Giga-Byte Technology (Giga Computing)
Vertically-integrated hardware manufacturer + ODM: designs and builds motherboards, graphics cards, laptops, and (via Giga Computing subsidiary) enterprise/AI servers and GIGAPOD rack-scale systems; sells through channel/OEM and increasingly direct to CSPs and enterprise. Low-margin, high-volume, component-cost-pass-through model.
Earnings, margins, COGS & capex
AI servers turned a mature, low-growth PC-component maker into a hypergrowth revenue story: FY2025 revenue hit a record NT$336.9B (+27%), TTM revenue +36.5%, with servers ~62% of FY2025 sales (rising to ~70% in Q1 2026) and AI servers over 80% of server revenue (~50% of total, ~56% in Q1 2026). But margins are structurally thin (gross ~10%, operating ~5%) because AI-server revenue is largely GPU/component pass-through — revenue scales far faster than profit. Net profit NT$12.19B (+25%), EPS NT$18.2 for FY2025 (up from NT$15.03 in FY2024); TTM EPS NT$20.1.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~90¢ is cost of goods and ~5¢ operating expense, leaving ~5¢ of operating profit (~4¢ net).
Revenue trend
Margins
down 0.15pp YoY; structurally pressured by AI-server pass-through mix and FX
up 0.16pp YoY; opex discipline (opex ratio 5.46%, -0.32pp)
stable; net profit +25% on +27% revenue
COGS structure
COGS ~89.6% of revenue (FY2025 gross margin 10.44%). Dominated by bought-in components — NVIDIA/AMD GPUs and CPUs pass straight through AI-server BOM, plus DRAM/HBM, PCB, power and cooling. GPU and memory pricing/allocation are the swing factors on both cost and availability.
Capex
Very low intensity: FY2025 NT$1.121B (~0.3% of revenue); 2026 plan ~NT$2.0-2.5B (roughly doubled) to expand AI-server production capacity, purchase office buildings, and develop internal AI computing infrastructure. Asset-light relative to pure-play server ODMs. (A larger NT$11.2B->NT$20-25B figure seen in some coverage is a units/currency error — the company results release states the NT$1.121B->NT$2.0-2.5B figures; verified.)
Latest earnings
Consensus beat/miss not cleanly disclosed for TWSE small/mid names; growth ran well ahead of the mature-hardware base rate
FY2026 revenue growth guided 'no less than' FY2025's +27%, despite flagged short-term impact from memory (DRAM/HBM) supply constraints
- AI servers % of total sales
- ~50% (FY2025), ~56% (Q1 FY2026)
- Server % of total revenue
- ~62% (FY2025), ~70% (Q1 FY2026)
- FY2025 EPS
- NT$18.2
- TTM EPS
- NT$20.1
- Trailing P/E
- ~16.7x; forward ~11.7x (consensus-implied)
Growth drivers
- AI server ramp via Giga Computing — NVIDIA GB200 NVL72/NVL4 and HGX B200 rack-scale (GIGAPOD) systems shipping to CSPs and HPC
- Server segment ~62% of FY2025 revenue (~70% in Q1 2026) — AI servers >80% of server revenue (~50% of total FY2025, ~56% Q1 2026)
- US AI-server production line (announced 2025) to serve North American hyperscaler demand and de-risk geography
- Legacy motherboard + graphics-card share gains (topped ASUS/MSI/ASRock in mid-2025 board revenue)
- 2026 capex doubling toward AI-server capacity and internal AI compute
Bull & bear
A cheap (forward ~11.7x P/E), net-cash way to play the AI-server buildout: Giga Computing is a credentialed NVIDIA rack-scale partner shipping GB200/HGX systems, revenue is compounding 30-60%, and the legacy board/GPU business is gaining share — with 2026 growth guided no-less-than 2025's +27%.
- AI-server revenue is ~50% of sales and rising (over 80% of server revenue), still ramping on GB200 NVL72/NVL4 and HGX B200 rack-scale, with a US line opening North American CSP demand
- TTM revenue +36.5% and Q1 FY2026 +59.8% YoY (NT$105B) — genuine hypergrowth, not a mature-hardware story
- Forward P/E ~11.7x and trailing ~16.7x is undemanding for 30%+ growth if any margin lift materializes from rack-scale/cooling attach
- Net cash ~NT$21B and low capex intensity mean the ramp is self-funded with optionality
- Legacy motherboard/GPU-card share gains (led ASUS/MSI/ASRock in mid-2025) add a cyclically-recovering second engine
- Monthly TWSE revenue disclosure (record April 2026 NT$52.27B, +73.7% YoY) gives a real-time confirmation of the ramp still accelerating
Thin-margin box-mover dressed as an AI winner: gross margin ~10% and operating ~5% mean revenue can double while profit barely moves, growth is hostage to NVIDIA GPU and DRAM/HBM allocation, and it's a mid-tier share-taker against far larger ODMs in a capex cycle that will eventually digest.
