ins-$1.0M · 13F 12+/12-×0.0 · short↑0.23
Current snapshotPublic companyResearch currentFresh
Fabrinet
Connectivity & Positioning · Physical AI
SignalPOOLReviewed 25 Oct 2026
Connectivity & Positioning exposure in Physical AI.
Fabrinet - optical transceiver EMS (800G/1.6T optics supply chain; pump lasers, InP substrates).
Catalysts
Next catalyst is Q1 FY2027 print (fiscal Q4 FY2026 reported 2026-08-17: $1.316B rev +45% YoY; FY2026 $4.64B +36%; DCI run-rate >$1B; capacity plan $12.5-14B). Watch 800G→1.6T merchant-transceiver second-customer production (management: Dec quarter) and NPO vs CPO mix. Parked pending panel.
Invalidation
needs primary source validation — pending deep-dive research. Placeholder: (1) thesis chain breaks on primary-source evidence, (2) named revenue/customer concentration risk, (3) margin compression or competitive displacement
Capital in context
Mapped funds and investors
Direct relationships supported by published evidence.
Reported holdings as-of
No sourced relationships mapped yet.
Disclosed investment history
No sourced relationships mapped yet.
Showing up to four relationships in each group. ETF weights describe the fund’s portfolio, not ownership of this company. Private investment history does not establish current ownership.
needs primary source validation — no X/Reddit signal yet
Lower-confidence snapshotTarget price: $599.33
Dated research targetMoat & position
- Evidence Grade
- Inferred
- Moat Durability
- Qualification barrier
- Toll Road Tier
- Strong proxy
Timestamped research update
Deep research update
3,254 words · Updated 19 Oct 2026 · 6 sources
Fabrinet is the precision optical-packaging and optical/electro-mechanical contract manufacturer (EMS) for the AI optical interconnect buildout — it assembles the 800G/1.6T optical transceivers, silicon-photonics modules, and coherent DCI (400ZR/800ZR/ZR+) subsystems that move data inside and between AI datacenters. Customers are the merchant transceiver makers (Cisco, Nokia, Lumentum) and hyperscalers/accelerator vendors (Amazon/AWS, NVIDIA) directly.
- Q4 FY2026 (ended 06-26): revenue $1,315.8M +44.6% YoY; GAAP net income $139.3M; GAAP EPS $3.83; non-GAAP EPS $4.10 (+54.7% YoY); non-GAAP GM 12.2%, non-GAAP op margin 10.9%.
- FY2026: revenue $4,641.1M +35.7%; GAAP net income $473.0M; GAAP EPS $13.05; non-GAAP EPS $14.09 (+38.6%).
- Q1 FY2027 guidance (ending 09-25-26): revenue $1.375–1.425B (+4.5–8.3% QoQ); non-GAAP EPS $4.10–4.25 (36.3M diluted shares).
- Balance sheet (06-26-26): cash + ST investments $875.1M; inventories $1,021.2M; PP&E net $615.1M; no long-term debt; shareholders' equity $2,454.3M. Pillar Two (OECD global minimum tax) provision $57.4M in FY26 — a Thailand-cost-structure headwind.
- monetization model: Contract manufacturing/services; build-to-order/consignment and turnkey optical/electro-mechanical/electronic assembly. Revenue recognized on product transfer; no subscription/SaaS. (10-K FY2025 Item 1, sec.gov ; accession: see EDGAR index; 10-Q Q3 FY2026 Revenue Recognition, sec.gov )
- revenue mix: Single reportable segment; disaggregated by market. Optical communications largest at ~79% of FY2025 revenue; non-optical remainder ~21% (automotive, industrial lasers, sensors, etc.). (10-K FY2025 Revenue/segment note, same EDGAR index)
- who pays: OEM customers that outsource complex manufacturing; Fabrinet does not sell directly to end users. (10-K FY2025 Item 1 Customers)
- margin_value_capture: Gross margin ~12%, operating margin ~10%; value accrues via process know-how, Thailand/scale, supply-chain and test integration. (10-K FY2025 MD&A; 10-Q Q3 FY2026 MD&A)
- customer classes: optical communications OEMs, Industrial Automation - OEMs (industrial lasers/sensors), automotive OEMs, medical/other OEMs.
