Quanta Computer
Contract design and manufacturing (ODM/ODM-direct): high-volume, low-margin build-to-order hardware for hyperscalers and PC brands; server business via subsidiary Quanta Cloud Technology (QCT); revenue scales with unit volume and bill-of-materials pass-through, profit with engineering mix and rack-level integration
Earnings, margins, COGS & capex
Quanta is in a violent revenue inflection driven by AI servers: FY2025 revenue rose 50.5% to NT$2.12T and Q1 2026 hit a record NT$809.2B (+66.6% YoY) with AI servers exceeding 75% of server revenue and servers over 80% of total revenue. The trade-off is classic ODM math at hyperscale: gross margin compressed to 4.78% as $3M+ GB300 racks with huge pass-through content (GPUs, memory) dilute percentage margins even as absolute profit dollars grow - FY2025 net income reached a record NT$75.0B (+25.6%, EPS NT$19.45) and Q1 2026 net profit was NT$21.19B (+8.7% YoY, second-highest quarter on record; EPS NT$5.50, a Q1 record). Order visibility reportedly extends through 2027.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~95¢ is cost of goods and ~2¢ operating expense, leaving ~3¢ of operating profit (~3¢ net).
Revenue trend
Margins
down - -154 bps QoQ, -314 bps YoY on GB300 mix + memory costs; management expects easing after Q2 2026 (FY2025 ~7.0%, FY2024 ~7.9%)
down in % terms - operating profit NT$23B, -4.4% QoQ; operating expense ratio compressed to a historically low 1.94%
stable-to-down in %; record absolute profits (FY2025 net NT$75.0B, +25.6%)
COGS structure
Dominated by pass-through components: NVIDIA GPUs/rack systems (GB300 racks >$3M unit price), server memory (DRAM/HBM price inflation was the named Q1 2026 margin drag), CPUs, networking, power and liquid-cooling components. Component cost is largely passed to customers, which mechanically dilutes gross margin percentage as AI mix rises; the consignment model (customer owns key components) being adopted on some high-end projects in H2 2026 shifts this off Quanta's P&L and balance sheet.
Capex
FY2026 target NT$30B (~$950M), with 80-85% going to US and Thailand capacity - a tariff/geopolitics hedge and proximity play for US hyperscaler racks. AI production capacity planned to roughly double vs 2025. Still under 1% of revenue: Quanta remains asset-light relative to its revenue base.
Latest earnings
Record top line; margin compression (memory costs, GB300 mix) was the debated item rather than demand
FY2026: AI servers triple-digit growth maintained; general servers raised from flat to double-digit growth; margin pressure expected to ease after Q2 as GB300 mix stabilizes and ASIC/non-NVIDIA projects scale; CFO said on the May call that management is 'now more optimistic about this market than we were 30 days ago'
- AI servers share of server revenue
- >75% (Q1 2026)
- Servers share of total revenue
- >80%
- FY2025 EPS
- NT$19.45 (vs NT$15.49 FY2024)
- Order visibility
- through 2027 (per company commentary)
- Dividend
- NT$15.60/share, ~4.2% yield (trailing)
Growth drivers
- AI server ramp — management guides triple-digit AI server revenue growth for 2026; S&P Global models AI server revenue +132% to ~NT$2.5T and total server revenue +103% to ~NT$3.1T
- NVIDIA GB300 rack-scale systems — shipments more than doubled QoQ in Q1 2026 at >$3M per rack; next-gen platform mass production begins Q3 2026
- General servers upgraded from flat to double-digit 2026 growth in both revenue and shipments
- Custom ASIC servers (multiple-fold H2 2026 increase expected, with better margins) and non-NVIDIA GPU projects broadening the customer/silicon mix
- US + Thailand capacity expansion enabling tariff-resilient supply to US hyperscalers
- Automotive electronics recovering with double-digit YoY growth (still small revenue share); notebooks (~10M units/quarter) a stable cash cow
Bull & bear
Quanta is the purest large-cap ODM play on the AI infrastructure buildout: triple-digit AI server growth guided for 2026, order visibility to 2027, revenue possibly reaching NT$3.7T (+74%), and absolute profit dollars compounding (record FY2025 EPS NT$19.45, record Q1 2026 EPS) - all at a low-teens forward P/E with a ~4% dividend yield.
- AI server revenue modeled to more than double in 2026 (+132% to ~NT$2.5T per S&P Global) with management maintaining triple-digit growth guidance
- GB300 rack shipments more than doubled QoQ at >$3M ASP; general server business also upgraded to double-digit growth - both engines firing
- Margin pressure is mix-and-memory driven, not structural: management expects easing after Q2 2026 as ASIC servers and non-NVIDIA projects scale, and consignment shifts pass-through off the P&L
- Valuation is undemanding for the growth: ~19x trailing FY2025 EPS, roughly 13-15x annualized 2026 earnings, plus ~4.2% trailing dividend yield
- US + Thailand capacity (80-85% of capex) converts tariff risk into a competitive moat versus less-diversified Asian rivals
- Custom ASIC wave (hyperscaler silicon) plays to Quanta/QCT's ODM-direct DNA and reduces NVIDIA single-vendor dependence over time
Quanta is a sub-5%-gross-margin pass-through business levered to a single spending cycle: if hyperscaler AI capex decelerates, revenue built to NT$3.7T can unwind as fast as it inflated, while memory inflation, customer concentration, and Taiwan/tariff risk squeeze a wafer-thin margin structure with no cushion.
