus despite record revenue — evidence server mix can dilute, not just grow, the bottom line
- Large China manufacturing base exposes it to US-China trade/geopolitical friction and tariff shifts
Opportunities
- AI-server rack shipments guided to roughly double in 2026; multi-year data-center buildout runway
- Move up-stack into full 'AI factory' / data-center systems, liquid cooling, and power — higher-value content per rack (esp. air-to-liquid transition and GB300 densities)
- US onshoring wave (customer + policy demand for domestic AI-hardware assembly)
- EV / MIH contract-manufacturing platform as a second S-curve
- Own-brand cloud/data-center services could lift blended margins over time
Threats
- Customer in-sourcing or dual-sourcing — hyperscalers and Nvidia can shift rack assembly to Quanta, Wiwynn, Wistron, Jabil, etc.
- AI-capex cyclicality: a hyperscaler digestion/pause would hit the fastest-growing segment first
- Apple unit softness / supply-chain diversification away from Foxconn (India competitors, Luxshare)
- Tariffs, export controls, and Taiwan-Strait geopolitical risk
- GPU/HBM component shortages or allocation shifts that throttle server throughput
Moats, dependencies & bottlenecks
Moats
No competitor can bring comparable AI-server capacity online as fast; scale is self-reinforcing with the largest customers.
Moderate-Strong Owns connectors, mechanicals, cooling, assembly — captures more content and de-risks supply, but each layer is individually contestable.
Deep Apple and Nvidia integration is sticky, but these customers deliberately cultivate second sources to cap Foxconn's power.
Best-in-class turns and utilization, but this is operational excellence, not structural pricing power — margins stay thin.
Dependencies
Supplier + customer GB200/GB300 GPUs are the dominant AI-server BOM and Nvidia effectively directs rack-assembly allocation; concentration cuts both ways.
Historically the single largest revenue source (~40-50%+); iPhone unit trends and Apple's supplier diversification (India, Luxshare) directly move Foxconn.
AI-server demand and capex cadence sit with a handful of cloud buyers; a digestion pause hits the growth segment first.
GPU/AI silicon and HBM/DRAM availability gate server throughput; allocation shifts throttle output.
Operational/geopolitical Tariffs, export controls and Taiwan-Strait risk over a large mainland footprint; mitigated but not eliminated by US/Mexico/India diversification.
Advantages
- Largest and fastest AI-server rack integrator globally (~40% share)
- Only player at true iPhone-scale that can also deliver hyperscale AI systems
- Vertically integrated content capture across connectors, cooling, mechanicals, assembly
- Anchor relationships spanning Apple, Nvidia and every major hyperscaler
- Record cash generation and dividend alongside double-digit growth
Weaknesses
- ~2.3% net margin with no pricing power — customers set terms
- Severe customer concentration (Apple + Nvidia)
- Server mix can grow revenue without growing profit (Q4 2025 miss the evidence)
- Rising capex intensity pressuring free cash flow
- Large China exposure and FX sensitivity
Bottlenecks
- GPU/HBM component allocation — server shipments are capped by silicon and memory supply, not by assembly capacity
- Liquid-cooling and power/data-center infrastructure capacity as racks scale to GB300 densities
- Rising capex vs thin margins — self-funding the AI capacity expansion strains FCF
- Skilled-labor and site build-out timelines for new US/Mexico capacity
Top signals & trends
Top signals
Structural, first-in-history mix shift toward the higher-growth AI segment.
Chairman Young Liu; anchors near-term growth, with US hub targeting ~2,000 racks/week by year-end.
Record Q1; margin recovery after the Q4 dip.
Margin dilution risk as server mix grows — watch bottom-line conversion (net margin 1.73% vs 2.17%).
Confidence and cash-return signal (52.9% payout, 7th straight year >50%).
Growth investment, but FCF drag against thin margins.
Trends
High positive · Foxconn is a prime physical-layer beneficiary; also its biggest single-cycle dependency.
Higher content and potentially higher margin per unit over time; air-to-liquid transition raises value per rack.
Wins onshoring demand but raises capex and execution complexity.
Negative for legacy mix · Consumer-electronics growth muted; offset by server pivot.
Applies the EMS model to autos; unproven at scale, upside optionality.
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.
GPUs are the core AI-server BOM (also the key customer/partner).
Fabs the advanced silicon inside servers and devices.
DRAM/HBM memory supplier for AI servers.
High-speed interconnects/connectors (Foxconn also builds its own via FIT).
HBM and memory; allocation gates server throughput (non-US-listed).
Largest historical customer — iPhone and device assembly.
GB200/GB300 AI-server rack integration — now a top revenue driver.
Hyperscale AI/cloud server buyer.
Azure AI infrastructure buildout.
Cloud/AI datacenter hardware customer.
AI-cluster and datacenter hardware buyer.
Server OEMs that both source from and compete with Foxconn.
Top-tier AI-server ODM and the closest rival for Nvidia rack programs.
Hyperscale AI/cloud server ODM (Wistron affiliate); direct competitor for hyperscaler racks.
Server and electronics ODM; competes in AI-server assembly.
US AI-server systems vendor; competes on liquid-cooled GPU systems.
US-listed EMS diversifying into AI/cloud infrastructure and datacenter content.
US-listed EMS/ODM with fast-growing hyperscale networking & AI hardware business.
Global EMS competitor across industrial, cloud and consumer.
Apple assembler and EMS rival competing for consumer + server work.
Mainland-China Apple-ecosystem assembler taking share of iPhone/AirPods work (context only, not a buy/own call).