Fabrinet is an optical-manufacturing company. Its business model is not that of a chip designer or an optical-component brand. It is a manufacturing and supply-chain partner to OEMs that design optical modules, transceivers, lasers, sensors, and related subsystems. The company generally operates build-to-print: the customer owns the design, and Fabrinet owns the process, assembly, alignment, testing, reliability, and logistics challenge of manufacturing at scale.
The exact current segment split must be taken from the Fabrinet 10-K MD&A and segment note. The table below shows planning-level memory estimates only; do not treat these as verified:
| Segment | Estimated revenue mix | Verification status | Source |
|---|---|---|---|
| Optical communications | approximately 75%–80% | needs primary source validation | Fabrinet 10-K index, CIK 0001408710, accessed 2026-05-09 |
| Lasers, sensors, and other | approximately 20%–25% | needs primary source validation | Fabrinet 10-K index, CIK 0001408710, accessed 2026-05-09 |
Source: Fabrinet EDGAR filing index:
https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001408710&type=10-K&dateb=&owner=include&count=40
Filing date/accession number: not asserted in this session; needs primary source validation.
Fabrinet’s revenue is concentrated in optical communications because it manufactures for optical module and component OEMs, including coherent optics, datacom transceivers, telecom transceivers, and related products. The non-optical portion typically includes industrial lasers, automotive sensors/lidar, and other precision devices.
Fabrinet is paid for manufacturing services and for the product it ships. Because it generally procures bill-of-materials components, its reported revenue includes significant material pass-through. This is structurally important: revenue may be high, but gross margin is lower than a fabless technology company because a large portion of revenue is the cost of materials.
Source: Fabrinet 10-K index,
https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001408710&type=10-K&dateb=&owner=include&count=40; filing details not asserted in this session.
Exact gross and operating margin percentages are not verified in this session. They must be pulled from the income statement and MD&A.
Planning-level memory estimate: Fabrinet’s gross margin has historically been in the low double digits, with operating margin below gross margin but still above many larger electronics manufacturing services peers. This would be consistent with a precision optical manufacturing mix, but the exact number is needs primary source validation.
| Metric | Memory-level estimate | Verification status | Source |
|---|---|---|---|
| Gross margin | low double digits | needs primary source validation | Fabrinet 10-K index |
| Operating margin | likely high single digits to low double digits | needs primary source validation | Fabrinet 10-K index |
Fabrinet’s paying customers are OEMs in optical communications, industrial lasers, and sensors. These OEMs generally pay Fabrinet for qualified production capacity, supply-chain management, assembly, test, and logistics.
Fabrinet’s contracts are negotiated with OEM procurement and supply-chain organizations. Pricing is generally SKU-specific and depends on material cost, volume, complexity, yield, and qualification requirements.
I will not invent customer names. The 10-K customer-concentration disclosure is the only acceptable source for top customers. That disclosure was not pulled in this session.
Fabrinet sells to OEMs, not to end consumers or enterprises directly.
The main customer classes are:
- Optical communications OEMs — suppliers of telecom and datacom optical modules.
- Telecom equipment vendors — through their optical module suppliers, not always directly.
- Datacom/hyperscaler supply chain — Fabrinet may manufacture modules that end up in cloud and AI datacenter networks, but it generally does not sell directly to the hyperscaler.
- Industrial laser OEMs — suppliers of fiber lasers, solid-state lasers, and laser systems.
- Sensor and lidar OEMs — automotive and industrial sensing suppliers.
The day-to-day buyer is usually:
- OEM supply-chain management
- OEM procurement
- OEM manufacturing engineering
- OEM quality/reliability engineering
The economic buyer is often the OEM’s executive team, because manufacturing outsourcing affects cost, gross margin, inventory, and time-to-market.