- Gross margin fell 314 bps YoY to 4.78% and operating profit declined 4.4% QoQ in Q1 2026 despite record revenue - growth is not currently translating proportionally to profit
- Net profit grew only 8.7% YoY against 66.6% revenue growth; percentage margins may keep compressing as $3M racks dominate mix (gross margin already halved from FY2024's ~7.9%)
- Customer concentration: a handful of hyperscalers (plus NVIDIA allocation decisions) control Quanta's fate; ODM contracts reprice against Quanta at renewal
- AI capex is cyclical: 2027+ visibility claims rest on hyperscaler budgets that can be cut in one earnings season; >80% server exposure means no diversification if it turns
- Working capital: financing GPU-laden inventory strains cash flow until consignment adoption is proven; memory cost spikes hit Quanta before pass-through catches up
- Geopolitics: Taiwan-strait risk plus evolving US tariff policy on Mexico/Thailand/Taiwan-assembled hardware could impose costs faster than the US plant buildout absorbs them
What it is worth
Earnings multiple on TWSE-listed shares; trailing and forward P/E cross-checked against growth and dividend yield
Hyperscaler capex digestion in 2027 plus sustained memory inflation stalls EPS near FY2025's NT$19-20; multiple compresses toward 12x on cyclicality fears, implying NT$230-250 (~35% downside)
Revenue grows ~60-74% but margin compression persists; EPS ~NT$24-27 at 14-15x holds the stock near NT$360-400 with the ~4% dividend as the carry
Triple-digit AI server growth lands, margins recover post-Q2 as ASIC/consignment mix improves, FY2026 EPS toward NT$28-30; a market re-rate to ~18x forward implies NT$500+ (~35%+ upside)
At NT$373-377 (Jul 8 2026) and FY2025 EPS of NT$19.45, Quanta trades ~19x trailing earnings with a ~4.2% trailing dividend yield (NT$15.60/share). Q1 2026 EPS of NT$5.50 suggests FY2026 EPS in the NT$24-30 range if the ramp holds, implying roughly 13-15x forward - inexpensive versus 60-74% modeled revenue growth, but the multiple correctly discounts sub-5% gross margins, hyperscaler concentration, and AI-capex cyclicality. Vintage 2026-07; not financial advice.
SWOT
Strengths
- Top-tier position in AI server ODM with rack-scale (GB300) integration capability few rivals match at volume
- Deep, decades-long relationships with US hyperscalers (AWS, Google, Meta, Microsoft) via QCT's ODM-direct model
- Record profitability in absolute terms: FY2025 net income NT$75B, order book visibility to 2027
- Asset-light model (<1% capex intensity) with US/Thailand/Mexico/Taiwan multi-region footprint reducing tariff exposure
- Diversified legacy businesses (notebooks ~10M units/quarter, MacBook-class ODM work, automotive) cushion server cyclicality
Weaknesses
- Structurally thin margins — 4.78% gross / ~2.8% operating leaves little buffer for component cost shocks (memory inflation already bit in Q1 2026)
- Extreme customer concentration in a handful of hyperscalers with strong pricing power over ODMs
- Heavy dependence on NVIDIA's product cadence and allocation for the majority of AI server content
- Working-capital intensity of $3M+ racks strains the balance sheet until consignment models fully adopt
- Revenue quality: much of the 66% growth is pass-through component value, not value-add
Opportunities
- Custom ASIC server programs (hyperscaler in-house silicon) and non-NVIDIA GPU projects diversify silicon exposure and can carry better margins
- US onshore manufacturing buildout (California, Nashville) positions Quanta for sovereign/onshoring-driven contracts
- General server refresh cycle upgraded to double-digit growth in 2026
- Liquid cooling and rack-level power integration move Quanta up the value stack
- Automotive electronics and edge AI as longer-term legs
Threats
- AI capex digestion: any hyperscaler pause hits an >80%-server revenue base hard
- US tariff regime on Taiwan/Mexico/Thailand-built hardware could disrupt cost structure despite footprint hedges
- Memory and component price inflation compressing already-thin margins
- Competitive intensity from Foxconn/Hon Hai (largest AI server rival), Wiwynn, Super Micro, Dell driving pricing pressure
- Taiwan geopolitical risk: concentration of engineering and some production in Taiwan
- NVIDIA shifting more integration in-house or reallocating share among ODMs
Moats, dependencies & bottlenecks
Moats
Decade-plus co-engineering with AWS/Google/Meta/Microsoft; switching an AI rack supplier mid-generation is costly, but ODM contracts reprice and dual-sourcing is standard
moderate-strong GB300-class racks (>$3M, power/cooling/networking integration) have far fewer qualified builders than board-level servers; Foxconn and Wiwynn are closing
Taiwan + US (California, Nashville) + Thailand + Mexico (Monterrey) lets Quanta quote tariff-resilient supply; capex of ~$950M/yr is trivial against its revenue but hard for smaller EMS players to match in AI-qualified capacity
