Phase Space AI

Financial Model Notes

Fabrinet [FN]

Fabrinet [FN] — Financial Model Notes

No .xlsx was produced this run (time-boxed; the brief's priority order puts accounting-quality and valuation work above model artefacts). These notes carry every figure the model would contain with its source, so the model is reproducible and every number traceable.

Per known-silent-failures.md §1–§3, had a workbook been built: no defined name matching ^[A-Za-z]{1,3}\d{1,7}$ (rev_fy27, not rev27; vol_252d, not VOL252 — the LITE memo lost a volatility-derived risk ratio to exactly that); open by absolute path and assert the filename matches the ticker before reading a cell; and treat the model as unverified until Excel has opened, computed and saved it.

Source hierarchy actually used

Verified inputs

Filed annual statements (10-K, $000)

FN has no R&D line. Customers own the designs. This is a structural feature and it is why a 12% gross margin still supports a 9.5% operating margin.

Quarterly, recent ($m, AV normalized, EDGAR-verified)

TTM: revenue $4,235.0m, gross margin 11.96%, operating income $414.0m (9.78%), D&A $60.6m, EBITDA $474.6m (11.21%). Prior-year TTM revenue $3,262.5m → +29.8%.

Use same-quarter DSO, not TTM-basis DSO. On this name the TTM measure reads 78.3d against 73.6d a year ago and AR growing 8.2pp faster than revenue — which mimics the AAOI signature and is an artifact of a 39% revenue acceleration against a trailing denominator. The same-quarter series (68.3d vs 68.9d vs 72.8d) is flat to improving. Applying the same correction to AAOI does not clear it (180.5d vs 156.4d vs 129.7d), which is what establishes the corrected measure as discriminating rather than exculpatory.

Balance sheet, 2026-03-27 ($m)

Lease convention stated per §9.7: operating leases excluded from net cash/debt across all three names in this cluster so the comparison is internally consistent. FN's are immaterial.

Forward, from guidance (8-K and call, 2026-05-04)

Model structure that would be built

1. Revenue by market, not by data rate. FN does not disclose data rates, so a 800G/1.6T mix model would be fabrication. Build: Optical Communications split telecom / datacom / DCI (all three disclosed with numbers in Q3FY26), plus automotive, industrial lasers and other.
2. Gross margin held flat at 12.0%, declining in the bear case. Do not model expansion: the three-year trend is down 0.75pp and management guides only to "similar dynamics." This is the single most important modelling discipline on the name.
3. SG&A as a declining percentage, 2.56% → 2.1% floor. There is no R&D line to model.
4. Tax at 6.4%, the FY2025 effective rate, with a note that Thai incentive expiry is the tail risk this model does not capture. A move to a 15% global minimum rate would cut NOPAT ~9%.
5. Zero debt, no interest expense. Interest income on $945m of cash and investments is material (~$40m/yr, 12% of pretax income) and must be modelled below the line, not netted into operating.
6. Shares 36.217m, flat. The buyback is authorised but barely used; do not model share shrinkage.
7. Terminal margin capped at the gross margin. Enforce m_EBIT,T ≤ m_gross,T as a hard model assertion that raises, not a flag. This is the check the screen lacked, which is how a 14.4% terminal margin was applied to a 12.09% gross-margin business.

Defects to carry forward