Phase Space AI

Valuation

Celestica [CLS]

Celestica [CLS] — Valuation

Spot $328.43 (2026-07-29). Two outputs: a 12-month target and the 5-year implied-path test.

Every income-statement and balance-sheet figure in this document comes from the filed 10-K (2026-02-27, FY2025) or 10-Q (2026-07-27, quarter ended 2026-06-30). Alpha Vantage was used for consensus estimates and for the split check only. Four AV figures were found wrong against the filings and are listed in §9.

---

1. THE INSTRUMENT QUESTION — answered first, because the task turns on it

The brief asked whether CLS is the Fabrinet case: an ODM at structurally low margins where an EV/EBIT exit multiple is inadmissible. On FN a sibling agent found a 14.4% terminal EBIT margin against an 11.96% gross margin — arithmetically impossible — with FN's own operating margin already consuming 82% of gross profit, leaving 2.2pp of total opex headroom, so no defensible point existed anywhere in the identifiable band. FN was valued instead on EV/gross profit against its own history.

CLS is NOT that case, and the reason is arithmetic rather than judgement

CLS's clean operating margin is 7.10% — between two and seven times the 1–3% band at which an EBIT multiple degenerates. The hard constraint m_EBIT,T ≤ m_gross,T holds with 3.60pp to spare, and the full opex bridge — which valuation.md correctly calls the real test — reconciles with 31.3pp of gross profit still available to operating expense.

Decision, stated explicitly as required:

• Primary instrument: EV/EBIT, on a TRS-cleaned EBIT. Admissible, and used.
• Co-primary: EV/gross profit. Not a fallback. It is the multiple in which the EMS class is genuinely comparable, it strips the pass-through material, and it is the only one of the three computable across CLS's own five-and-a-half-year price history — so it is the one that can carry a percentile.
• EV/Sales: INADMISSIBLE. Cost of sales is 88.0% of revenue ($4,121.1m of $4,698.6m in Q2 2026). A contract manufacturer's revenue line is mostly customer-specified material bought and passed through, so EV/Sales measures material content as much as business value. It is reported for continuity with the shared EMS set and used nowhere in the conclusion.

Both instruments were run to completion and both give the same sign. That is the point of running two.

The reason the reported EBIT may not be used — and this is the headline finding

CLS's reported earnings from operations includes mark-to-market gains on a total return swap over its own shares. §2 below. On the reported figure CLS trades at 28.06x EV/EBIT. On the clean figure it trades at 34.69x — the reported multiple understates how expensive the name is by 23.6%. Anyone valuing CLS on a screen's EBIT is valuing a swap on Celestica stock as if it were Celestica's business.

---

2. THE TOTAL RETURN SWAP — 19.1% of reported operating profit is a bet on own equity

What it is

CLS is party to a Total Return Swap Agreement with a third-party bank over 1.25 million of its own Common Shares, entered into to fund the cash settlement of SBC awards (it cash-settled $326.3m of SBC in 1H 2026). It does not qualify for hedge accounting. Mark-to-market changes are recorded in cost of sales and in SG&A — so they flow through reported gross profit and reported earnings from operations.

The magnitudes, filed

FY2023's gain sat below the operating line, so FY2023's operating margin is clean. From FY2024 the classification moved into cost of sales and SG&A.

What it does to the reported margins

$261.8m of the TTM $1,369.8m of reported operating profit — 19.1% — is the swap. In FY2025 it was 24.3%. CLS itself puts a number on the per-share effect: the 10-Q states that GAAP EPS for Q2 2026 "included $0.90 per share (pre-tax) positive impact attributable to TRS FVAs" — 28.4% of the $3.17 GAAP diluted EPS.

Why it cannot be underwritten — and why it reverses next quarter

The strike history is disclosed and the arithmetic closes exactly:

• Re-struck March 2025 at $91.58; CLS received $98.6m. 1.25m × ($91.58 − $12.70) = $98.6m → the prior strike was $12.70, a legacy position set years below market.
• Re-struck December 2025 at $288.87; CLS received $246.6m. 1.25m × ($288.87 − $91.58) = $246.6m.
• At the 2026-06-30 close of $364.80, embedded value = 1.25m × $75.93 = $94.9m, which ties to the $94.3m TRS asset on the balance sheet.

