FN
Fabrinet
Summary
What they do:
Contract manufacturer specializing in precision optical, electro-optical, and electronic assemblies — building the optical transceivers, silicon photonics modules, and photonic components that carry data between AI servers at the speed of light, for customers including Coherent, Lumentum, NVIDIA, and other optical leaders.
Why they matter:
Fabrinet is the hands that build the optical layer. When Coherent designs an 800G transceiver or iPronics develops a silicon photonics optical circuit switch, Fabrinet manufactures it at scale with the precision and yield required for data center deployment. Record Q2 FY2026 revenue of $1.13B (+36% YoY) — optical comms $833M (telecom $554M, datacom $278M) plus a surging non-optical segment at $300M driven by an $86M HPC program with AWS — confirms Fabrinet is riding the AI optical infrastructure wave while diversifying into adjacent compute manufacturing.
Recent performance:
Q2 FY2026 (Dec quarter) record revenue $1.13B, up 36% YoY. Non-GAAP EPS $3.36 (beat $3.25 consensus). Q3 guided $1.15–1.20B with EPS $3.45–3.60. Stock ~$554, market cap ~$20B. Stock up 224% over the past year.
Our Verdict
The premier optical contract manufacturer riding 36% revenue growth on 800G/1.6T transceiver demand and silicon photonics expansion — but at ~5x trailing revenue and ~40x forward P/E after a 224% run, the stock prices in near-perfect execution.
Structural trends
Structural
77
/ 100
Moat
6/10
Optical specialization + yield track record + customer qualification, but CM model limits pricing power
AI Exp.AI Exposure
High~40% AI
Play Type
EstablishedAI Growth
~35%
Rel. Value
76
COMPELLINGInside every AI data center, thousands of optical transceivers convert electrical signals from GPUs and switches into light pulses that travel through fiber optic cables at the speed of light. These transceivers are extraordinarily precise devices — each one contains lasers, photodetectors, driver ICs, and optical elements that must be assembled with micron-level accuracy. Fabrinet builds them.
The company operates as a precision contract manufacturer based in Thailand, with major campuses at Pinehurst and Chonburi. Unlike a generic electronics manufacturer, Fabrinet's cleanrooms, alignment equipment, and skilled workforce specialize in the exacting requirements of photonic devices. In Q2 FY2026, optical communications contributed $833M (~74% of revenue) — broken into telecom at $554M (record, +59% YoY; includes DCI modules at $142M, +42% YoY, primarily 400 ZR and 800 ZR coherent modules) and datacom at $278M (-7% YoY but inflecting, +2% QoQ). Non-optical communications contributed $300M (~26% of revenue, +61% YoY), driven by a breakout high-performance computing program ($86M, up from $15M in Q1), automotive ($117M), and industrial lasers ($41M).
The AI optical story is straightforward. As AI clusters scale to tens of thousands of GPUs, the interconnect bandwidth required between racks, switches, and storage grows exponentially. This bandwidth is carried by optical transceivers — currently 800G and 1.6T, both at the 200G-per-lane node. Fabrinet manufactures these transceivers for companies like Coherent, Lumentum, and increasingly for silicon photonics customers. Management stated 800G and 1.6T are produced on the same production line and are "very similar products" — the company is indifferent to mix between the two. Supply has been constrained by EML laser availability at the leading 200G-per-lane node, but a second laser source was qualified during Q2, which should ease constraints going forward. The iPronics partnership announced in March 2026 added a dedicated manufacturing line for silicon photonics-based optical circuit switching systems designed for AI data centers, and Fabrinet disclosed it is working on OCS projects with multiple customers.
Beyond optical, the HPC program with AWS (Fabrinet is a second source) is ramping rapidly across fully automated production lines, with management guiding to a >$150M quarterly run rate when fully ramped over the next two quarters. Two automated lines are qualified, with additional lines in qualification. The relationship is non-exclusive, and Fabrinet is pursuing other HPC customers. CEO Grady noted that if Fabrinet exceeds expectations on cost, quality, and delivery, it could earn a larger share of the current AWS program.
Q2 FY2026 results were exceptional — record revenue of $1.13B, up 36% YoY, with non-GAAP EPS of $3.36 beating consensus. Q3 guidance of $1.15–1.20B implies continued acceleration (~35% YoY growth at midpoint). Management stated visibility is "the furthest we have ever had." Wolfe Research upgraded the stock to Outperform, and the analyst consensus target sits at $582 with zero sell ratings.
Supply Chain Dependencies
The Catch
Fabrinet doesn't own the designs — it's a contract manufacturer. This means customer concentration risk (Coherent is the largest optical customer; AWS is now a major non-optical customer), limited pricing power (12.4% gross margins with 20-30bp FX headwinds), and vulnerability to insourcing or competitive bids from lower-cost Asian CMs. The AWS HPC program is explicitly a second-source position — the incumbent could retain or recapture share. Free cash flow turned negative in Q2 as Building 10 CapEx ramps (~$130M total for 2M sq ft). At ~40x forward P/E and a 224% one-year run, the stock is priced for sustained 25%+ growth. If the AI optical buildout slows or a major customer shifts volume, the correction would be severe.
If They Win
If Fabrinet becomes the default precision manufacturing partner for every 800G, 1.6T, CPO, and optical circuit switching product entering AI data centers, while HPC manufacturing scales to multiple hyperscaler customers, revenue reaches $6B+ by FY2029. Building 10's 2M sq ft ($2.5B+ capacity) fills, and Buildings 11 and 12 (already zoned on the same Chonburi campus) begin construction. Operating leverage continues improving as scale grows faster than OpEx. Fabrinet is either a $25B+ standalone company or an acquisition target for a larger technology company seeking vertically integrated optical and precision compute manufacturing.
Others in Transmit the Data
Not financial advice. All scores generated via AI algorithms using public data.