Inventec
High-volume, low-margin contract design + manufacturing. Builds servers, notebooks, and smart/automotive devices to spec for OEM/hyperscaler customers; earns thin per-unit margins on massive volume, with value-add concentrated in engineering, integration, and supply-chain scale rather than IP or brand.
Earnings, margins, COGS & capex
A classic thin-margin ODM operating at ~US$21-23B revenue scale where a ~5% gross margin and ~1.8% operating margin are the norm. FY2025 revenue grew ~6.9% to a record NT$691.1B with net income up 19.65% to over NT$8.6B (EPS NT$2.42); FY2026 has accelerated sharply, with Q1 revenue up 27.6% YoY (+17% QoQ) on AI-server strength and servers crossing 50% of total revenue for the first time. Management explicitly steers investors to absolute profit (bottom line) over gross-margin percentage, since server-system shipments dilute the margin ratio even as they add dollars. Net income and EPS are growing faster than revenue (Q1 EPS +44.7% YoY), reflecting operating leverage as volume scales.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~95¢ is cost of goods and ~3¢ operating expense, leaving ~2¢ of operating profit (~1¢ net).
Revenue trend
Margins
declined QoQ in Q1 on product-mix shift as low-value-add server systems grow as a share of mix
roughly flat to slightly up (+0.1pp YoY in Q1); scale expected to lift it modestly
improving on operating leverage
up
COGS structure
COGS runs ~95% of revenue — dominated by bought-in components (GPUs, CPUs, memory, boards, chassis). In AI-server systems the customer often consigns or dictates the GPU, so Inventec captures value on integration/assembly, not the silicon, which is what mechanically depresses the gross-margin percentage as system revenue scales.
Capex
Asset-light by ODM standards (~1-2% of revenue). Investment is rising for AI-server capacity, geographic diversification (US/Mexico footprint to serve North American AI demand and hedge tariff/geopolitical risk), and the newer smart-device/automotive lines still in an investment (loss-absorbing) phase.
Latest earnings
Record quarterly revenue NT$200.3B (+27.6% YoY, +17% QoQ); EPS NT$0.68 (+44.7% YoY) — strong absolute growth, margin ratio roughly flat as expected; beat consensus on both revenue and EPS
Q2 revenue expected above Q1 with servers sustaining momentum; operating margin expected to improve modestly on scale; gross margin similar to Q1; PC/notebook volumes flat to low-single-digit decline for the year (revenue may still rise on component price increases); next-gen GPU platforms (Rubin, AMD MI) in EVT with mass production targeted H2 2026
- Q1 revenue
- NT$200.3B (+27.6% YoY, +17% QoQ)
- Q1 operating margin
- 1.8% (+0.1pp YoY)
- Q1 EPS
- NT$0.68 (+44.7% YoY)
- Server share of revenue
- >50% (first time)
- China clients' share of AI-server revenue
- >50%
Growth drivers
- AI-server rack/board assembly (Nvidia GB200/GB300 generation — next-gen Nvidia Rubin and AMD MI in EVT, mass production expected H2 2026)
- General-purpose server refresh cycle (shipments +20-25%, revenue +30%+ in Q1)
- Servers crossing 50% of total revenue for the first time — durable mix shift toward datacenter
- Smart devices / automotive electronics expected to at least double YoY (small base, still investment phase)
- Geographic capacity expansion (North America) to win US-based AI-server and enterprise demand
Bull & bear
Inventec is a real, profitable, cheaply-valued way to play the AI-server buildout, with revenue re-accelerating to +27% and the bottom line compounding faster than the top line as datacenter mix takes over the business.
- Servers now >50% of revenue and growing 30%+ — a structural shift from commodity notebooks to datacenter systems
- EPS +44.7% YoY in Q1 shows the model converts volume into disproportionate profit growth despite flat margin ratios
- Qualified across Nvidia's current and next-gen (Rubin) and AMD MI server platforms — durable design-win pipeline into H2 2026
- Forward P/E ~22 with ~3% dividend yield is undemanding for a company growing Q1 revenue ~27% and EPS ~45%
- North American capacity expansion positions it to win US AI-server share and diversify away from China concentration
A ~1.8% operating-margin commodity assembler with a small AI-server share, dangerously concentrated in Chinese clients, whose growth is entirely levered to a capex cycle it cannot control and whose margins get worse, not better, as it grows.
