Fabrinet (FN)
Outsourced precision manufacturing, assembly and test on a build-to-customer-design basis; revenue is per-unit assembly fees plus largely pass-through materials, making it a low-margin, high-volume price-taker that monetizes labor and process know-how rather than its own product IP.
The thesis on this name
State of AI Compute
Structural loser of the within-layer optics margin reallocation — capped at ~12% GM, price-taker, supply-gated, and the pluggable module it assembles is exactly what CPO/optical-I/O disintermediate; avoid as a Phase-C long (the rent accrues upstream to laser/engine/DSP owners), though not yet a clean short while AI volume + supply-gated demand still lift re…
State of AI Compute
Avoid Fabrinet over a 12–24mo horizon: it is the structurally capped (~12% GM, price-taker) pluggable-module assembler whose value the 2027–29 CPO/LPO transition reallocates upstream to silicon-photonics, CW-laser and DSP owners — but AI volume + supply-gated demand still lift revenue, so it is not yet a clean short;…
State of Data-Center Power
AVOID (adjacent-stack read-across, not a power name) — the structural loser of the within-layer margin reallocation in the interconnect/optics chain that feeds AI datacenters. Capped at ~12% gross margin (Q3 FY26, fact), a price-taker, supply-gated, and the pluggable module it assembles is exactly what co-packaged optics / optical-I/O disintermediate over 2028-29. Included as the avoid-archetype for any pure-assembly toll in the broader AI-infra buildout: the rent accrues upstream to the scarce input (lasers, gear), not to the assembler. Not a near-term short (a Nvidia/Cisco program can keep assembly volume growing); a long-side avoid.
Earnings, margins, COGS & capex
Fabrinet is compounding AI-driven optical volume at ~39% YoY (Q3 FY26 revenue $1,214.3M) while holding non-GAAP gross margin almost perfectly flat near 12.1% — the defining tension of the name: explosive top-line, structurally fixed margin. Telecom (incl. DCI/800ZR) is now the growth engine at a record $628.3M, datacom is supply-gated (could have shipped materially more), and the balance sheet is pristine (~$941M net cash). The model converts revenue growth to EPS via operating leverage and buybacks, not margin expansion.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~88¢ is cost of goods and ~1¢ operating expense, leaving ~11¢ of operating profit (~10¢ net).
Revenue trend
Margins
flat — 12.0% YoY; structurally pinned near 12% for a decade
up slightly — from 10.2% YoY via operating leverage
flat-to-up
down — pressured by inventory/supply build + capacity capex
COGS structure
COGS is dominated by pass-through materials (lasers, DSP/ASICs, photonic ICs, memory, optics, connectors) bought to customer spec, plus low-cost Thai direct labor and process overhead. Because Fabrinet earns a thin conversion fee on top of customer-specified bill-of-materials it cannot mark up, gross margin is mechanically capped ~12% — richer-content 800G/1.6T modules add revenue dollars but not margin points. Material availability (not labor) is the current binding constraint.
Capex
Historically light (~2-4% of revenue) for an EMS at this scale, funding cleanrooms, automated assembly/test lines and packaging tooling. Now stepping up: Building 10 (2M sq ft, adds ~$3.0B revenue capacity; first floor live ~June 2026) and the Navanakorn facility (~$11M, $250M initial / up to $500M capacity) lift total capacity toward ~$8.5B. Plus a $32M / ~14% strategic stake in Raytech Semiconductor for advanced/CPO packaging.