- ~10% gross / ~5% operating margin: AI-server revenue is mostly GPU/component pass-through, so the top-line growth is not translating to proportional profit
- Demand and cost both gated by NVIDIA GPU allocation and by DRAM/HBM supply — management already flagged memory constraints as a 2026 headwind
- Competes for the same CSP wallet as Foxconn, Quanta, Wiwynn, Super Micro and Wistron — all larger, all pressuring server margins
- Any AI-capex digestion or hyperscaler pause hits the ~62-70% server mix directly with little cushion from low-growth legacy PC lines
- FX and component-price volatility can erase the razor-thin margin in a single quarter
- EPS grew only +21% while revenue grew +27% in FY2025 — the pass-through model structurally caps profit leverage even in a boom
What it is worth
Peer P/E + reverse read on current price. Trades ~16.7x trailing / ~11.7x forward EPS (TTM EPS NT$20.1; price NT$336.5, mktcap NT$225.4B). Peer AI-server ODMs (Quanta, Wiwynn, Foxconn) trade low-to-mid teens forward; SMCI richer but more volatile. At ~$7.0B market cap on ~$11.8B TTM revenue, the stock is priced at ~0.6x sales — a thin-margin box-maker multiple, not a premium AI multiple.
~NT$220-260
AI-capex digestion and HBM/DRAM cost squeeze cut growth and margin; multiple compresses to high-single/low-double-digit forward P/E on flat EPS.
~NT$340-380
~25-30% growth per guidance with flat ~10% gross margin; ~12-14x forward P/E holds as EPS grinds higher.
~NT$450-500
growth stays 30%+ and rack-scale/cooling attach lifts blended gross margin toward 12%, earning a mid-teens forward multiple on rising EPS.
The forward ~11.7x P/E embeds continued 25-30% revenue growth but essentially no margin expansion — the market is treating Gigabyte as a low-margin AI-server pass-through, not a margin-compounder. Re-rating requires blended gross margin to lift above ~10-11% (via rack-scale/cooling/attach) or growth to durably beat guidance; de-rating comes from AI-capex digestion or memory-driven margin compression. Analyst 12-month targets cluster ~NT$416-422. Not financial advice.
SWOT
Strengths
- End-to-end AI-server capability under Giga Computing — GB200 NVL72/NVL4, HGX B200, GIGAPOD rack-scale with direct liquid cooling
- Trusted brand across the full stack — motherboards, GPU cards, laptops, servers — with decades of manufacturing and channel depth
- Net-cash balance sheet (~NT$21B) funds the AI ramp without leverage stress
- Asset-light capex model converts AI demand to revenue quickly without heavy fixed-cost drag
- Explosive top-line growth (+36.5% TTM) with market-share gains in legacy board/GPU lines
Weaknesses
- Structurally thin margins (gross ~10%, operating ~5%) — AI-server revenue is largely GPU/component pass-through, so profit lags revenue badly
- Deep dependence on NVIDIA GPU allocation — availability and pricing dictate both revenue and cost
- Exposed to DRAM/HBM supply constraints and pricing (management-flagged near-term headwind)
- Smaller AI-server scale than Foxconn/Quanta/Wiwynn/Super Micro — a share-taker, not the volume leader to top hyperscalers
- Legacy PC-component demand (motherboards, gaming GPUs) is cyclical and structurally low-growth
Opportunities
- Rack-scale GB200/GB300-class systems and liquid cooling command higher value per unit than bare servers
- US/North America AI-server line taps hyperscaler capex and reduces China-geography risk
- Enterprise/sovereign AI and HPC buildouts beyond the top-4 CSPs, where a mid-tier ODM can win
- Attach of own GPU cards, cooling and power into integrated racks to lift blended margin
- Next NVIDIA platform cycles (Blackwell Ultra / Rubin) refresh the entire installed base
Threats
- AI-server capex digestion / any hyperscaler spending pause hits the growth engine directly
- Intense ODM competition (Foxconn, Quanta, Wiwynn, Super Micro, Wistron, Inventec) compresses already-thin server margins
- GPU/memory shortages or allocation shifts throttle shipments regardless of demand
- NT$/USD FX swings move gross margin (management cited FX drag on 2025 margin)
- US-China tech/tariff friction and export controls on advanced-compute supply chains
Moats, dependencies & bottlenecks
Moats
HGX, MGX, GIGAPOD) must be re-won every GPU platform cycle Real engineering credential (liquid cooling, rack integration) but shared with several ODM peers; not exclusive.