So the $253.0m FY2025 gain was the harvest of a strike set at $12.70 against a share price that reached $288.87. That cannot repeat. And on current marks it is already reversing: at the 2026-07-29 spot of $328.43 the embedded value is 1.25m × $39.56 = $49.5m, against $94.3m ten weeks earlier. Q3 2026 books a TRS loss of roughly $44.8m on current spot — a 0.9pp headwind to the reported operating margin, from the same line that gave a 2.2pp tailwind in Q2.

A further classification point: the $98.6m and $246.6m of settlement proceeds were recorded in cash provided by financing activities, while the income sits in operating profit. Reported operating margin is credited with a gain whose cash never touches operating cash flow.

CLS agrees with this adjustment. Its own non-GAAP adjusted operating margin excludes TRS FVAs on the stated grounds that they "do not represent our ongoing operating performance" — Q2 2026 adjusted operating margin 8.2% against a GAAP earnings-from-operations margin of 9.8%. The adjustment made here is the issuer's own, applied consistently to the historical series and to the terminal margin.

---

3. COMPANY STATE — C, with a declared B overlay

Not A

The reported operating margin ran 1.60% (FY2018) → 2.54% → 2.23% → 2.98% → 3.99% → 4.25% (FY2023) → 6.21% (FY2024) → 8.40% (FY2025) → 8.78% (TTM). That is a 7.2pp structural move, not the low-variance band State A requires — and it coincides with a business-mix regime change: CCS went from 58% of revenue in FY2023 to 79% of TTM revenue, and the largest customer from 22% to 34%.

Not D

The reverse DCF returns a solution at every exit multiple tested from 9.71x to 34.69x. The instrument is not degenerate.

C — thin-margin but economically observable

valuation.md's State C test is "pre-profit or thin-margin, but with usable evidence: positive gross margin, identifiable contribution margin, visible expense scaling, cohort data, or a formally guided long-term model." CLS satisfies it on three of the five:

The B overlay, declared rather than blended away

CLS's revenue driver is exogenous: hyperscaler AI capital expenditure, concentrated in three unnamed customers at 63% of Q2 2026 revenue. That is somebody else's capex cycle. But CLS's own margin is mix- and scale-driven rather than cycle-driven — it expanded straight through the 2023–24 enterprise-hardware downturn during which Enterprise revenue actually fell — so the primary state is C and the overlay is recorded on the revenue line.

Consequence, per State C: the terminal margin is built from the opex bridge and never assumed, and the instrument is the two-dimensional expectations surface, so all three of required margin at underwritten growth, required growth at underwritten margin, and the sensitivity surface are reported below.

Evidence grade: A−. Filings current and reconciled line by line; consensus available with real analyst counts. Downgraded from A because CLS names no customer and quantifies receivable concentration only against a 10% threshold, and because transcript mention-frequency was not completed inside the time box.

---

4. TERMINAL MARGIN — 7.90%, built from the bridge

The constraint first

m_EBIT,T  ≤  m_gross,T        →      7.90%  ≤  11.50%   ✓     headroom 3.60pp

The bridge, every line from the 10-K/10-Q

None of these lines came from AV's sellingGeneralAndAdministrative, which the record shows is G&A only and was 4.7x wrong on two unrelated names.

Where 7.90% sits — both directions stated

• 0.80pp ABOVE the clean trailing actual of 7.10%. Justified: the demonstrated direction is up 2.85pp in two and a half years on visible SG&A leverage that has been delivered three years running.
• 0.88pp BELOW the reported trailing actual of 8.78%. Justified: the reported figure contains the TRS gain, which is 1.68pp of it, so 7.90% is above the reported figure once the swap is removed.
• Adding back TTM SBC of $78.4m (0.50% of revenue) puts the terminal at 8.40% on CLS's own non-GAAP definition, 0.20pp above the Q2 2026 actual of 8.2%. Modestly better than today on the issuer's own measure.
• Industry data used as a sanity band only, never as an override. The EMS peer set runs 4.3–5.1% operating margin (JBL 4.32%, FLEX 5.13%, SANM 5.09%) and FN 9.78%. CLS at 7.90% sits between them — consistent with a gross margin (11.2%) that also sits between theirs (9.0–9.3%) and FN's (12.0%). No percentile clamp was applied.

