FN · Fabrinet · Stock research

Current Thesis

Fabrinet builds the transceivers and optical modules that sit underneath the AI-interconnect buildout, and the operating story has not deteriorated: trailing revenue is $4.24B, +29.8% YoY, with a fiscal-Q4 guide of $1.25B–$1.29B implying roughly 40% growth at the midpoint. The problem is that the stock has stopped tracking its own theme. Through mid-July 2026 the optical basket re-accelerated hard — Applied Optoelectronics +390% YTD, Lumentum +138%, Coherent +107%, with LITE printing $806.81 on 2026-07-14 as the Texas-expansion narrative pulled fresh money into the trade — while FN trades near $478 (2026-07-17), up 0.7% on the year and about 36% below the $748.89 52-week high. A name that lags its cluster by 100-plus points of performance during the cluster's best month is not being ignored; it is being priced for a different mix. The next hard resolution is the fiscal-Q4 print on 2026-08-24, and the setup into it is a broken chart attached to an accelerating theme.

Bullish and bearish views on Fabrinet

The model's bull view on Fabrinet (FN), in brief: Telecom is running at records and is genuinely multi-customer. The bear view: Relative strength has failed outright. The cluster's July bid did not lift this name in proportion. When the group leaders are compounding triple digits YTD and FN is flat, the market is signalling that pluggable-module assembly is not where the incremental AI dollar is landing.… Both cases follow in full.

Bull Case

• Telecom is running at records and is genuinely multi-customer. Q3 FY26 (2026-05-04) telecom revenue hit $628M, +55% YoY and +13% sequentially, with Data Center Interconnect at $197M, +91% YoY and +39% sequentially — the 400ZR/800ZR coherent buildout spread across several accounts rather than one.
• The datacom shortfall is a supply problem with a stated fix. Datacom fell ~6% sequentially in Q3 on laser, ASIC and memory constraints while demand exceeded shippable supply; management framed the gap as deferred backlog and guided all major product categories higher in Q4 (2026-05-04 call).
• Two hyperscale 800G programs cleared qualification and are shipping. Both scale-out transceiver programs began direct shipment to a hyperscale customer, with management pointing to meaningful contribution in fiscal 2027 — a step up from anonymous contract assembler to named-program supplier.
• The guide survived the constraint. Q4 FY26 revenue $1.25B–$1.29B and non-GAAP EPS $3.72–$3.87, gross margin near Q3 levels. The bottleneck was acknowledged on the call and the revenue line still steps up ~40% YoY.
• Multiple has already compressed. Forward P/E is ~29 against a trailing ~41 (2026-07-17), versus the ~50x forward the stock carried into the May print. The de-rate has done real work on valuation.
• Raytek adds a CPO hedge. The mid-2026 minority stake in Taiwan's Raytek Semiconductor (advanced wafer-level packaging) puts FN inside the co-packaged-optics supply chain rather than purely opposite it; Rosenblatt lifted its target to $750 on the repositioning (2026-05-05).

Bear Case

• Relative strength has failed outright. The cluster's July bid did not lift this name in proportion. When the group leaders are compounding triple digits YTD and FN is flat, the market is signalling that pluggable-module assembly is not where the incremental AI dollar is landing.
• The chart is still repairing damage. The late-June $555 shelf and the rising 50-day are both gone, and the June–July drawdown ran roughly 25% inside a month. Price closed at $478.39 on 2026-07-17 after an oversold reading below 30 was flagged on 2026-07-08 — a bounce off a washout, not yet a reclaimed base.
• Thin margins turn a revenue wobble into an EPS event. Non-GAAP gross margin ran 12.1% with 10.7% operating margin in Q3. At that structure there is no cushion; a 3% revenue shortfall does disproportionate damage to the EPS line the multiple is built on.
• The market has already shown how it reacts. FN fell 12.88% on the 2026-05-04 session despite beating on revenue, punishing the datacom commentary and the multiple together. That reaction function is unchanged heading into 2026-08-24.
• Target dispersion is a warning. One analyst panel averages ~$749 (high $850, low $635); another shows a $544 median with a $269 low. That spread, unresolved since the de-rate, more often closes by the high marks coming down than by price running 55% to meet them.
• Customer concentration cuts both ways. Cisco is ~16% of FY sales and Lumentum ~15%, with Nvidia and Infinera close behind. Lumentum is both a customer and a cluster peer taking share of the same narrative dollar — its strength does not automatically flow through.
• CPO and linear-drive optics remain a structural overhang. If co-packaged optics displaces pluggable transceiver volume at the hyperscale tier, the unit base FN assembles shrinks; Raytek is optionality against that, not insurance.

