Earnings Call Highlights: Fabrinet (FN) | Q4 FY2026 — Quarterly Revenue Tops $1.3B, YoY Growth Accelerates to 45%, and the Long-Term Capacity Roadmap Stretches to $14B

2 days ago

9 min read

What Fabrinet delivered this quarter was not just a "pretty good report" but a manual that recalibrates its capacity ceiling for the next three years. Quarterly revenue reached $1.316B, up 45% YoY, the sixth consecutive quarter of accelerating growth. But the number to circle is the long-term capacity ceiling: last quarter management talked about $11.5B; now it is $12.5B–$14B. When a contract manufacturer starts telling its story in terms of land rather than orders, the demand it sees has outgrown what its existing buildings can absorb.

1. Three Key Takeaways

First, growth is not riding on a single customer. Four customers each exceeded 10% of FY2026 revenue: Cisco 20%, NVIDIA 16%, Nokia 11%, Amazon 11%. An analyst pointed out in Q&A that NVIDIA actually declined more than 20% for the full year. In other words, excluding NVIDIA, the rest of the business grew close to 60%. The market has long treated Fabrinet as an "NVIDIA proxy" contract manufacturer; this quarter's numbers formally retire that label.

Second, the revenue breakdown changed, and in an interesting way. Starting this quarter, Fabrinet reports three segments: Data Center, Communications Infrastructure, and Automotive/Industrial/Other. DCI (data center interconnect), previously under telecom, has moved into Data Center. This is more than an accounting move; management is telling the market: our end customers are hyperscale data center operators, not telcos.

Third, NPO was explicitly framed as a nearer-term opportunity than CPO. CEO Seamus Grady put it plainly: Near-Packaged Optics (NPO) sits between pluggable transceivers and Co-Packaged Optics (CPO), and Fabrinet has capabilities on both sides. It has built tens of millions of pluggable modules and is already producing CPO components for a small number of customers (not yet at full volume). The line "NPO is a nearer-term opportunity than CPO" carries real weight for how the whole silicon photonics supply chain should think about timing.

2. Revenue and Financials

Core quarterly figures (non-GAAP, the basis the market typically uses for Fabrinet):

For the full year, FY2026 revenue was $4.64B (+36% YoY) and non-GAAP EPS was $14.09 (+39% YoY). Earnings grew faster than revenue, evidence that operating leverage is real. Operating expenses were just 1.3% of revenue, and the 10.9% operating margin is a three-year high. CFO Csaba Sverha stressed a line worth remembering: this growth model does not require a proportional increase in operating expenses.

On a GAAP basis, two one-time items offset each other this quarter and need to be separated: first, a $56.7M non-cash gain from revaluing the Raytek investment (a book gain, no cash received); second, a $57.4M top-up tax provision recorded in Thailand under the OECD global minimum tax (no cash actually paid in FY2026). Net of the two, GAAP net income was $139.3M and EPS $3.83. The second item is worth tracking: Thailand's implementing rules and offsetting investment incentives are still evolving, future tax burden and related compensation could shift again, and a meaningful part of Fabrinet's cost advantage comes from Thailand.

Cash flow turned negative this quarter: operating cash flow was $55M while capex jumped to $92M (Building 10 construction plus $11M to buy the Navanakorn facility), for free cash flow of −$37M. For the full year, operating cash flow was $257M and free cash flow only $4M. This is not deteriorating fundamentals but a deliberate choice to pour all cash back into capacity. Cash and short-term investments ended at $876M, down $70M QoQ; buybacks were nearly zero this quarter, with about $169M of authorization remaining.

Three Segments Broken Down

Data Center: $669M, +68% YoY, +13% QoQ, 51% of total revenue. It is now the largest segment for the first time, and it houses three engines: DCI, high-performance computing (HPC), and data center optical transceivers. DCI is already above a $1B annualized run-rate, and the CFO added a more vivid comparison: Q4 DCI revenue alone now equals the size of the entire historical Datacom business.
Communications Infrastructure: $413M, +40% YoY, +1% QoQ, 31% of revenue. Growth came from telecom systems, satellite communications, and telecom components: broad-based rather than single-point.
Automotive/Industrial/Other: $234M, +8% YoY, +9% QoQ, 18% of revenue. Sequential growth was driven mainly by EV charging infrastructure, with a small contribution from LiDAR customers. This segment is finally no longer a drag.