Latest earnings
Beat — non-GAAP EPS $3.72 vs ~$3.54 consensus (~+5%); revenue above the guided range (fact)
Q4 FY26: revenue $1.25–1.29B, non-GAAP EPS $3.72–3.87 (~+40% YoY at midpoint); management flagged persistent datacom component/material supply constraints as the gating factor, not demand
- Non-GAAP EPS
- $3.72 (record, beat $3.54)
- Optical communications revenue
- $888.7M (73% of total)
- Datacenter interconnect (DCI)
- $196.9M, +90% YoY
- ROIC
- 41.8% (up from 39.1% YoY)
Growth drivers
- AI datacenter optics — 800G transceivers ramping, two direct hyperscaler 800G programs shipping with a steady ramp through FY2027
- Datacenter interconnect (DCI) — 400ZR/800ZR coherent modules, +90% YoY to $196.9M in Q3 FY26
- Telecom super-cycle — record $628.3M, roughly doubled over two years on coherent/AI-backbone demand
- Capacity unlock — Building 10 + Navanakorn lift ceiling from ~$4.8B to ~$8.5B, converting backlog/demand into shippable revenue
- Next-gen content — 1.6T modules, silicon-photonics integration, and early CPO/LPO packaging programs (3 active CPO customers, modest revenue)
- Share gains from OEMs outsourcing more assembly as module complexity rises
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-08-18. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’26
Bull & bear
Fabrinet is the indispensable, debt-free toll-collector on the entire Western optical-networking supply chain at the exact moment AI datacenter, DCI, and coherent-telecom demand are all inflecting — growing ~39% YoY with demand far exceeding shippable supply, a ceiling lifting toward ~$8.5B in capacity, and ~42% ROIC.
- Demand vastly exceeds supply — management explicitly said they 'could have shipped a lot more,' so revenue is gated by components/capacity, not orders; Building 10 + Navanakorn unlock that backlog
- Telecom + DCI are structurally re-rating — telecom a record $628.3M, DCI +90% YoY to $196.9M; coherent/800ZR is a durable second leg beyond datacom
- Operating leverage + buybacks compound EPS even with flat ~12% gross margin — record $3.72 non-GAAP EPS, Q4 guide $3.72-3.87, ~42% ROIC, zero debt
- Direct hyperscaler relationships (two 800G programs) + early CPO packaging (Raytech, 3 programs) position FN to participate in, not just be disrupted by, the next architecture
- Fortress balance sheet (~$941M net cash) self-funds a near-doubling of capacity with no dilution and no refinancing risk
Fabrinet is a structurally capped (~12% gross margin), price-taking pluggable-module assembler whose value the 2027-29 CPO/LPO transition reallocates upstream to silicon-photonics, CW-laser and DSP owners — and it is paying ~35x forward earnings near record highs for a labor-arbitrage conversion business with ~35% single-customer (Nvidia) concentration. Avoid over a 12-24mo horizon.
- Margin is mechanically stuck near 12% — richer 800G/1.6T modules add revenue but not margin; FN cannot capture the value its customers and component suppliers do, so EPS depends entirely on volume + buybacks
- CPO/LPO disintermediation risk — co-packaged optics fold the discrete pluggable module FN assembles into the switch package; LPO+CPO share is projected to roughly double to ~60% by 2030, reallocating value to Broadcom/Nvidia/Coherent/Lumentum silicon, not the assembler
- Valuation is rich and late-cycle — ~34.7x forward P/E, ~40x EV/EBITDA, +219% in 52 weeks, on a ~12%-margin EMS whose multiple assumes the AI optics super-cycle persists uninterrupted
- Concentration + cyclicality — Nvidia ~35% of FY24 sales; an AI-capex air-pocket or a mega-customer in-sourcing/second-sourcing decision hits a high-fixed-cost, freshly-expanded footprint
- FCF is thin (~1% TTM) — supply hoarding and capacity capex consume operating profit, so the 'cash machine' narrative is weaker than headline net cash implies
What it is worth
Forward P/E and EV/EBITDA cross-check against an EMS-with-secular-tailwind frame, anchored to FY26E EPS run-rate (~$14-15 non-GAAP annualizing the $3.72-3.87 Q4 guide).
~$350-420
multiple de-rates to ~20-25x on CPO/LPO disintermediation evidence, an AI-capex air-pocket, or a Nvidia in-sourcing/second-sourcing event
~$560-600
roughly current; ~34x forward EPS with growth decelerating from the supply-gated peak and margins flat
~$700+
capacity fills toward ~$8.5B, supply constraints ease, hyperscaler/DCI programs compound, multiple holds (~35x+ on rising EPS)
Trades ~34.7x forward P/E / ~40x EV/EBITDA (Jun 2026) — a premium multiple for a structurally ~12%-margin price-taker; the multiple is the bet that the AI-optics super-cycle persists and CPO/LPO does not disintermediate FN before FY29.