No customer names are cited here as verified because I have not pulled the current Fabrinet 10-K customer-concentration disclosure in this session.
The correct procedure is:
- Pull Fabrinet’s 10-K from EDGAR.
- Find the customer-concentration disclosure, usually in the risk factors or revenue recognition note.
- Extract only names that the filing itself discloses as greater than 10% of revenue.
- Verify the same disclosure in the most recent 10-Q if available.
Until that is done, any named customer is needs primary source validation and should not be used for scoring.
Fabrinet - optical transceiver EMS (800G/1.6T optics supply chain; pump lasers, InP substrates).
Next catalyst is Q1 FY2027 print (fiscal Q4 FY2026 reported 2026-08-17: $1.316B rev +45% YoY; FY2026 $4.64B +36%; DCI run-rate >$1B; capacity plan $12.5-14B). Watch 800G→1.6T merchant-transceiver second-customer production (management: Dec quarter) and NPO vs CPO mix. Parked pending panel.
- Capacity is the constraint (run-rate vs capacity): Q4 FY2026 revenue $1,315.8M → ~$5.26B annualized; CEO (Rosenblatt "Age of AI" conf., 08-20) pegged current capacity at ~$5.8B → ~91% utilization. Building 10 adds ~$3.0–3.5B, plus Nava Nakorn + acquired Santa Clara ops ~$250M each → total capacity ~$8.8–9.3B. Headroom to roughly double revenue is gated by construction/qualification, not demand. (Capacity figures = CEO statement via secondary; revenue = 8-K EX-99.1, measured.)
- The margin ceiling — profit per AI dollar: GAAP gross margin 12.0%, GAAP operating margin 10.0% (FY2026, EX-99.1) → for every $1 of optical transceiver value Fabrinet captures ~$0.10 of operating profit. The IP rent accrues upstream (DSPs to Marvell/Broadcom, EML lasers to Coherent/Lumentum); Fabrinet's 12% GM is flat YoY despite +36% revenue — the EMS no-pricing-power signature.
- NVIDIA de-concentration (the real story): top customers FY2026 = Cisco 19.9%, NVIDIA 16.3%, Nokia 10.7%, Amazon 10.5% (10-K, = 57.4% top-4). NVIDIA trajectory: 35.1% (FY24) → 27.6% (FY25) → 16.3% (FY26). In dollars: NVIDIA ≈ 27.6% × $3,419M = $943.6M (FY25) → 16.3% × $4,641M = $756.5M (FY26) = -20% absolute decline while total revenue +36%. The entire growth is non-NVIDIA (Cisco SiPh, Amazon warrant-locked, Nokia, and DCI).
- Working-capital intensity of the ramp: inventories $1,021.2M, +76% YoY (22% of revenue) and AR $1,017.9M — cash is locked in EML-laser and component inventory for the 800G/1.6T ramp (10-K balance sheet + cash-flow: -$442.9M inventory change).
- Cost of the moat: capex $252.5M (5.4% of rev) in FY26 vs $121.1M (3.5%) FY25 — the Thailand campus build-out; FCF collapsed to $4.2M (vs $207.3M FY25) as capex consumed operating cash. Post-FY-end, a THB 2.5B (~$75M) term loan with Bank of Ayudhya (Krungsri) dated 2026-08-17 funds the expansion (8-K EX-10.2).
- Warrant economics ("cost of moat vs moat defended"): $39.3M warrant fair value (~1.1% dilution at 36.3M sh) defends ~$487M of Amazon revenue (10.5% × $4,641M) = 0.08× — a cheap retention decision; Amazon is effectively underwriting Fabrinet's capacity ramp.
