$FN EXECUTIVE OVERVIEW
The 2 OFC 2026 Fabrinet sessions materially strengthen the view that the company’s current acceleration is being driven by multiple live programs rather than by a single cyclical optical spike. Official Q2 FY26 results already showed revenue of $1,132.9M, up 36% y/y, non-GAAP gross margin of 12.4%, non-GAAP operating margin of 10.9%, and Q3 revenue guidance of $1.15B-$1.20B, which implies another quarter of roughly 35% y/y growth at the midpoint. The OFC sessions add granularity on where the growth is coming from, which bottlenecks are temporary versus structural, and how much incremental capacity is already being put in place. The most important conclusion is that Fabrinet can now be analyzed as an advanced AI interconnect and compute manufacturing platform with 4 visible growth engines, namely non-DCI telecom, DCI, datacom normalization, and HPC, plus 3 material but later options in OCS, CPO, and hyperscaler-direct designs. (Fabrinet)
That conclusion does not justify treating Fabrinet like an upstream optical component supplier. The economic model remains unmistakably contract-manufacturing-like. Gross margin is still low-12%, customer pricing remains disciplined, and future margin improvement was framed as a function of mix, vertical integration, and operating leverage rather than opportunistic repricing. The correct investment question is therefore not whether Fabrinet can suddenly become a 20%+ operating-margin photonics company. The correct question is whether it can sustain elevated revenue growth, asset turns, and content gains across a much larger AI-oriented opportunity set than the market likely appreciated 12 months ago. (Fabrinet)
WHAT IS ACTUALLY NEW IN THE OFC TRANSCRIPTS
The 2 sessions were directionally consistent on the most important points, which raises confidence that the remarks were deliberate rather than isolated conference rhetoric. Management repeatedly emphasized 2-3 years of demand visibility from customers, while still maintaining only 1-quarter formal guidance. It described the current datacom softness as supply-constrained rather than demand-constrained. It was unusually explicit in naming AWS as the core current HPC customer and NVIDIA, Ciena, Cisco, and Lumentum as major current or prospective vectors. Historically, a key objection in the FN debate has been that the company was difficult to underwrite beyond the next 1 or 2 quarters. These sessions do not eliminate opacity, but they narrow it materially.
The other major change is conceptual. The company is no longer describing itself primarily through legacy segment labels. The more accurate framing is a neutral manufacturing platform serving AI networking, AI compute, and advanced optical packaging across multiple architectures. That matters because it reduces dependence on any single form factor. The base case no longer requires immediate CPO monetization. The near-term thesis can already be supported by businesses that are either in the numbers today or directly constrained by supply rather than by end demand.
REVENUE MIX: THE KEY MISREAD
The quantitative mix shift is the key analytical point. In Q2 FY26, optical communications were $832.6M, or 73.5% of total revenue, and non-optical were $300.3M, or 26.5%. Within optical, telecom represented 67% and datacom 33%. Telecom was $554.4M, or 48.9% of total revenue; datacom was $278.1M, or 24.5%; DCI, which is reported inside telecom, was $142.2M, or 12.6% of total and 25.6% of telecom; HPC was $85.6M, or 7.6% of total. On that basis, datacom plus DCI plus HPC already accounted for about 44.7% of Q2 revenue. The practical implication is that the headline datacom line materially understates Fabrinet’s AI, hyperscaler, and interconnect exposure. (Fabrinet)
Non-optical revenue also changed meaningfully. It increased to $300.3M from $186.5M 1 year earlier, a 61.0% y/y increase, showing that Fabrinet is not only participating in optical growth but also expanding into compute-adjacent manufacturing. Automotive at $117.0M and industrial laser at $41.4M still provide ballast, but they are no longer the marginal drivers of the story. The company is diversifying by manufacturing category while converging more tightly toward the same AI and cloud capex end markets. (Fabrinet)
TELECOM AND DCI