- ~5% gross / ~1.8% operating margin means almost no error budget — a small cost, FX, or pricing miss wipes out profit
- >50% of AI-server revenue comes from Chinese clients, squarely in the path of US-China export controls and tariff escalation
- Only ~5-7% of AI-server rack assembly vs Foxconn (~40%) and Quanta (~25-30%) — persistent scale and negotiating disadvantage
- Management's own framing — rising system revenue mechanically pushes the gross-margin ratio down — concedes the mix shift is margin-dilutive
- Earnings are cyclical and hyperscaler-capex-dependent; a post-build-out digestion phase would hit the stock hard after its run (52-wk high NT$88.60 vs NT$67.60 now)
What it is worth
Peer-relative (P/E and forward P/E vs Taiwanese server-ODM comps like Quanta, Wistron/Wiwynn, Compal) cross-checked against earnings growth; DCF is unreliable given thin, cyclical margins and capex-cycle dependence.
AI-server capex digestion and/or export-control disruption to China clients stalls growth; a ~1.8%-margin assembler de-rates toward a mid-teens multiple, giving back much of the 2025-26 re-rating.
Revenue grows high-teens to ~20%+ into 2026 on AI servers, operating margin stays ~1.8%, EPS grows ~15-20%; stock roughly tracks earnings at a ~20-22x forward multiple.
Sustained AI-server ramp (Rubin/MI in H2 2026) + China-diversifying US capacity keeps EPS compounding 20-40%; multiple holds and earnings carry the stock back toward/through the NT$88 high.
Trades at ~25.8x trailing / ~22x forward earnings with ~3% dividend yield at NT$67.60 (~US$7.6B market cap, last close Jul 6 2026). For a ~1.8%-operating-margin ODM this is a growth multiple, justified only if AI-server volume and EPS growth (Q1 EPS +44.7%) persist. The stock has already re-rated hard off its lows (52-wk range NT$38.55-88.60), so the multiple embeds continued AI-server momentum.
SWOT
Strengths
- Established, at-scale server ODM with decades of relationships across major OEMs (HPE, Dell) and hyperscalers
- Qualified participant in Nvidia's Blackwell/GB200 and RTX PRO server ecosystems and next-gen Rubin/AMD MI development
- Strong revenue acceleration (+27.6% YoY Q1) with EPS growing faster than revenue on operating leverage
- Diversified across servers, notebooks, and emerging smart-device/automotive lines
Weaknesses
- Razor-thin margins (~5% gross, ~1.8% operating) leave little cushion for cost, FX, or pricing shocks
- Limited pricing power — customers dictate specs and often the highest-value component (GPU)
- Smaller AI-server share (~5-7%) than Foxconn, Quanta, and Wiwynn — a scale disadvantage in a scale game
- Heavy revenue concentration in Chinese AI-server clients (>50% of AI-server revenue) creates geopolitical and demand-concentration risk
Opportunities
- Secular AI-server capex cycle (TrendForce projects ~28% AI-server shipment growth in 2026)
- Next-gen Nvidia Rubin / AMD MI ramps in H2 2026 as a fresh volume wave
- North American capacity build to capture US enterprise/hyperscaler demand and mitigate tariff exposure
- Smart-device/automotive segment scaling toward profitability (revenue expected to double)
Threats
- US-China export controls / tariffs that could disrupt the >50% China AI-server client base
- Intense ODM price competition compressing already-thin margins
- AI-server demand air-pocket or hyperscaler capex digestion after the current build-out
- TWD appreciation (a headwind for USD-denominated exporters reporting in TWD)
- Customer in-sourcing or consolidation toward the largest ODMs (Foxconn/Quanta)
Moats, dependencies & bottlenecks
Moats
Decades of high-volume server/notebook manufacturing and component-sourcing scale — real but not unique; shared with Quanta, Wistron, Foxconn.
Being a qualified ODM for Nvidia platforms and OEMs like HPE/Dell requires multi-year engineering integration; sticky but customers multi-source.
Design ownership (L6 board and system design for next-gen GPUs) adds value above pure contract assembly, but is table stakes among top Taiwanese ODMs.