| Parameter | Value | Source | Tier | Measured/derived/assumed | Cross-verified |
|---|---|---|---|---|---|
| Fabrinet fiscal-year revenue | needs primary source validation | Fabrinet 10-K index, CIK 0001408710, accessed 2026-05-09 | Tier 1 | measured | no |
| Optical communications revenue mix | needs primary source validation | Fabrinet 10-K index | Tier 1 | measured | no |
| Lasers, sensors, other revenue mix | needs primary source validation | Fabrinet 10-K index | Tier 1 | measured | no |
| Gross margin | needs primary source validation | Fabrinet 10-K index | Tier 1 | measured | no |
| Operating margin | needs primary source validation | Fabrinet 10-K index | Tier 1 | measured | no |
| Top customer concentration | needs primary source validation | Fabrinet 10-K risk factors/concentration note | Tier 1 | measured | no |
| Competitor gross margin | needs primary source validation | Sanmina 10-K index, CIK 0000898182, accessed 2026-05-09 | Tier 1 | measured | no |
| SKUs re-qualified, \(n\) | 10–20 in worked scenarios | none currently | Tier 3/assumed | assumed | no |
| Qualification time, \(q\) | 6–12 months in worked scenarios | none currently | Tier 3/assumed | assumed | no |
| Monthly qualification cost, \(m\) | $50,000–$100,000 in worked scenarios | none currently | Tier 3/assumed | assumed | no |
| Annual revenue at risk, \(r\) | $50,000,000 scenario | none currently | Tier 3/assumed | assumed | no |
| Fabrinet price premium, \(p\) | 3%–5% in worked scenarios | none currently | Tier 3/assumed | assumed | no |
| Switching-cost payback, \(T\) | 0.6–16.0 years in sensitivity | derived from formula | Tier 3 | derived | no |
Fabrinet controls the qualified precision optical-assembly capacity (sub-micron fiber alignment, die-attach, wire-bonding, SiPh packaging) for 800G/1.6T optical modules in the Western-aligned supply chain. The binding constraint is manufacturing capacity and its build-out timeline, not demand: the company ended FY2026 at a ~$5.3B revenue run-rate against ~$5.8B of current capacity (~91% utilized), so the growth ceiling until Building 10 (Thailand, 2M sq ft) ramps is the capex/capacity curve, not the order book. Fabrinet sits DOWNSTREAM of the real laser bottleneck (EML capacity at Coherent/Lumentum), which is why its capture is a manufacturing toll, not an IP rent.
- Advanced optical packaging at scale — "advanced optical packaging and precision optical, electro-mechanical and electronic manufacturing services" (10-K business description); 800G/1.6T SiPh assembly requires sub-micron alignment and clean-room die-attach that few merchant EMS (vs. Chinese Innolight/Eoptolink) can qualify at hyperscaler volume.
- Amazon warrant (revealed-preference lock-in): on 2025-03-12 Fabrinet issued a warrant to Amazon.com NV Investment Holdings LLC for up to 381,922 shares @ $208.48 (expiring 2032, fair value $39.3M, Black-Scholes: 45.2% vol, 7yr term), vesting in tranches tied to Amazon payments; ~$4.8M/yr amortized as a revenue reduction (10-K Note 7 + "Customer Warrant" cash-flow line). Amazon is paying ~1% dilution to lock in Fabrinet's AWS 800G manufacturing — the same pattern as ATEN's Microsoft warrant.
- Strategic private investment: non-marketable equity securities of $89.1M (a private company, unnamed in the 10-K) with $32.4M net additions + $56.7M unrealized gain in FY2026 — a strategic optical/photonics bet (Note 7).
- Thailand manufacturing base + diversification: facilities in Thailand, US, China, Israel; ~$875M cash + ST investments, zero long-term debt at FY-end (10-K balance sheet) funding expansion organically.
bypass 2 (qualified optical EMS is sticky but bypassable — Chinese EMS, hyperscaler in-sourcing, CPO) / concentration 2 (tight 3-player merchant optical manufacturing oligopoly, Western-aligned leader but not a sole chokepoint) / hyperscaler 2 (hyperscalers are both deepening customers AND the CPO displacement threat; diversification mitigates single-hyperscaler dependence) = 6 STRONG PROXY (consistent with PA Layer-6 6+ range; capped below 7–8 DIRECT BOTTLENECK by the EMS 12%-GM margin ceiling).