Setup & Price Structure

Price ~$478.39 (2026-07-17, +3.95% on the session), against a 52-week range of $272.49–$748.89 and an all-time closing high of $746.47 set 2026-05-14. Market cap $17.14B. The structure is a broken uptrend attempting a first repair: the $555 shelf that held through late June gave way, the 50-day is overhead, and the stock has spent July grinding a low base between roughly $470 and $495 after the RSI-sub-30 flush. The 2026-06-23 cluster shakeout (COHR −9%, LITE −8%, AAOI −13%), which followed a 2026-06-02 melt-up of similar magnitude in the other direction, did not turn out to be a one-session reset for this name the way it did for the leaders — those recovered and made new ground in July; this one did not.

That divergence is the whole read. A base can form here and the stock can retrace a large part of a 36% drawdown into a strong print. But buying a laggard because its peers are working is exactly the reasoning that turns a bounce trade into a bag. The clean structural condition to wait for is a reclaim of the $555 shelf on volume, or a higher low above the July base that holds through the earnings blackout. Neither exists yet.

Catalyst Calendar (next 30 days)

• 2026-08-24 (confirmed, after close) — Fiscal Q4 FY26 results. Fiscal year ended ~2026-06-26. Guide to beat: revenue $1.25B–$1.29B, non-GAAP EPS $3.72–$3.87, gross margin near the 12.1% Q3 level. The line that matters is datacom sequential direction, not the headline.
• ~2026-08-19 — Three trading days ahead of the print; the point past which any fresh position is an earnings bet rather than a narrative one.
• Ongoing, through August — Peer prints and component commentary from Coherent and Lumentum are the real-time read on whether the laser/ASIC/memory constraint is loosening. A peer confirming supply relief before 2026-08-24 is the single most useful pre-print datapoint available.
• Ongoing — Hyperscaler capex updates. Any downward revision to 2027 network capex hits the pluggable-volume assumption directly.

Nothing dated falls inside the next 30 days. The window from here to 2026-08-19 is unusually empty, which is why relative performance versus the cluster carries the information.

What Would Change Our Mind

• A weekly close below $450 would confirm the July bounce failed and mark the de-rate as structural rather than a washout — at that point the drawdown stops being a valuation reset and starts being a re-rating of the business model.
• A second consecutive sequential datacom decline on 2026-08-24, particularly alongside revenue under the $1.25B guide floor, breaks the deferred-backlog argument. Supply constraints that persist for three quarters stop being constraints and start being lost share.
• Constructively: a reclaim of the $555 shelf with the cluster still bid, plus a Q4 print showing datacom turning up sequentially and management pulling the fiscal-2027 hyperscale ramp forward, would rebuild the case — the multiple has already compressed to ~29x forward, so an inflection would not need multiple expansion to work.
• A downward revision cluster from the $700–850 target cohort before the print would confirm the dispersion is resolving toward price rather than away from it.
• Peer commentary confirming component availability ahead of 2026-08-24 would materially raise the odds on the Q4 datacom line and is the cheapest available signal.

Correlation Notes

FN trades inside the optical basket with Coherent, Lumentum, Applied Optoelectronics and, more loosely, Marvell — a group that has been repriced as a single thematic block on AI data-center capex expectations. Two features complicate the read. First, Lumentum is simultaneously a ~15% customer and a direct peer, so cluster strength routed through LITE is not cleanly additive here. Second, the July divergence broke the historical correlation: the group made new highs while FN did not, which means the basket beta that carried the stock up from the $272 low is no longer reliably carrying it. Secondary exposures run to Cisco (~16% of FY sales) on the enterprise/telecom side and to Nvidia on the datacom program side, giving the name a hybrid sensitivity — telecom capex cycles on one leg, AI accelerator shipments on the other. The theme itself reads ACCELERATING; this particular expression of it does not, and that gap is the position the market is currently taking.