On the parameter that has failed six times

No stored scan_v2/scan_v3 record was located for CLS. Had the universe-wide 14.4% been applied it would have breached the 11.50% terminal gross margin by 2.90pp — arithmetically impossible, exactly as on FN, and CLS would have been the eighth name in the corpus to violate m_EBIT,T ≤ m_gross,T. Recorded because the same constant was applied to 70 of 111 names while labelled sector-derived.

---

5. EXIT MULTIPLE — 12.0x derived, and the flip point is what matters

Derived from the identity

EV_T / EBIT_T  =  (1 − t)(1 − g/ROIC) / (WACC − g)

t = 20% is CLS's adjusted effective tax rate in Q2 2026 (GAAP 16%; FY2025 GAAP 15.3%; management guides ~21%). Current clean ROIC is 32.2% — clean NOPAT $886.4m over invested capital $2,754.3m (equity $2,479.6m + net debt $274.7m) — but it is stepped down to 25% at terminal because 1H 2026 capex ran 7.1x the prior year and net PP&E rose 75% in six months, so invested capital is about to grow much faster than NOPAT. That step-down is adverse to the name.

Reference class — declared under-populated

EMS operating model, matched on build-to-customer-specification, pass-through material content, gross margin under 13%:

n = 4 against a requirement of 5 — declared marginally under-populated rather than papered over. It has genuine dispersion in both matched dimensions (growth 8.1%–58.6%, operating margin 4.3%–9.8%), so it is not the §9.9 failure of a precisely identified uninformative anchor. CLS's +47.3% growth is bracketed by SANM above and FN/JBL below.

A caveat carried forward and not buried: those four peer EV/EBIT figures are on a reported basis. The CLS work establishes that a reported EBIT in this class can be 23.6% too favourable. If any peer carries an analogous non-operating credit, the peer column is likewise understated. Treat it as a soft anchor.

The base 12.0x sits inside the anchor range (11.4x–34.0x), so valuation.md's prohibition on a base below every stated anchor is not triggered — but only just, and it is 13.2 turns below the median. Both are carried: 12.0x as the base, 25.2x as the upper sensitivity bound.

Percentile against CLS's own history — and the regime-change trap

Reported on EV/gross profit, because a five-year EV/EBIT percentile is meaningless for a business whose operating margin was 3.2% in Q1 2023. Point-in-time series: 1,398 daily observations from 2021-01-04, the denominator lagged 45 days to reporting availability, price/gross-profit as the proxy (share count fell 121.6m → 116.2m, −4.4%, so the proxy is close and if anything overstates the multiple's rise).

The full-window percentile is declared uninformative. A 5.5x re-rating has occurred, and measured to twelve months ago the current multiple sits at the 100th percentile — every observation in the top decile was created inside the last year. That is precisely the signature valuation.md flags on the legacy semicap names, where all four sat at the 99th–100th percentile of their own 9.5-year history measured to twelve months prior.

Within the regime-consistent windows the picture inverts: CLS sits at the bottom quartile — the 25.4th percentile — of its own post-regime multiple range, at $328.43 against a 2026-06-30 close of $364.80. Both readings are stated; the regime-consistent one is used, and the substitution is declared.

---

6. IMPLIED-PATH TEST — the Valuation Criteria

Terminal value is 88.6% of EV, so the reverse DCF is the mandatory primary long-horizon output.

Held fixed and named: terminal EBIT margin 7.90%; WACC 10.0%; horizon 5 years; TTM revenue $15,594.5m; diluted shares 116.2m; net debt +$274.7m (total debt $810.4m including finance leases, less cash $535.7m — lease-exclusive, and adding the operating lease liability would raise EV ~0.4%); EV $38,438.3m; fcf_margin = 0.033, CLS's own demonstrated TTM FCF margin.

The required parameter

A tooling defect corrected in the open: reverse_dcf.py accepts fcf_margin in solve() and project_ev() but does not expose --fcf-margin on the command line, so a CLI-only caller silently gets the terminal-only behaviour the record identifies as overstating the required CAGR. On CLS that is 33.16% terminal-only against 29.99% with the demonstrated FCF margin passed — a +3.17pp overstatement, adverse to the name. The module was called directly.

The margin the strategy ranks on

Demonstrated − required = 47.3% − 29.99% = +17.31pp.

