Wistron
Contract design + manufacturing (ODM/EMS): builds AI-server compute boards, GPU baseboards, full liquid-cooled racks and IT products for hyperscalers, NVIDIA and OEMs on thin per-unit margins at massive scale; revenue scales with AI-server unit volume and rising per-rack ASPs.
Earnings, margins, COGS & capex
FY2025 revenue doubled to a record NT$2.19T (~$69B) with net profit NT$27.4B (~$866M, +57% YoY) and EPS NT$9.04 (+48%) — both records — driven by AI products reaching ~70% of revenue. Q1 2026 accelerated: revenue NT$846.3B, net profit NT$9.63B (record quarter, +80.7% YoY), EPS NT$3.06 (vs NT$1.85 a year earlier), with servers 79% of sales. The story is scale over margin: gross margin compressed to ~6% (Q4 2025 5.62%) as complete-rack shipments (lower % margin, higher absolute $) dominate the mix.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~94¢ is cost of goods and ~2¢ operating expense, leaving ~4¢ of operating profit (~1¢ net).
Revenue trend
Margins
declining (Q4 2025 5.62%) as full-rack mix rises
thin, roughly stable
structurally thin ODM economics
sharply rising on volume
COGS structure
~94% of revenue — dominated by bought-in components (NVIDIA GPUs/HGX boards, CPUs, HBM memory, power, liquid-cooling, PCB/substrate). Value-add is design, integration, test and manufacturing yield, not component margin; ASP is largely pass-through of expensive GPU content, which mechanically depresses gross-margin % even as absolute gross profit grows.
Capex
2026 guided ~NT$60B (~$1.9B), +69% YoY from NT$35.5B, most earmarked for AI-related capacity in Taiwan and North America (compute-board lines, liquid-cooling rack integration); modest ~2% of revenue but rising fast in absolute terms.
Latest earnings
Beat on both lines: EPS of NT$3.06 versus NT$2.90 aggregated sell-side consensus (+5.5%), and consolidated revenue of NT$846.30bn versus NT$767.83bn consensus (+10.2%) — per Investing.com's consensus aggregation, a secondary source; no primary sell-side research note was obtained. Both revenue and EPS were single-quarter records. Net income attributable to parent was NT$9.631bn, +17.9% QoQ and +80.7% YoY, versus NT$1.85 EPS in Q1 2025. Against the company's own prior periods the print was a sharp acceleration driven by AI server demand.
Wistron gave qualitative segment guidance on the May 8 2026 call rather than a consolidated revenue/EPS range: AI server momentum continues to strengthen, with H2 2026 expected to outperform H1 2026; general-purpose servers expected to grow more than 20% both QoQ and YoY; networking products maintained at triple-digit percentage YoY growth; notebooks expected to decline modestly QoQ in Q2 2026 on rising DRAM prices feeding into system pricing and potential end-demand impact; desktops and displays expected roughly flat sequentially. No consolidated numeric guidance range was given.
- Consolidated revenue
- NT$846.30bn (Q1 2026), +17.4% QoQ / +144% YoY
- Gross margin
- 5.21% (Q1 2026), -0.41pp QoQ / -2.6pp YoY
- Operating income
- NT$29.14bn (Q1 2026), +14.4% QoQ / +92.3% YoY
- Net income attributable to parent
- NT$9.631bn (Q1 2026), +17.9% QoQ / +80.7% YoY — record quarter
- EPS
- NT$3.06 (Q1 2026) vs NT$1.85 in Q1 2025 — single-quarter record; beat NT$2.90 consensus by +5.5%
Growth drivers
- NVIDIA GB200/GB300 (Grace Blackwell) rack and compute-board volume ramp — Wistron is one of a handful of NVIDIA-certified full-rack ODMs
- Rising per-rack ASP as content shifts to full liquid-cooled NVL-class systems
- Hyperscaler AI capex supercycle (Microsoft, Meta, AWS, neocloud demand)
- Reported next-gen Vera Rubin / VR200 compute-tray design-win pipeline expected to carry the ramp into 2027
- Enterprise & Cloud / server segment now ~79% of sales, displacing legacy notebook/PC EMS
Bull & bear
A cheap (~10.6x forward earnings) pure-play on the NVIDIA AI-rack build-out, with scarce certification, doubling revenue, and record profits — operating leverage as absolute gross profit compounds even at thin margins.