Hypothesis verdict: INTACT — the differentiated evidence validates the thesis on growth (record $1.316B quarter +45%, FY26 +36%, ~91% capacity utilization, Amazon warrant, DCI $1B run-rate, Building 10 +$3.0–3.5B) but tempers it on capture (12% GM EMS economics, NVIDIA wallet -20% absolute while total +36%, FCF ≈ 0, CPO transition risk). Fabrinet is a real precision-manufacturing toll on the AI optical buildout, but it is a 6-STRONG-PROXY / low-margin toll, not an IP-monopoly toll — the -22% price drawdown is the market repricing that distinction.
- product_service: Advanced optical packaging and precision optical, electro-mechanical, and electronic manufacturing services.
- technical function: Assembles, packages, integrates, and tests optical modules/transceivers, lasers, sensors, and complex electro-mechanical assemblies to customer designs.
- why it is hard: Requires advanced optical alignment, hermetic/cleanroom packaging, automation, process qualification, and scale—difficult for OEMs to replicate cost-effectively.
- substitutes: OEM in-house manufacturing; other EMS/ODM providers (e.g., Jabil, Flex, Sanmina, Celestica, Asian optical contract manufacturers).
- pricing power: LOW — contract manufacturer with customer concentration and low-teens gross margin; pricing is quote/cost-driven.
Fabrinet sells manufacturing and supply-chain services for advanced optical and precision products. It does not sell a consumer product under its own brand.
Fabrinet’s output includes:
- Optical communication components and modules
- Coherent optical transceivers
- Datacom transceivers
- Telecom transceivers
- Active optical cable assemblies
- Photonic integrated circuit packaging
- Industrial laser systems and subassemblies
- Sensors and lidar modules
- Precision electro-mechanical assemblies
These products are manufactured to customer specifications. Fabrinet provides manufacturing process development, new product introduction, supply-chain management, assembly, active and passive optical alignment, test, reliability screening, packaging, and shipping.
Source: Fabrinet EDGAR filing index, CIK 0001408710, accessed 2026-05-09. Specific product list is from the company’s general SEC description; exact current product mix needs primary source validation.
The core manufacturing challenge is optical alignment. Optical modules require precise alignment of fibers, lenses, lasers, photodetectors, and waveguides. Alignment tolerances are often sub-micron. Fabrinet uses active alignment, passive alignment, automated test, and cleanroom assembly.
The technical function can be summarized as:
- Customer supplies a design and qualification package.
- Fabrinet develops the process and purchases materials.
- Fabrinet assembles and aligns the device.
- Fabrinet tests for optical and electrical performance.
- Fabrinet packages and ships under the customer’s brand.
Substitutes include:
- In-house manufacturing by the OEM
- Other electronics manufacturing services providers, such as Sanmina, Benchmark, Jabil, Flex, and Celestica
- Lower-cost assembly locations
- Automated photonic packaging competitors
- Original design manufacturers that offer both design and manufacturing
The key substitution risk is not simply moving to another factory. Optical manufacturing requires re-qualification. The OEM must qualify the new production line, materials, and test process. This creates a switching cost.
Fabrinet is hard to replicate because:
- Precision optical alignment requires specialized process know-how.
- Customer qualification cycles are long and expensive.
- High-mix, low-volume optical production reduces the attractiveness for commodity EMS players.
- Yield and reliability are process-dependent; a lower-yield line erases the labor-cost advantage.
- Fabrinet has a large Thailand manufacturing campus and trained optical workforce.
These factors are structural and generally described in Fabrinet’s 10-K risk factors and business description. Specific capex, headcount, capacity, and qualification-cycle figures are needs primary source validation.
Fabrinet likely has moderate pricing power, but it is constrained:
- When optical capacity is scarce, Fabrinet can push pricing and prioritize higher-margin programs.
- However, OEMs are large, sophisticated buyers and can qualify alternative suppliers over time.
- Because material pass-through is large, percentage gross-margin changes can be driven by raw-material and component pricing.