3. Capacity Roadmap: The Math from $5.3B to $14B

The most information-dense part of the call came when Seamus was asked "when will you build Building 11?" and simply laid out the whole capacity roadmap:

・Today: Q4 revenue annualized = a $5.3B capacity run-rate

・Pinehurst: converting 120,000 sq ft of office space into production lines → up to $5.5B–$5.8B

・Building 10 (Chonburi, completion early 2027, 2 million sq ft in total): +$3B–$3.5B → cumulative $8.5B–$9.3B. The first floor's 250,000 sq ft is already qualified and producing, and another 250,000 sq ft on the third floor is expected to qualify this quarter, meaning most of this capacity is still ahead

・Navanakorn (acquisition closed in Q4, trial operations just completed, +200,000 sq ft): +$200M–$250M, contributing from Q1

・Santa Clara Great America Place (two office buildings plus ~130,000 sq ft of production, more than doubling the Silicon Valley footprint): +$200M–$250M

・Two more land parcels in Chonburi, each able to host a 1.2 million sq ft building, each worth +$1.8B–$2.1B

Add it all up: $12.5B–$14B. Last quarter the figure was $11.5B. Besides the new Navanakorn and Santa Clara additions, the other half of the gap comes from rising revenue per square foot. DCI products are small and high-priced, classic "revenue dense" items, and together with better space utilization and efficiency, the same buildings now generate more revenue.

Note that Santa Clara is not a volume manufacturing site. Fabrinet West serves as the entry point for new product introduction (NPI); many customers are just down the street, and success is defined by how fast products move to volume production in Thailand, not by how much is built locally. Doubling down in Silicon Valley signals that more new projects are queuing at the front end.

4. Technology and Business Highlights: NPO, Multi-Rail Amplifiers, OCS

NPO and CPO. Seamus's comments on NPO deserve a line-by-line read: NPO technology "sits between pluggable modules and CPO"; Fabrinet has built tens of millions of pluggables; on CPO it is already working with a few customers and producing components, though not yet at full volume. Because NPO blends elements of both, he believes Fabrinet is "well positioned to be a leader in NPO manufacturing and packaging." As NPO scales to 6.4T, 12.8T and beyond, manufacturing complexity and yield become decisive, and turning advanced optical components into high-yield, manufacturable systems is exactly what this company has done for decades. For a full breakdown of how to think about the CPO vs. NPO debate, see NVIDIA's Shift from CPO to NPO: Bad News for the Laser Supply Chain?.

Why the Raytek move matters. Management said Raytek will add capacity inside Fabrinet's Thailand campus. The industry implication outweighs the financial one: Fabrinet wants every packaging capability needed for these future products under its own roof, either by building lines itself or by bringing partners onto the campus. It is vertical integration, executed through a joint-venture-style partnership.

Multi-rail amplifier architecture. Asked about this, Seamus essentially read out his own menu: high-density fiber routing and management, high-volume fusion splicing and connectorization, and a full suite of precision optical manufacturing and assembly processes. Multi-rail packs fiber pairs into highly integrated optical systems with many difficult process steps; he said bluntly that "these process steps are our secret sauce." Fabrinet is already deeply engaged with multiple customers on several multi-rail programs.

Optical circuit switching (OCS). The stance is unchanged but more conservative: shipments are still small, but the technology closely resembles existing products, and once the industry leader's large-scale pull-in begins, Fabrinet is "quite confident it will participate." That is consistent with Lumentum writing OCS and lasers into its reported numbers this quarter. OCS has moved from slides into shipment schedules, just not yet into the main body of the P&L.

Full-system integration. Fabrinet's playbook has always been "components first, then up to PCBA, subsystems, and finally full systems." It already runs two complete network-system businesses, for Infinera (now part of Nokia) and Cisco, with one or two more programs in progress. The key to this model is a high enough share of in-house components; otherwise it is not cost-effective enough for the customer, nor sticky enough for Fabrinet.