All three windows reported, as required:

TTM YoY is primary because the backward 3-year window spans the mix regime change — over FY2022–FY2023 ATS was 42% of revenue and no customer exceeded 22%. The backward figure is disclosed, not discarded, and it is the only window that fails.

Why this result is robust rather than marginal — three independent checks

1. Flip point: exit multiple 5.98x. The multiple would have to fall to 5.98x for the price to require more growth than CLS has demonstrated. Every EMS comparator trades far above it — the lowest, SANM, is at 11.4x — and the identity floor at g=3%/ROIC=20% is 9.71x. There is no defensible parameterisation of the identity that produces a fail. This is the mirror image of FN, whose flip point (14.8x) sat above its identity base (14.0x).
2. Required terminal margin at demonstrated growth: 3.94%. At 47.3% growth and a 12.0x exit, the price requires a terminal operating margin of 3.94% — roughly half the clean trailing 7.10%, and below anything CLS has earned since FY2022. The price does not require margin improvement at all; it requires the growth not to stop.
3. The co-primary instrument agrees, and by more. Required gross-profit CAGR 30.66% (the required revenue CAGR compounded with clean gross margin moving 11.20% → 11.50%) against +53.11% demonstrated — clean TTM gross profit $1,746.8m versus prior-TTM clean $1,140.9m. Margin +22.45pp, wider than the revenue instrument's +17.31pp, because clean gross margin expanded 0.42pp over the year.

Sensitivity — on the exit multiple, never on scenario probabilities

The table above is the sensitivity. A 1.0x change in the exit multiple moves the required CAGR by ~2.1pp near the base. A ±1pp change in the terminal margin moves it by ~3.5pp, but the terminal margin is boxed between the 11.50% gross ceiling above and 7.10% of clean trailing actual below, so its live range is about ±0.8pp. The exit multiple again carries most of the variance — but unlike on FN it cannot carry the answer, because the flip point sits below every anchor.

Verdict: PASS

Not "pass with argument." At +17.31pp on the primary window, +5.65pp on the forward 3-year, a flip point below every anchor, and a required terminal margin at half the trailing actual, the valuation criterion passes on the arithmetic without needing a narrative to carry it.

---

7. TWELVE-MONTH TARGET

Target $495. +50.7% to spot $328.43.

Build

A consistency note in the target's favour: the historical multiple's denominator included TRS gains from FY2024 onward, so the historical multiple is understated. Applying it to a clean forward denominator is therefore conservative.

Sanity band and caveat, stated plainly

The target implies 32.8x FY2027E adjusted EPS of $15.08 (20 analysts, range $12.20–$19.90) against 21.8x at spot. That is the aggressive reading and is disclosed as such: the entire target is a gross-profit-growth statement at a flat multiple. At the trailing-twelve-month median multiple of 24.99x it would be $562; at the full-history median it collapses to a fraction of spot. No external professional price target was retrieved — recorded as absent rather than substituted.

Named events inside twelve months

Dated in CLS_Catalyst_Calendar.md. Q3 2026 results with the updated 2026 annual outlook (late October 2026); Q4/FY2026 results, the first 2027 guide, and the FY2026 10-K carrying the refreshed top-10-customer percentage and A/R concentration count (late January / February 2027). Within Q3 2026, the TRS mark reverses to a ~$44.8m loss on current spot — a 0.9pp reported-margin headwind for reasons unrelated to the business. No dated product event is disclosed by the company and none is asserted here.

---

8. Liquidity Criteria and vehicle — PASS, equity

Vehicle: EQUITY. Ladder rung 1.

Absent a measured IV − RV spread and a quoted size, no derivative vehicle is admissible and the vehicle is equity by default. IV − RV is recorded as UNIDENTIFIED rather than as a zero or a peer-implied estimate — the framework record contains a case of a VOL252 field silently resolving to 0 and zeroing a risk ratio.

---

9. Data-layer defects found — full list with magnitudes in CLS_analysis.json

av_vs_edgar was run by hand rather than via the script — a full line-by-line reconciliation of FY2025, Q3 2025, Q1 2026 and Q2 2026 against the filings, which is a stronger test than the script's revenue/margin comparison and is what surfaced all four AV statement errors. DISAGREE on FY2025 revenue and gross profit and on Q3/Q4 2025 revenue, gross profit and operating income; AGREE on Q1 2026, Q2 2026, FY2023 and FY2024 and on both balance sheets.

---

10. Summary card