Decades of manufacturing and distribution; hard to replicate but doesn't confer pricing power in commodity hardware.
Top-tier board share, but a mature, cyclical, low-margin category.
Enables self-funded growth but is a financial buffer, not a competitive moat.
Dependencies
Supplier / design partner GPU allocation drives both AI-server revenue and BOM cost; the platform cadence (Blackwell/Rubin) sets the refresh cycle.
Company-flagged 2026 headwind; HBM allocation and DRAM pricing directly gate AI-server output and margin.
Grace/EPYC/Xeon and Radeon/GeForce lines feed servers and consumer boards.
Upstream supplier Indirect — gates GPU/CPU availability across the whole BOM.
~62-70% server mix means a capex pause or digestion hits revenue directly.
Advantages
- Credentialed NVIDIA rack-scale (GB200/HGX/GIGAPOD) design wins with liquid cooling
- Full-stack hardware brand — can integrate own GPU cards, cooling, power into racks
- Net-cash, low-capex model funds AI ramp without leverage
- Fast revenue conversion of AI demand (+36.5% TTM) and share gains in legacy boards
- US AI-server line diversifies geography toward North American demand
Weaknesses
- Razor-thin gross (~10%) and operating (~5%) margins
- Profit growth structurally lags revenue growth (pass-through economics)
- Sub-scale in servers vs Foxconn/Quanta/Wiwynn/Super Micro
- Heavy single-supplier (NVIDIA) and memory dependence
- Legacy PC-component exposure is cyclical and low-growth
Bottlenecks
- NVIDIA GPU allocation — cannot ship AI servers faster than GPUs are supplied
- HBM/DRAM availability and pricing (management-flagged near-term constraint)
- Thin margin structure caps profit conversion from revenue growth
- Advanced-packaging/HBM upstream capacity at TSMC/memory makers
- Skilled integration + liquid-cooling deployment capacity for rack-scale delivery
Top signals & trends
Top signals
Each new platform refreshes the rack-scale order base; being an early partner protects share.
Bearish near-term · Company-flagged headwind; tight memory throttles AI-server shipments and margin.
Sustained CSP capex underwrites the ~62-70% server mix; a pause is the key downside trigger.
Taiwan monthly revenue disclosure gives a real-time read on the AI-server ramp.
Any lift above ~10-11% from rack-scale/cooling attach would re-rate the thin-margin bear case.
Trends
High positive · The core growth driver — rack-scale GB200/HGX demand from CSPs and enterprise.
Higher value-add per rack; plays to Gigabyte's GIGAPOD DLC capability.
Negative near-term · Raises cost and gates output through 2026.
US line aligns with hyperscaler onshoring and export-control geography.
Crowded field of larger ODMs keeps AI-server margins thin.
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.
Primary GPU + platform (GB200/GB300, HGX, Grace, MGX) supplier and rack-scale design partner — the single most critical dependency.
EPYC CPUs, Instinct accelerators, and Radeon GPUs for servers and consumer cards.
Xeon CPUs and platform silicon for servers, boards, and laptops.
Upstream foundry for the GPUs/CPUs Gigabyte integrates — indirect but foundational.
DRAM/HBM supplier; memory allocation is a flagged 2026 constraint (also Samsung, SK Hynix).
Amazon AWS, Google, Meta, Oracle) AI-server and rack-scale buyers driving the ~62-70% server mix; buy directly and via ODM channels.
GIGAPOD and GB200/HGX systems for on-prem and sovereign AI / supercomputing.
Motherboards, graphics cards, and laptops sold through distribution and gaming retail.
Closest listed comparable — AI-server/rack-scale specialist with liquid cooling; larger AI-server scale and US-centric.
Largest AI-server ODM; dominant GB200 rack integrator for top hyperscalers.
Top-tier AI-server ODM to hyperscalers; far larger server scale.
Hyperscale/cloud server ODM (Wistron affiliate); direct CSP AI-server competitor.
Server ODM and GPU-module maker; competes across AI-server and components.
Direct rival in motherboards, graphics cards, laptops, and increasingly AI servers.
Branded enterprise AI-server leader competing for the same CSP/enterprise wallet.
Enterprise/HPC server incumbent (Cray) in AI and supercomputing.
Head-to-head in motherboards and gaming GPU cards; expanding into servers.
AI-server ODM to hyperscalers; competes on GPU baseboards/systems.