FN · Fabrinet · Stock research

Current Thesis

The binary this coverage was tracking resolved, and it resolved against the price. Fabrinet reported fiscal Q4 after the close on 2026-08-17: revenue $1.316B, +45% YoY, above the top of its own $1.25B–$1.29B guide and above the $1.275B consensus; non-GAAP EPS $4.10 against $3.82 expected. The fiscal-Q1 FY2027 guide came in above the record just printed — $1.375B–$1.425B revenue, non-GAAP EPS $4.10–$4.25 on roughly 36.3M diluted shares. Shares fell 18.5% on 2026-08-18 (Benzinga), after opening the session down about 9% pre-market.

That is the whole read now. The operating narrative — contract-manufactured optics under AI interconnect — accelerated: data center revenue $669M, 51% of the quarter, +68% YoY and +13% sequentially, with DCI products crossing a $1B annualized run rate. The market narrative did not follow. The prior note's stated break condition, a weekly close below $450, was met: the 2026-08-14 close of $570.22 has been replaced by $436.67 on 2026-08-21, leaving the stock 41.5% under its $746.47 52-week high and down 38.0% over three months, with RSI(14) at 47.6.

The narrative is saturated. Coverage is mainstream (a corrected 2026-08-18 wire piece exists purely to name the customer list; Roundhill launched a photonics ETF on 2026-08-06), estimates are rising, and the response to a beat-and-raise was a one-day 18.5% decline. The theme did not fail — Fabrinet's own numbers say the opposite — but the marginal bid for this specific name did not appear on the best news it has had. Multiple compression, not estimate compression, is doing the work.

Bullish and bearish views on Fabrinet

The model's bull view on Fabrinet (FN), in brief: The print cleared on every reported line (2026-08-17). The bear view: A beat-and-raise produced an 18.5% single-day decline (2026-08-18). Both cases follow in full.

Bull Case

• The print cleared on every reported line (2026-08-17). Revenue $1.316B versus a $1.25B–$1.29B guide; non-GAAP EPS $4.10 versus $3.82 consensus; FY2026 revenue $4.64B, +35.7% over FY2025's $3.42B.
• The forward guide steps up from a record. Fiscal Q1 FY2027: $1.375B–$1.425B revenue, GAAP EPS $3.39–$3.54, non-GAAP EPS $4.10–$4.25 — the mid-point of the non-GAAP range sits above the Q4 result.
• Data center is now the majority of the business. $669M, 51% of revenue, +68% YoY, +13% QoQ, spanning transceivers, DCI, high-performance computing assemblies, silicon photonics and co-packaged optics (Q4 FY26 call, 2026-08-17).
• Communications infrastructure held its level: $413M, 31% of revenue, +40% YoY on the 400ZR/800ZR coherent cycle, though only +1% sequentially.
• Gross margin ticked up sequentially. Non-GAAP gross margin 12.2% in Q4, +10bp versus Q3 FY26 (2026-08-17 call).
• Sell-side raised into the fall. Barclays (Tim Long) kept Overweight and lifted the target to $739 from $702; JPMorgan (Joseph Cardoso) stayed Neutral and lifted to $695 from $680; Needham reiterated Buy at $650; Rosenblatt and BNP Paribas reaffirmed Buy (all 2026-08-18).
• The customer roster broadened. Cisco, Nvidia, Amazon and Nokia were identified as the largest customers in the Q4 disclosure cycle (Benzinga, corrected item, 2026-08-18).

Bear Case

• A beat-and-raise produced an 18.5% single-day decline (2026-08-18). When better numbers and a higher guide are met with that, the disagreement is about the multiple, and no upcoming company event forces a re-vote until early November.
• Datacom went flat again. Management described the datacom business as "somewhat flat" in Q4 on component supply, program transitions and demand timing (2026-08-17 call) — following the roughly 6% sequential decline disclosed on 2026-05-04. Two consecutive quarters without sequential growth in that line is now the record, whatever the framing.
• Margin is not scaling with revenue. 12.2% non-GAAP gross margin is 30bp below the year-ago quarter even as revenue grew 45%. A 2026-08-18 market wrap attributed part of that day's broad chip selloff to "Fabrinet's margin miss" — that framing sits awkwardly against a 10bp sequential improvement, and the public summaries do not reconcile the two; the GAAP Q1 EPS guide of $3.39–$3.54 was separately flagged as below expectations.
• The target-versus-price gap is extreme and has a history of closing downward. Needham cut from $800 to $650 on 2026-07-22 before the print. A panel average near $766.91 against a quote of $446.14 (chartmill FN analyst-ratings page, accessed 2026-08-23) is a spread that historically resolves through target cuts as often as through price.
• Macro is pressing the same window. Sessions on 2026-08-18 and 2026-08-19 saw the Nasdaq fall on elevated bond yields with Brent near $91 — the discount-rate input that hits a high-multiple hardware name hardest.
• Concentration cuts both ways now that Nvidia is on the customer list. The 2026-08-18 wire framing was explicit: datacom weakness tied to Nvidia programs overshadowed telecom strength.