Whether Fabrinet can raise prices without losing share depends on capacity scarcity, qualification lead times, and customer concentration. This is not directly verifiable without the 10-K/10-Q and customer disclosures.
needs primary source validation — pending deep-dive research. Placeholder: (1) thesis chain breaks on primary-source evidence, (2) named revenue/customer concentration risk, (3) margin compression or competitive displacement
- Merchant optical EMS vs. Chinese Innolight + Eoptolink (hold ~60% of NVIDIA's 800G wallet — secondary/industry); Fabrinet is the leading Western-aligned precision-optical assembler.
- CPO displacement threat: NVIDIA Quantum-X/Spectrum-X and Broadcom Tomahawk 6 push co-packaged optics on-package, which could shrink the merchant pluggable-transceiver TAM Fabrinet serves (secondary). Notably, Fabrinet's 10-K has zero mentions of "co-packaged optics," "CPO," "silicon photonics," or "linear pluggable" — a disclosure gap relative to Lumentum/Coherent, who actively discuss NPO/CPO.
- Upstream, the true bottleneck is EML laser capacity (Coherent/Lumentum duopoly, NVIDIA-pre-allocated) — Fabrinet assembles what those lasers enable.
- CPO in-package transition accelerates → merchant 800G/1.6T pluggable TAM shrinks, and Fabrinet's $5.8B→$9B capacity build-out becomes over-built (fixed-cost drag on a 12% GM business).
- NVIDIA wallet rotation continues (already -20% YoY in dollars) → if the 1.6T Blackwell transition moves NVIDIA orders to Chinese EMS or CPO, Fabrinet's Cisco/Amazon/Nokia diversification must fully absorb the gap.
- Margin ceiling holds at 12% → no operating leverage despite +36% revenue (GM flat YoY); if inventory +76% build-out precedes a demand pause, obsolescence/write-downs hit a 12% GM P&L hard.
- FCF ≈ 0 ($4.2M FY26) while capacity capex stays elevated → the "debt-free expansion" is already being funded by shrinking FCF + a new THB 2.5B term loan.
- Pillar Two tax ($57.4M) compresses the Thailand cost advantage that underpins the EMS model.
The following would materially weaken or invalidate a bullish Fabrinet thesis. Each item should be checked against primary filings before making a conviction decision.
If Fabrinet’s 10-K shows one customer above 20%–30% of revenue and that customer is in-sourcing or qualifying a second source, Fabrinet’s revenue and margin could fall sharply. This would directly contradict the “diversified optical manufacturing bottleneck” thesis.
If Fabrinet’s gross margin declines materially over several quarters while optical demand remains strong, it would suggest pricing pressure or inability to pass through component costs. That would weaken the pricing-power assumption.
If Sanmina, Benchmark, or another EMS comparator reports optical-manufacturing margins equal to or better than Fabrinet’s, then Fabrinet’s process advantage may be smaller than assumed. The moat would depend more on customer qualification inertia than superior economics.
If Fabrinet’s optical revenue growth is concentrated in a single datacom transceiver generation or a single customer program, then the revenue base is more cyclical than it appears. The switching-cost model may overstate stickiness once that program ends.
Fabrinet’s optical communications exposure likely includes datacom transceivers tied to AI datacenter buildouts. If hyperscaler capex slows, or if transceiver inventories correct, Fabrinet’s growth could reverse even if Fabrinet gains share.
Fabrinet’s manufacturing is heavily concentrated in Thailand. A natural disaster, political disruption, labor shortage, or supply-chain disruption could halt production. If the 10-K risk factors disclose more concentration than the market assumes, the risk premium should rise.
Even if reported net income is strong, the thesis would weaken if operating cash flow is materially lower than net income due to inventory build, receivables, or customer-owned inventory changes. This must be checked against the statement of cash flows.
Any announcement by a major optical OEM that it is expanding in-house manufacturing would directly threaten Fabrinet. The switching-cost calculation shows that in-sourcing can be expensive, but it is not impossible for large OEMs with enough volume.
Primary and supporting material
Sources
6 sources preserved from the latest qualified research update.