- Forward P/E ~10.6 (Jul 2026) is undemanding for a business growing revenue triple-digits with record EPS
- Scarcity value: only a handful of ODMs are NVIDIA-certified for full GB200/GB300 racks; Wistron is reported in the next-gen VR200 pipeline
- Absolute net profit and EPS are compounding fast (Q1 net profit +80.7% YoY; EPS NT$3.06 vs NT$1.85) even as % margins compress
- AI-server ASPs keep rising with liquid-cooled full-rack content, lifting revenue per unit
- Multi-year visibility: hyperscaler AI capex guidance and NVIDIA roadmap extend the ramp into 2027+
A thin-margin, working-capital-hungry assembler whose fortunes are wholly downstream of NVIDIA allocation and hyperscaler capex — a single-digit gross-margin business priced for a cycle that will eventually digest.
- ~6% gross / ~1.3% net margin leaves no room for error; a mix or yield slip erases profit
- Cash flow is structurally pressured — huge inventory of GPUs/components financed with debt; FCF likely thin-to-negative
- No pricing power: squeezed between NVIDIA's component pricing and hyperscalers' bid pressure, with Foxconn/Quanta undercutting
- Revenue is lumpy across generational transitions (management already flagged flat sequential compute-board quarters at GB200->GB300)
- Cyclical AI-capex risk: any hyperscaler pause hits a high-fixed-cost ramp; the low forward multiple reflects real cyclicality, not a bargain (trailing P/E is ~16.7x)
- Taiwan geopolitical tail risk and TWD/FX translation on USD-denominated hardware flows
What it is worth
Forward P/E on a cyclical thin-margin ODM, cross-checked vs Taiwan AI-server peers (Foxconn, Quanta, Inventec)
AI-capex digestion or a GPU-transition air-pocket flattens volumes against a high fixed-cost, inventory-heavy base; thin margins swing profit sharply lower and the low multiple compresses further — the cheapness is a value trap in a downturn.
High-double-digit revenue growth continues into 2026-27 at ~6% gross / low-single-digit net margin; stock roughly tracks EPS growth at a stable ~10-12x forward, modest appreciation with generational lumpiness.
Sustained AI-rack demand + next-gen (Rubin/VR200) wins keep EPS compounding; multiple re-rates toward mid-teens as cash conversion proves out — meaningful upside from ~10.6x on rising earnings.
Trades ~10.6x forward earnings (~16.7x trailing) at ~NT$158 / ~$16B market cap (Jul 6 2026) — a low forward multiple reflecting single-digit gross margins, working-capital intensity, and NVIDIA/hyperscaler cyclicality despite triple-digit revenue growth. Valuation is volume-and-ASP-driven, not margin-driven; the durable question is FCF conversion and margin stability across GPU generations, not top-line growth. Not financial advice.