Telecom is now the largest and most underappreciated near-term engine. Officially, telecom grew 17.0% q/q and 59.3% y/y to $554.4M in Q2 FY26. DCI grew only 3.0% q/q to $142.2M. That means about $76.5M of the $80.6M sequential telecom increase came from non-DCI telecom, which aligns with management’s comments around Ciena, Cisco, ROADMs, tunable lasers, and satellite communications. This matters because it shows the growth base is broader than coherent pluggables alone. It is increasingly systems, components, and network infrastructure content tied to AI-driven traffic growth across the broader optical stack. (Fabrinet)
The external read-through from Ciena strengthens this interpretation. Ciena’s Q1 FY26 materials show direct cloud provider revenue up 76% y/y to 42% of total revenue, RLS revenue and shipments up more than 80% y/y due to AI-driven cloud expansion, and explicit investment in contract manufacturing capacity expansion with benefits targeted for H2 2026 and 2027. Ciena’s OFC materials also show WaveLogic 6 Nano 800G ZR/ZR+ and Vesta CPO products moving forward. This does not prove exact Fabrinet share, but it strongly supports the idea that Fabrinet’s non-DCI telecom ramp is connected to a real, expanding cloud-optical capex cycle rather than to isolated customer programs. (Ciena Corporation)
DCI itself should be interpreted as paused, not impaired. Official DCI revenue of $142.2M in Q2 FY26 was still up 42.2% y/y. Management attributed the modest sequential growth to simultaneous customer transitions from 400ZR to 800ZR. That explanation is credible in the context of broader industry roadmaps that are already commercializing 800G coherent pluggables and moving toward higher lane speeds. The correct inference is not that DCI has peaked. The better inference is that the product-transition cadence has become fast enough to create quarter-to-quarter air pockets even inside a strong secular ramp. (Fabrinet)
DATACOM AND EML NORMALIZATION
Datacom is the area where reported revenue most clearly lags underlying demand. Official datacom revenue was $278.1M in Q2 FY26, up only 1.8% q/q and down 7.0% y/y. Management’s explanation was a shortage of 200G-per-lane EML lasers for the main NVIDIA program, with a 2nd source now approved and improvement expected to begin in the June quarter, more meaningfully in the September quarter. The broader ecosystem supports the notion that this is a supply-qualification issue inside a rapid node transition rather than a demand problem. Coherent is already demonstrating 1.6T pluggables using 200G EML and 3.2T pathways using 400G-per-lane links, while NVIDIA’s Rubin networking roadmap is built around 200G SerDes and photonic fabrics available from partners in 2H 2026. Datacom therefore looks delayed, not broken. (Fabrinet)
The important caveat is that supply bottlenecks are a recurring structural exposure for FN, not a 1-time event. The 10-Q explicitly states that critical materials are often sourced from a single source or limited number of suppliers and that historical shortages have already impaired timely manufacturing. The current EML issue is therefore best viewed as a current manifestation of a durable business risk rather than as a freak event. Datacom recovery should be viewed as probable, but not riskless. A resolved EML bottleneck can simply be replaced by the next constrained photonic input if industry volumes continue to accelerate faster than the upstream supply base. (Fabrinet)
HPC: THE NEW ADJACENCY
HPC is the most important incremental non-optical vector. Official reporting shows HPC revenue moving from $15.4M in F1Q26 to $85.6M in F2Q26, a 456% sequential increase. Management identified AWS as the core current customer and indicated that the present scope is contract manufacturing rather than ODM content. The 10-Q further corroborates a strategic Amazon relationship through a customer warrant issued in March 2025 for up to 381,922 shares at a $208.48 exercise price, with 49,652 shares vested by December 26, 2025. The financial impact of the warrant is small. The strategic implication is larger: Fabrinet is now an aligned manufacturing partner on a meaningful hyperscale compute program, not only an optical supplier to that ecosystem. (Fabrinet)
The HPC expansion matters for 3 reasons. It widens the TAM beyond optics. It proves that Fabrinet’s automation and yield discipline can travel into more complex compute-adjacent assemblies. It also creates a new path to content expansion if AI infrastructure increasingly shifts toward rack-scale systems. At the same time, management made clear that pure rack assembly without substantial Fabrinet content would be margin-dilutive. That is strategically encouraging because it suggests discipline on revenue quality, but it also means Fabrinet may underparticipate in portions of the rack-scale opportunity where it does not own meaningful content.