Setup & Price Structure

• Reference close 2026-08-21: $436.67. Distance from the $746.47 52-week high: −41.5%. Three-month price change: −38.0%. RSI(14): 47.6.
• The 2026-08-18 session gapped down roughly 9% pre-market and closed −18.5%. That leaves unfilled supply overhead running back toward the pre-print $570 area — the level a repair attempt has to work through.
• RSI at 47.6 is neutral. There is no oversold washout comparable to the sub-30 reading of 2026-07-08 that preceded the July bounce, so the momentum series is not signalling capitulation; equally, no base has formed in the sessions since the gap.
• Positioning observables, not verdicts: Citadel Advisors and Millennium sharply increased Fabrinet stakes in Q2 per 13F filings surfaced 2026-08-18 — those holdings are struck as of 2026-06-30 and predate both the print and the gap. The published target panel sits far above the tape. There is no earnings date inside 30 days to force repricing. The recent-filings feed carries no new insider transactions.
• The prior published break level did its job: the $450 weekly-close condition set on 2026-08-15 fired on the week ending 2026-08-21. The structure that framed the July bounce is gone.

Catalyst Calendar (next 30 days)

• ~2026-08-26 (est.) — Nvidia fiscal Q2 FY2027 results. Date unconfirmed here. Nvidia was named among Fabrinet's largest customers on 2026-08-18, so networking and optics commentary is the nearest read-through on whether the datacom flatness is program timing or digestion.
• ~2026-09-04 (est.) — Ciena fiscal Q3 results (last year's comparable print landed 2025-09-04). A demand check on the coherent/DCI cycle that produced $413M of communications-infrastructure revenue.
• ~2026-11-02 (est.) — fiscal Q1 FY2027 results, outside the 30-day window and the next date at which the $1.375B–$1.425B guide and "data center up sequentially" are tested.

Elapsed catalysts

• No Fabrinet-specific company event is confirmed inside the window. The Rosenblatt Age of AI Technology Summit appearance was 2026-08-18 and has passed; the FY2026 Form 10-K is filed. (passed 8d ago)

What Would Change Our Mind

The structure that held the July low is already lost — the $450 weekly-close condition published on 2026-08-15 was met at $436.67 on 2026-08-21, and this note treats that leg as broken rather than pending. What matters from here is whether the de-rate finds a floor while estimates are still rising. A weekly close below $400 confirms the post-print de-rate is extending rather than basing, and would argue the label should move from saturated to a failed narrative outright. Secondarily: the data center category failing to grow sequentially at the fiscal-Q1 print (~2026-11-02, est.), after management guided it up on the 2026-08-17 call, would break the supply-timing explanation for good and make the datacom flatness a demand fact.

The reverse case is equally observable. A weekly close back above the pre-print $570 area would mean the gap has been absorbed and the beat is being re-priced late rather than rejected. So would two or more targets holding above $700 through a datacom-negative Nvidia print, since the target panel converging downward — the pattern of the 2026-07-22 cut from $800 to $650 — is the mechanism by which the current $766.91 average would validate the tape instead of contradicting it.

Correlation Notes

• Optical cluster: AAOI, LITE and COHR remain the peer set. The relative-strength failure flagged in July persists — Fabrinet is now 41.5% below its own 52-week high while the theme continues to generate fresh product news. Cluster strength has not transferred mechanically to this name in either direction.
• Nvidia beta: the 2026-08-18 reaction tied Fabrinet's datacom line directly to Nvidia program timing. The name should be expected to trade off Nvidia networking commentary until the November print.
• Rates: the 2026-08-18/19 tape moved on bond yields and $91 Brent. A contract manufacturer carrying 12.2% gross margin has no revenue-side buffer against multiple compression driven by the discount rate.
• Theme wrappers: the Roundhill photonics ETF (LYTE) launched 2026-08-06 gives theme flow a diversified route that does not require buying the single lagging name.
• Geography: manufacturing is concentrated in Thailand. Reported US restrictions on Chinese data-center devices (Reuters, 2026-08-04) can move the name in either direction depending on final product scope.