SWOT
Strengths
- Scale + reputation moat in precision optical assembly/test — the default outsourcing partner for nearly every Western optical-networking OEM (Nvidia, Cisco, Coherent, Lumentum, Marvell)
- Pristine balance sheet — ~$941M net cash, essentially zero debt, ~42% ROIC, self-funds capacity without dilution
- Riding the strongest secular demand in its history — AI datacenter + DCI + coherent telecom all inflecting together; +39% YoY in Q3 FY26
- Deep, sticky qualification barriers — multi-year customer qualifications and co-located cleanroom processes make switching slow and costly
Weaknesses
- Structurally capped ~12% gross margin — a price-taker conversion-fee model with no product IP to expand margin on
- Extreme customer concentration — Nvidia ~35% of FY24 sales; loss or in-sourcing by one mega-customer is an outsized hit
- Thin FCF conversion (~1% TTM margin) — supply hoarding and capacity capex absorb most of the operating profit
- Geographic single-point-of-failure — manufacturing concentrated in Thailand (Chonburi/Pinehurst), exposed to local disruption and tariff geography
Opportunities
- 1.6T modules + silicon-photonics integration add far higher per-unit content as AI scale-up generations advance
- Capacity expansion (Building 10, Navanakorn) lifts the revenue ceiling toward ~$8.5B, converting supply-gated demand into shipments
- Direct hyperscaler programs (two 800G transceiver wins) deepen the customer base beyond merchant OEMs and ramp through FY2027
- Position itself as the assembly/packaging partner for the CPO/LPO transition (Raytech stake, 3 active CPO programs) rather than be disintermediated by it
Threats
- CPO/LPO architecture shift (2027-29) reallocates value upstream to silicon-photonics, CW-laser and DSP owners — co-packaged optics collapse the discrete pluggable module Fabrinet assembles into the switch package
- Hyperscaler/OEM vertical integration or second-sourcing to Chinese/SE-Asian assemblers to de-risk supply
- AI-capex air-pocket — a pause in datacenter buildout would expose the high fixed-cost capacity expansion
- Component supply shortages (lasers, ASICs, memory) cap shippable revenue even when demand is strong
- Tariff/geopolitical shifts to optical supply chains and any China-exposure scrutiny
Moats, dependencies & bottlenecks
Moats
multi-year customer qualifications + co-located process IP make switching slow, but ties to a product architecture (pluggables) that may shift OEMs qualify FN lines over 12-24 months; once in a program, designs are sticky through that product generation.
largest dedicated optical EMS; breadth of customers and processes is hard to replicate The default outsourcing partner for nearly the entire Western optical-networking roster; reputation for yield/quality on hard-to-build modules.
real today, replicable by other SE-Asian/China assemblers over time Low-cost geography + automation underpins the ability to assemble at a thin fee profitably; not unique long-term.
High while net-cash — ~$941M net cash self-funds capacity others must finance Lets FN add ~$3.5B of capacity without dilution, a real but not customer-facing advantage.
FN builds to customer designs and owns little of the underlying photonics IP The structural ceiling: no proprietary silicon-photonics, laser, or DSP IP to defend margin or capture architecture-shift value.
Dependencies
~35% of FY24 sales; AI-systems optics demand drives the cycle but creates outsized single-customer exposure to in-sourcing or roadmap change.
Coherent, Lumentum, Marvell, Infinera) Cisco historically >10%; FN's volume tracks these OEMs' design wins and their own outsourcing decisions.
The binding constraint today; datacom -6% QoQ purely on shortages — FN 'could have shipped a lot more.'
Core dependency the bear case targets — CPO/LPO (2027-29) could disintermediate the discrete module FN assembles.
~39% growth is cycle-fueled; a buildout pause exposes freshly-expanded fixed-cost capacity (~$8.5B ceiling).
Geographic single point of failure; optical supply-chain tariffs and China-exposure scrutiny are watch items.