SWOT
Strengths
- One of a handful of NVIDIA-certified GB200/GB300 full-rack ODMs — high barrier, hard-won thermal/liquid-cooling and yield expertise
- Revenue scale doubled past NT$2T with record profits — proven ability to ramp AI-server volume fast
- Diversified beyond a single customer after exiting Apple iPhone assembly (sold India plant to Tata)
- Reported design-win pipeline extending toward next-gen NVIDIA Vera Rubin / VR200 trays
Weaknesses
- Razor-thin margins (~6% gross, ~1.3% net) — little cushion if volumes or ASPs wobble
- Heavy working-capital/inventory financing to carry expensive GPU content; cash conversion is a structural risk
- Profit % mechanically diluted as full-rack pass-through content grows
- Limited pricing power against both NVIDIA (supply) and hyperscaler customers (demand)
Opportunities
- AI-server capex supercycle still early; per-rack content and unit volumes rising
- Liquid-cooling and full-system integration raise value-add vs bare-board assembly
- US/Mexico/SE-Asia manufacturing footprint positions it for supply-chain-localization demand
- Affiliate cloud-ODM Wiwynn provides adjacent hyperscaler exposure
Threats
- Extreme customer concentration around NVIDIA's roadmap and a few hyperscalers
- Intense ODM competition (Foxconn, Quanta, Inventec) compressing margins in bids
- GPU allocation, HBM supply, and generational transition (GB200->GB300->Rubin) timing risk
- AI-capex digestion / any hyperscaler spending pause would hit a high-fixed-ramp cost base
- Taiwan geopolitical and FX (TWD) exposure
Moats, dependencies & bottlenecks
Moats
NVIDIA full-rack certification + liquid-cooling/thermal integration know-how real but shared with ~3 rivals and dependent on staying on NVIDIA's roadmap Solving GB200 thermal/liquid-cooling qualification was a genuine engineering barrier
capacity to absorb multi-billion-$ GPU throughput) Volume begets allocation and cost, but Foxconn is larger
Switching cost / design-in with hyperscalers and NVIDIA per platform generation resets each GPU generation Design wins must be re-earned at each roadmap transition
Dependencies
Supplier + demand driver GPU/HGX component supply, certification, and roadmap timing all controlled by NVIDIA; Wistron's AI revenue is downstream of NVIDIA allocation
End demand for racks; a capex-digestion pause hits volumes directly
HBM allocation and liquid-cooling components gate rack completion
Constrains total GPU availability feeding the rack build
Advantages
- Scarce NVIDIA full-rack ODM qualification
- Fast, proven volume ramp (revenue doubled in one year)
- Reported next-gen (VR200/Vera Rubin) design pipeline
- Post-Apple diversification into higher-growth AI infrastructure
Weaknesses
- Structurally thin, still-compressing margins
- Weak pricing power vs both suppliers and customers
- Cash-flow/working-capital intensity
- Concentration on NVIDIA roadmap and a few hyperscalers
Bottlenecks
- GPU/HBM component allocation (upstream CoWoS + HBM supply caps buildable rack volume)
- Liquid-cooling and thermal-integration engineering/yield at GB300 and Rubin transitions
- Working-capital/cash to finance ballooning GPU inventory
- Skilled integration/test capacity and power at manufacturing sites
Top signals & trends
Top signals
AI-server ramp translating to record absolute profit
Growth is volume/ASP-led, not margin-led
Management investing behind sustained AI-platform demand
Confirms lumpiness across generational transitions
Market pricing cyclicality/thin margins, not a growth premium
Trends
Primary tailwind; multi-year hyperscaler + neocloud rack demand
Positive on revenue, negative on margin % · Raises ASP and value-add but dilutes reported gross-margin percentage
Continuous design-win opportunity but recurring transition lumpiness and requalification cost
Aligns with hyperscaler and policy demand for non-China assembly
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 / HGX baseboards — the dominant BOM component and platform owner
Alternative GPU/CPU (Instinct, EPYC) supply
HBM memory (gating rack completion)
HBM/DRAM supply
Upstream advanced-node + CoWoS packaging that caps GPU availability
Buys/integrates Wistron compute boards into reference racks; both supplier and channel
Hyperscaler AI-rack buyer
Hyperscaler AI infrastructure buyer
Cloud AI capacity buyer
OEM channel reselling integrated racks
Largest AI-server ODM and lead NVIDIA rack partner; sets the competitive/price benchmark
Direct GB200/GB300 rack ODM rival; Q1 2026 revenue NT$809.2B, net profit NT$21.19B, close peer
NVIDIA-certified AI-server ODM competing on the same platforms
Hyperscaler cloud ODM; Wistron affiliate but competes for AI-rack/board share
US-listed AI-server maker; competes for hyperscaler/enterprise rack business
OEM shipping NVIDIA racks; both a competitor and integration partner
US/Canada-listed ODM winning hyperscaler networking/compute share
Smaller Taiwan AI-server maker