5. Management Outlook: Conservative Numbers, Bold Wording

Q1 FY2027 guidance: revenue of $1.375B–$1.425B, implying +43% YoY at the midpoint; non-GAAP EPS of $4.10–$4.25. Management also flagged the usual Q1 expense seasonality, a brief margin headwind, while the direction of operating leverage remains intact.

The wording is worth reading in layers:

・Highest certainty: "All three segments are expected to grow in Q1" — language backed by orders already in view

・Medium: "Customer visibility extends into FY2027 and beyond," though the CFO immediately added "these long-term forecasts are not order commitments" — a standard disclaimer, but also a sign that visibility has genuinely lengthened

・Softest but most informative: an analyst asked directly whether FY2027 growth could exceed FY2026's 36%, and Seamus answered "it's not out of the question." For a company that strictly guides one quarter at a time, letting that line stand in the transcript is itself a signal

The timeline for new transceiver programs was quite specific: a hyperscaler-direct program could begin ramping as early as this quarter, a merchant program starts in the December quarter, and the rest ramp in early 2027. This is a staggered ramp spanning the whole fiscal year, not a one-quarter spike.

6. Supply Chain and Customer Clues

Cisco is the largest customer at 20%. Cisco's own report this quarter was also bullish on DCI, and the two corroborate each other. For context, see Earnings Call Highlights: Cisco (CSCO) | Q4 FY2026.

Nokia becoming a 10% customer for the first time is the most easily overlooked change this quarter. Seamus explained it in two layers. First, after Nokia acquired Infinera, Fabrinet's existing Infinera business was not lost to product-line consolidation (he noted this often happens in acquisitions and is a risk vendors cannot control). Second, Fabrinet historically did little with Nokia's core business and is now winning new programs there. He said "we feel it's just the beginning." Nokia itself doubled its AI & Cloud revenue this quarter, so the growth vectors line up.

Amazon, at 11%, is the HPC customer. Seamus named AWS directly, saying Fabrinet is introducing the customer's next-generation silicon platform while adding capacity to support the technology transition and new products. This line is "performing above expectations."

Component supply remains a constraint. Management declined to speak for customers or suppliers on specific high-profile parts but acknowledged demand for some components exceeds available supply, stressing that any potential shortfall is already reflected in guidance. Keep this in mind: in the near term, Fabrinet's growth ceiling may not be floor space but the laser or chip upstream.

China optical module ban. The stance was pragmatic to the point of indifference: Fabrinet does no contract manufacturing for Chinese suppliers, so in theory it benefits; but he also said outright that a blanket ban on transceivers from China would "bring the whole industry to a halt," so it is "definitely not settled." We walked through the scenarios in One Rumor Reshuffles the Supply Chain: A US Ban on Chinese Optical Modules.

LEO satellites. Reported under Communications Infrastructure, customers include the two major players in the space plus one or two newcomers. The technology overlaps heavily with existing capabilities: business Fabrinet can capture directly once volumes ramp.

7. Conclusion

Here is how I would rank the three metrics to watch next quarter:

1. Whether Data Center can hold above 50% of revenue, and specifically whether the transceiver business outside DCI keeps pace. New program ramps start this quarter, so Q1 is the first checkpoint.

2. Whether gross margin holds the 12% line. The three-year-high operating margin came from squeezing the expense ratio to 1.3%, but gross margin actually slipped slightly YoY. With new programs ramping and new buildings qualifying, gross margin is where friction shows up first.

3. Follow-through on Thailand's global minimum tax. The $57.4M provision is a book entry with no cash outflow so far, but implementing rules and investment-incentive offsets are still evolving, one of the few directionally unclear variables in Fabrinet's cost structure.

The real point of this quarter fits in one sentence: Fabrinet's narrative has shifted from "who its big customers are" to "how much capacity it can build." When a contract manufacturer lifts its long-term capacity ceiling from $11.5B to $14B in a single quarter, and can map each parcel of land and each building to a revenue figure, its customer forecasts have grown long enough to require planning in land. The risk sits in the same sentence: the math assumes sustained demand, adequate component supply, and no further change to Thai tax rules. If any of those premises loosens, $14B is just a very nice slide.

This article is for technology and industry trend analysis only and does not constitute investment advice.