Being a lead ODM for each NVIDIA generation is valuable but at NVIDIA's discretion, not a durable structural asset
Dependencies
supplier/platform GPU allocation and rack-architecture roadmap (GB300, successors) drive the majority of AI server content and timing
Amazon (AMZN), Microsoft (MSFT), Alphabet (GOOGL), Meta (META) Concentrated buyer set with strong pricing power; AI capex budgets are the single biggest revenue variable
supplier/input cost Server DRAM/HBM price inflation was the named Q1 2026 gross-margin drag; pass-through lags cost spikes
indirect supplier GPU/ASIC availability upstream constrains rack shipments
Assembly spans Taiwan, Mexico, Thailand, US; tariff shifts change landed-cost math faster than plants can move
HQ, engineering core, and part of production concentrated in Taiwan
Advantages
- One of only a few ODMs shipping NVIDIA rack-scale AI systems at hyperscale volume today
- ODM-direct model (QCT) cuts out brand-vendor margin, historically making Quanta the cost leader for hyperscaler fleets
- Order visibility through 2027 per management - unusual depth for an EMS/ODM
- Balance across AI servers, general servers, notebooks, and automotive smooths (somewhat) the cycle
- Early mover on US onshore AI server assembly relative to Taiwanese peers
Weaknesses
- Sub-5% gross margin leaves near-zero shock absorption
- Percentage-margin dilution is structural to the AI rack mix (pass-through BOM dominates)
- Revenue concentration: servers >80% of revenue, AI >75% of that
- Limited brand/IP ownership - value capture depends on staying designed-in each silicon generation
- Foreign-listing friction for US investors (TWSE listing, thin OTC liquidity, TWD currency exposure)
Bottlenecks
- GPU and HBM/DRAM allocation - upstream component supply caps rack output regardless of demand
- Working capital to finance $3M+ GPU-laden racks through the build cycle (consignment shift is the mitigation, not yet fully in place)
- Data-center power and liquid-cooling component availability for rack-scale systems
- AI-qualified plant capacity: US/Thailand buildout must double AI output vs 2025 to meet guided growth
- Engineering talent for rack/power/cooling integration across three continents
Top signals & trends
Top signals
Demand signal strengthened, not softened, into mid-2026; CFO 'more optimistic than 30 days ago'
Next-gen NVIDIA transition executing on schedule; next platform mass production starts Q3 2026
Watch Q2/Q3 2026 for the promised easing; failure would signal structural margin erosion
Reduces working-capital strain and inventory risk; also slightly deflates reported revenue
Tariff hedge and hyperscaler proximity; execution risk on doubling AI capacity
Backlog depth supports the 2026 ramp
Operating leverage not yet showing; profit growth lags top line badly
Trends
The dominant driver; S&P Global models Quanta AI server revenue +132% in 2026
positive for revenue, negative for margin % · Raises server ASP ~75% YoY in 2026 (per S&P Global) while diluting gross margin percentage
ASIC server programs named by management as a H2 2026 margin and diversification lever
Costs and complexity rise, but Quanta's US buildout can win share from less-hedged rivals
Named cause of Q1 2026 margin compression; persists while AI demand outruns memory supply
Adds engineering value-add per rack for qualified integrators
~10M units/quarter, industry-average growth; now a minority of revenue (below 20% and falling)
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 and rack-scale reference architectures (GB300); the critical input
Instinct GPUs and EPYC CPUs for non-NVIDIA projects
Xeon CPUs for general servers
Networking silicon and hyperscaler custom-ASIC ecosystem
Server DRAM/HBM; memory pricing directly moves Quanta's gross margin
Upstream foundry for the GPUs/ASICs Quanta integrates
Power supplies and thermal components for racks
Hyperscaler cloud server customer
Hyperscaler AI infrastructure customer
Long-standing ODM-direct server customer
OCP-aligned data-center hardware customer
Notebook (MacBook-class) ODM relationship on the legacy PC side
Notebook ODM customer
PC ODM customer (also a server-side competitor)
Largest EMS globally and the biggest AI server rival, also building GB-series racks and US capacity
Hyperscaler-focused ODM-direct pure play, direct competitor for cloud server contracts
US-based AI server systems vendor competing for enterprise and neocloud AI deployments
Branded AI server share gains with enterprises and neoclouds (e.g. large GPU cluster deals)
Branded server/HPC rival, strong in sovereign and enterprise AI
Taiwanese ODM in servers and notebooks, smaller AI rack presence
Closest notebook-ODM peer, pushing into servers
ODM diversifying into AI servers
EMS winning hyperscaler networking/compute programs
US EMS expanding in AI infrastructure manufacturing