Essentially none — customers own the brand and dictate the highest-value components; margins reflect this.
Dependencies
Component supplier + platform gatekeeper GPU allocation, platform qualification, and design cadence (Blackwell/Rubin) drive Inventec's AI-server revenue and timing.
Customer concentration >50% of AI-server revenue; exposed to China capex swings and US export-control policy.
Long-standing server/notebook ODM relationships; multi-sourced, so share is contestable.
Component supplier CPU/GPU (MI accelerators) roadmaps for general-purpose and alternative-accelerator servers.
Upstream silicon/memory Component availability and pricing (esp. HBM/DRAM) gate AI-server build volumes and costs.
Revenue growth is levered to datacenter capex it does not control; a digestion phase hits hard.
Advantages
- Qualified, at-scale ODM already inside Nvidia's current and next-gen server ecosystems
- Revenue re-accelerating (+27.6% YoY) with earnings growing faster than sales
- Diversified base (servers + notebooks + emerging smart/auto) smooths single-segment shocks
- North American capacity build to capture US demand and hedge geopolitics
- Undemanding valuation (fwd P/E ~22, ~3% yield) relative to growth
Weaknesses
- ~1.8% operating margin — almost no error budget
- No pricing power or brand; customers own the value-add layer
- Sub-scale in AI servers vs the top two ODMs
- Heavy China AI-server client concentration exposed to export controls
- Cyclical, capex-dependent earnings with limited visibility beyond a few quarters
Bottlenecks
- Structurally thin margins cap reinvestment and cushion against shocks
- GPU/HBM component allocation gates how many AI-server systems it can actually ship
- Scale gap versus Foxconn/Quanta limits negotiating leverage and top-tier rack allocation
- China client concentration is a policy-driven ceiling on part of the AI-server book
- Smart-device/automotive segment still loss-absorbing (investment phase), not yet a profit contributor
Top signals & trends
Top signals
Confirms durable mix shift from commodity PCs to datacenter.
Operating leverage is real despite flat operating-margin ratio.
Concentration + policy risk on the fastest-growing revenue line.
Dollars up, ratio down — the ODM growth paradox; watch absolute profit.
A fresh volume wave already qualified into the pipeline.
Already had a large run; leaves it exposed to any capex-digestion scare.
Trends
High positive · TrendForce projects ~28% AI-server shipment growth in 2026; primary revenue driver.
Higher system content per unit lifts revenue dollars even as margin ratio dilutes.
Tariff/geopolitics push AI-server assembly toward North America — an opportunity to diversify off China.
Directly threatens the >50% China AI-server client base.
Legacy segment flat to slightly declining; no longer the growth engine.
FX headwind for a USD-billing exporter reporting in TWD.
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 + platform qualification (Blackwell/GB200, next-gen Rubin) — the key upstream gatekeeper.
EPYC CPUs and MI-series AI accelerators.
Xeon CPUs for general-purpose servers and notebooks.
Foundry for the silicon that gates AI-server build volumes.
DRAM/HBM memory — availability and pricing drive server BOM cost.
HBM/DRAM/NAND supplier for AI-server memory (alongside SK Hynix).
Long-standing server ODM OEM customer and Nvidia rack-scale partner.
Server/notebook OEM customer (also a competitor via in-house builds).
>50% of AI-server revenue; specific names not disclosed. China-market context only, not a buy/own call.
Legacy consumer-PC customers for the notebook ODM segment; specific mix not disclosed.
Dominant AI-server rack assembler (~40% share); scale leader across EMS; Q1 FY2026 revenue ~US$66.6B on ~30% AI-server surge.
One of the two largest AI-server ODMs (~25-30% share); Q1 FY2026 revenue NT$809.2B (+66.6% YoY), AI-server >75% of server revenue, gross margin ~4.78%.
Major server ODM; parent of cloud-native specialist Wiwynn (~8-10% AI-server rack share).
Hyperscaler-focused cloud server ODM (Wistron majority-owned); Q1 FY2026 revenue NT$276.5B (+62% YoY), PAT NT$14.1B.
Notebook + server ODM peer; overlapping customer base.
US-listed AI-server vendor/integrator competing for the same GPU-server designs.
OEM that both buys from ODMs and competes with in-house AI-server builds.