Advantages
- Scale leadership — the largest, most-qualified dedicated optical-photonics contract manufacturer in the world
- Customer breadth — supplies nearly every Western optical-networking OEM plus direct hyperscaler programs, diversifying program risk within the cycle
- Fortress balance sheet — ~$941M net cash, ~zero debt, ~42% ROIC; self-funds expansion and buybacks
- Process/yield know-how on hard-to-build precision modules that OEMs choose not to manufacture in-house
- Demand-gated backlog visibility — orders exceed capacity, giving unusually high near-term revenue confidence
Weaknesses
- Capped ~12% gross margin — price-taker conversion-fee model with no product IP to expand margin
- Heavy customer concentration — Nvidia ~35% of FY24 sales
- Thin FCF conversion (~1% TTM) — working-capital build and capex absorb operating profit
- No proprietary photonics/laser/DSP IP — captures none of the value the CPO/LPO shift reallocates upstream
- Single-geography manufacturing (Thailand) — concentration and tariff/geopolitical exposure
- Cyclical, capex-heavy expansion just as it adds ~$3.5B of capacity into a demand level that depends on AI capex continuing
Bottlenecks
- Component/material supply — lasers, ASICs/DSP, memory shortages cap shippable revenue even with demand far exceeding it (the explicit Q3 FY26 datacom constraint)
- Cleanroom/line capacity — gated until Building 10 floors and Navanakorn come online through 2026
- Skilled-labor and qualification throughput — new lines must pass customer qualification before they can ship revenue
- Structural margin ceiling — no lever to expand the ~12% gross margin regardless of volume, so profit growth is bottlenecked by unit economics
- Architecture risk — pluggable-module value pool itself is the long-run bottleneck if CPO/LPO migrates assembly into the switch package
Top signals & trends
Top signals
If component supply frees up, datacom snaps back from the -6% QoQ trough to record levels — the single biggest near-term revenue swing factor.
Full run-rate guided for mid-FY27; meaningful new revenue leg and a deeper, more-direct customer relationship if it lands.
+90% YoY to $196.9M; durable second growth engine beyond datacom and beyond a single customer.
The core thesis risk — watch whether FN's Raytech stake + 3 CPO programs convert to real revenue or whether value migrates to silicon owners; LPO+CPO share projected to ~60% by 2030.
Capacity ceiling rising to ~$8.5B — bullish if demand fills it, a fixed-cost drag if AI capex stalls.
Watch the next 10-K — rising concentration amplifies single-customer risk; falling concentration signals healthier diversification.
Trends
Primary demand driver; each generation raises per-module content and volume, fueling FN's +39% growth.
Telecom + DCI re-rating into a second secular leg; DCI +90% YoY.
Co-packaged + linear-drive optics fold the discrete pluggable module into the switch package, reallocating value to silicon-photonics/laser/DSP owners — the structural threat to the assembler.
Raises content per device (good for revenue) but concentrates value in PIC/laser owners, not the assembler (bad for FN's share of the pie).
Caps near-term shippable revenue but also reflects how strong underlying demand is; resolution is a revenue catalyst.
Mega-customers may in-source assembly or qualify Chinese/SE-Asian alternatives to de-risk supply.
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.
Lasers, InP chips, transceiver components used in modules FN assembles.
Lasers and photonic components; both supplier-adjacent and customer.
DSP/PAM4 and coherent DSP silicon inside the transceivers FN builds.
Switch ASICs, SerDes and CPO/silicon-photonics engines that define the architecture FN packages around.
Advanced/CPO packaging partner — FN holds a ~14% / $32M strategic stake (private).
Upstream PIC and CW-laser supply (incl. GlobalFoundries-type SiPh foundries) gating module content.
Largest customer, ~35% of FY24 sales; AI-systems optics and networking demand.
Long-standing >10% customer; routing/switching optical modules.
Outsources assembly/test of some transceiver lines to FN.
Outsources portions of module assembly to FN.
Customer for assembled optical modules incorporating its DSPs.
Two direct datacom transceiver programs shipping, ramping through FY2027 (customer unnamed).
EMS scaling fast into 800G/1.6T optical and hyperscaler networking (HPS segment); the most direct listed peer capturing AI-networking assembly share.
Large diversified EMS with photonics/optical capabilities; can bid for outsourced optical assembly programs.
Vertically integrated optical OEM (lasers, transceivers, InP/silicon photonics) — both a customer-adjacent partner and a vertical-integration threat that captures the upstream value FN cannot.
Laser/transceiver OEM; insources some assembly and owns the photonic IP/lasers that the CPO/LPO shift favors over the assembler.
Chinese optical-module makers (named for analysis only, not a recommendation) compete on transceiver assembly/cost and could be second-sourced by hyperscalers.
Not a company but the structural competitor — Nvidia/Broadcom CPO designs and OEM in-house assembly disintermediate the discrete-module step.