OCS, CPO, AND HYPERSCALER-DIRECT
The medium-term optionality stack should be ranked by monetization proximity. OCS appears closer than CPO. Lumentum has already disclosed OCS backlog well beyond $400M and described itself as only at the starting line of that opportunity. By contrast, Lumentum’s disclosed incremental multi-hundred-million-dollar CPO order is deliverable in 1H calendar 2027. Ciena’s Vesta 200 CPO engine is at customer-sample stage in Q2 CY2026, and NVIDIA’s Rubin-era photonic networking products are targeted for partner availability in 2H 2026. The ecosystem data therefore support management’s view that CPO is now real, but still later in the revenue curve than OCS and certainly later than the current telecom, DCI, datacom, and HPC ramps. (Lumentum Investor Relations)
That ecosystem validation should not be confused with Fabrinet revenue attribution. Lumentum’s OCS backlog proves demand for OCS, not Fabrinet share. CPO roadmaps at Ciena, Coherent, and NVIDIA prove commercialization momentum, not Fabrinet content scope. Outsourcing capture will depend on make-versus-buy choices, qualification timing, and how far up the stack Fabrinet can climb from deep packaging into larger system content. This is the critical modeling distinction. OCS and CPO are real upside vectors, but they should not yet carry the base case. (Lumentum Investor Relations)
Management’s refusal to choose between pluggables, CPO, copper, and other scale-up or scale-out architectures appears analytically sound. The external roadmaps are clearly multi-track, not winner-take-all in 2026. Coherent is simultaneously demonstrating 1.6T and 3.2T pluggables, XPO pathways toward 12.8T, and multiple CPO approaches. Ciena is advancing both 800G ZR/ZR+ coherent pluggables and pluggable CPO engines. NVIDIA is advancing co-packaged photonic switches. The near-term industry reality is coexistence. That favors a neutral manufacturing platform with broad packaging and assembly capabilities rather than a company overcommitted to a single interconnect architecture. (Coherent Inc)
The most important unanswered question in CPO is scope of content, not technology adoption. Management explicitly analogized the opportunity to the prior move from telecom components toward larger telecom systems. If Fabrinet can move from deep packaging into broader switch or system assembly, the addressable BOM expands materially. If participation remains limited to packaging or optical-engine subassemblies, revenue still grows, but the TAM expansion is smaller. That is the correct way to model CPO: as a scope-of-participation question, not merely a market-size question.
Hyperscaler-direct transceivers are potentially very large but should be modeled with patience. Management described about 1 year to 18 months from real customer need to production shipment on new direct programs. That implies meaningful revenue from newly engaged hyperscaler-direct wins is more likely a FY27-FY28 issue than an immediate FY26 catalyst. The strategic appeal is obvious. If hyperscalers allocate even a modest share of their transceiver volume to customer-owned designs, the content pool is large. The analytical discipline is to treat this as credible option value rather than near-term forecast revenue.
CAPACITY, CAPITAL EFFICIENCY, AND WORKING CAPITAL
Capacity is no longer the obvious limiting factor. The 10-Q and investor materials show cash, cash equivalents, short-term investments, and restricted cash of $961.5M, 0 debt, and ongoing construction of an approximately 2.0M square foot Chonburi facility at an expected cost of about $132.5M. The same filing notes that current manufacturing capacity is sufficient for anticipated production requirements for at least the next few quarters. At OFC, management went much further, stating that Building 10 should support about $3B of additional annualized revenue capacity and that existing campus land could ultimately support about $11.5B of capacity. Even if those figures are treated conservatively, the point remains that Fabrinet is self-funding large additions from a position of balance-sheet strength rather than from leverage or emergency customer financing. (Fabrinet)
This capital profile is a real strategic differentiator. NVIDIA’s strategic partnerships with Lumentum and Coherent each include $2B investments to support capacity, R&D, and U.S. manufacturing, underscoring how capital-intensive upstream optical component production has become. Fabrinet’s expansion problem is structurally lighter because it is primarily adding buildings, process capability, and customer-specific equipment rather than new laser fabs or photonic foundries. That does not eliminate execution risk, but it does imply a much higher revenue-capacity increment per dollar of balance-sheet commitment than in the upstream optical supply base. (Lumentum Investor Relations)
Working capital behavior corroborates demand. Inventories increased by $221.0M in the 1st 6 months of FY26, and management explicitly explained in the 10-Q that the inventory increase was to support higher demand in the next quarter. Operating cash flow remained positive at $148.8M despite that inventory build, and PP&E purchases nearly doubled y/y to $96.9M in the 1st 6 months. Purchase obligations and other commitments also reached $1.94B, with total capital expenditure commitments of $193.7M. This is consistent with a business deliberately leaning into customer ramps rather than reacting after the fact. (Fabrinet)
Management’s claim that an unfilled Building 10 would create only about a 15 bps gross-margin headwind is directionally helpful but should be treated as a simplified sensitivity, not a full downside model. Underutilization can also affect labor absorption, startup yields, and program mix. Even so, the broader conclusion remains valid: relative to the size of the balance sheet and cash generation, the company does not appear overextended on capacity.
MARGIN STRUCTURE AND VALUE CAPTURE
The margin discussion sets the correct valuation frame. Non-GAAP gross margin was 12.4% in Q2 FY26, unchanged from the prior-year quarter, while non-GAAP operating margin improved to 10.9% from 10.6%. That pattern is exactly what a high-quality contract manufacturer should exhibit during a strong ramp: modest gross-margin movement, stronger operating leverage, and incremental gains from mix and vertical integration rather than from price extraction. The implication is that the equity case should be underwritten primarily on revenue durability, utilization, capital efficiency, and incremental operating margin, not on a thesis of supplier-like gross-margin expansion. (Fabrinet)
The correct way to interpret low gross margin is not low quality. Presentation materials show ROIC at 42.0%, which is exceptionally strong and reflects the combination of high asset turns, tight overhead, and disciplined capital allocation. The business can therefore create substantial equity value even without product-company gross margins. What should not be expected is a wholesale change in the margin regime unless Fabrinet materially changes where it sits in the stack, which management has repeatedly indicated it does not intend to do through an ODM or proprietary-IP strategy. (Fabrinet)
The strategic refusal to become an ODM or product company is both an advantage and a limit. It likely improves trust with OEMs and suppliers that would not want their contract manufacturer to become a competitor. It also means that if future hyperscaler economics increasingly favor in-house IP paired with ODM-style system design or turnkey rack integration where the manufacturer owns more architecture, Fabrinet will voluntarily cede part of that margin pool. That discipline is strategically coherent. It is not free.
RISK FRAMEWORK
The primary risk is concentration, not demand scarcity. The 10-Q states that 4 customers represented 59.1% of quarterly revenue in the December 2025 quarter, and 3 customers represented 48.9% of revenue for the 1st 6 months of FY26. That concentration is rising even as the business appears more diversified by segment label. The underlying reason is that many of the growth vectors, including datacom, DCI, HPC, and an increasing share of telecom systems, are all ultimately tied to a relatively small group of hyperscaler and cloud-linked programs. Segment diversification is real. End-demand diversification is less so. (Fabrinet)
The 2nd risk is working capital and inventory timing. Inventory increased $221.0M in 6 months, purchase obligations reached $1.94B, and the 10-Q reminds that most customer-specific materials cannot easily be repurposed across programs. If demand timing slips, if a major customer changes architecture, or if a ramp is overestimated, the balance sheet absorbs the first impact even if contractual protections eventually reduce losses. This is not an immediate bear-case signal, but it is the correct place to look for early evidence of demand digestion. (Fabrinet)
The 3rd risk is geographic and regulatory concentration. Long-lived assets are overwhelmingly located in Thailand, with $414.4M of $465.0M total long-lived assets there as of December 26, 2025. The 10-Q also notes that export-control restrictions could in principle be extended to Thailand, which would directly affect the feasibility of manufacturing certain products there. This concentration has historically been a competitive advantage. It is also a single-region operating dependency. (Fabrinet)
The 4th risk is timing error on the optionality stack. OCS, CPO, and hyperscaler-direct transceivers are all plausible, but none should be capitalized aggressively into near-term estimates until formal revenue conversion becomes visible. Management itself declined to guide CPO by category and framed 1-year to 18-month engagement cycles for new hyperscaler-direct wins. The correct interpretation is that these are credible options, not yet core forecast pillars.
BOTTOM LINE
The central conclusion from the OFC transcripts is that Fabrinet’s near-term growth case is stronger and broader than a superficial reading of the datacom line would suggest. The base case no longer needs speculative CPO success. It can already be underwritten by 1) continued non-DCI telecom strength, especially Ciena, Cisco, and optical component content, 2) resumed DCI growth after 400ZR-to-800ZR transitions normalize, 3) datacom recovery as EML supply improves, and 4) AWS HPC ramp with possible further share gains. OCS, CPO, and hyperscaler-direct designs are upside to that base rather than prerequisites for it. (Fabrinet)
The correct debate for an investment committee is therefore not whether Fabrinet participates in AI infrastructure. It already does at scale. The real debate is whether the company can sustain elevated growth while preserving CM-like discipline, avoiding component bottlenecks, and converting future architectural shifts into additional content rather than watching value migrate upstream to IP owners or downstream to full-rack integrators. The OFC evidence materially improves confidence on the 1st point, is constructive but not definitive on the 2nd, and leaves the 3rd as the major medium-term open question. On balance, the transcripts are clearly positive for growth durability, moderately positive for medium-term optionality, and largely unchanged on the structural margin ceiling.
