Phase Space AI

Financial Model Notes

Celestica [CLS]

Celestica [CLS] — Financial Model Notes

Every figure below is reconstructed from the filed 10-K (2026-02-27) and 10-Qs (2026-07-27, 2026-04-27, 2025-10-27). Alpha Vantage was used for consensus estimates and the split check only. This file is the audit trail for CLS_analysis.json.

---

1. Quarterly reconstruction — how each series was derived

CLS reports three-month and cumulative (six/nine-month, annual) columns. Individual quarters must be differenced. Every quarter below is either directly filed or differenced from two filed cumulative figures.

Revenue ($m)

Q4 2024 = FY2024 9,646.0 − 9M 2024 7,100.3 = 2,545.7.

Gross profit ($m)

Q4 2025 gross profit = FY2025 1,494.1 − 9M 2025 1,061.0. Q4 2025 TRS in COS = FY2025 109.3 − 9M 2025 81.6. Q1 2026 TRS in COS = 1H 2026 40.9 − Q2 2026 48.4 = −7.5 (a loss).

Earnings from operations ($m)

Q1 2025 TRS = 1H 2025 78.3 − Q2 2025 97.4 = −19.1 (a loss), so clean Q1 2025 = 128.8 + 19.1 = 147.9. Note what the clean series shows that the reported series hides: clean operating margin rose monotonically 5.58% → 6.05% → 6.63% → 6.92% → 7.14% → 7.54% across six consecutive quarters, while the reported series zig-zagged 4.86% → 9.42% → 10.29% → 8.60% → 6.72% → 9.75% purely on the share price. The clean series is the one that supports a terminal margin; the reported series is noise.

Q3 2025 = 9M 2025 726.3 − 1H 2025 401.3. Q4 2025 = FY2025 1,040.7 − 9M 2025 726.3. Q1 2026 = 1H 2026 730.4 − Q2 2026 458.3. Q4 2025 total TRS = FY2025 253.0 − 9M 2025 191.6.

The annual series, filed

Longer history of reported operating margin, from AV annual reports (revenue and operating income only, which matched the filings on FY2023 and FY2024): FY2018 1.60%, FY2019 2.54%, FY2020 2.23%, FY2021 2.98%, FY2022 3.99%. Used only to establish that the margin is not low-variance, i.e. that State A does not apply.

---

2. Terminal margin bridge — the opex reconciliation

TTM clean, every line from the filings:

Reported SG&A TTM = Q3'25 38.4 + Q4'25 70.1 + Q1'26 117.4 + Q2'26 58.1 = 284.0. TRS in SG&A TTM = 64.8 + 33.7 + (9.5) + 55.7 = 144.7. Q4'25 reported SG&A = FY2025 259.9 − 9M 189.8. Q1'26 reported SG&A = 1H 175.5 − Q2 58.1. Q1'26 TRS in SG&A = 1H 46.2 − Q2 55.7 = −9.5.

Clean SG&A history for the trend: FY2023 303.2 + 0 = 303.2 → 3.81%; FY2024 293.5 + 51.4 = 344.9 → 3.58%; FY2025 259.9 + 143.7 = 403.6 → 3.26%; TTM 2.75%.

Clean gross margin history: FY2023 754.1/7,961.0 = 9.47% (TRS was below the line); FY2024 (1,033.7 − 39.6)/9,646.0 = 10.30%; FY2025 (1,494.1 − 109.3)/12,390.9 = 11.18%; TTM 11.20%.

Terminal: 11.50% − 2.30% − 1.00% − 0.15% − 0.15% = 7.90%. Constraint 7.90% ≤ 11.50% satisfied with 3.60pp of headroom. Operating profit consumes 68.7% of gross at terminal against 63.4% today.

---

3. Balance sheet and enterprise value, at 2026-06-30

Net debt = 26.4 + 784.0 − 535.7 = +$274.7m. Lease-exclusive: operating lease liabilities are not included (ROU asset $152.7m as a proxy would raise EV ~0.4% to ~$38,591m). Stated explicitly, per the record of a lease-exclusive anchor invalidating a seven-name ladder.

Cash was cross-checked against the cash-flow statement: closing cash $535.7m at 2026-06-30, opening $595.6m at 2026-01-01, net decrease $59.9m. Ties.

EV = 116.2m diluted shares × $328.43 + $274.7m = $38,163.6m + $274.7m = $38,438.3m.

On the 114,982,086 shares on the 10-Q cover instead of diluted, EV would be $38,038.7m, 1.0% lower. Diluted is used; both are recorded.

Debt is not zero and AV's longTermDebt was not treated as missing. CLS has a real revolver and term loans: 1H 2026 saw $600.0m borrowed and $600.0m repaid under revolving loans, $250.0m borrowed and $233.7m repaid under term loans, and $4.8m of debt issuance costs paid. Interest paid $39.3m in 1H 2026.

---

4. Working capital and cash flow

Cash flow, filed

D&A is taken from the cash-flow statement, never from the income statement. AV's income-statement D&A returned $15.5m for Q4 2024 against a $45–53m quarterly run-rate.

EBITDA (clean) = clean EBIT $1,108.0m + D&A $176.2m = $1,284.2m. AV's own ebitda field was not used at any point.

Same-quarter working-capital ratios (× 91 days)

Same-quarter basis throughout. TTM-basis DSO is not used: it produces a false AAOI-style positive on any accelerating name, and CLS's revenue accelerated 62% year on year.

---

5. Segment reconstruction

ATS annual figures are derived as total less CCS (CLS discloses CCS by end market and ATS as one line). TTM CCS = FY2025 9,188.5 − 1H 2025 3,915.7 + 1H 2026 7,051.3. TTM ATS = 3,202.4 − 1,626.3 + 1,694.3. Sum ties to $15,594.5m exactly, which is the check that the derivation is right.

Segment income and margin, filed:

Segment income is defined by CLS as segment revenue less its cost of sales and its allocable portion of SG&A and R&D. It therefore excludes SBC, TRS FVAs, amortisation of intangibles and restructuring — all of which appear as reconciling items to earnings before income taxes. Q2 2026 check: 386.3 − 21.2 + 3.8 + 0.7 − 19.3 + 104.1 − 9.9 − 2.9 = 441.6 = earnings before income taxes. Ties.

Cross-check on the clean EBIT figure from the segment side: 386.3 − SBC 19.3 − amortisation 11.9 − restructuring 2.9 = 352.2 against 354.2 from the income-statement side. The $2.0m difference is computer-software amortisation, which the segment reconciliation excludes from its amortisation line ("excluding computer software", $9.9m) but the income statement includes ($11.9m). Both routes agree to within 0.04% of revenue.

---

6. Own multiple history — construction

Point-in-time series, 1,398 daily observations, 2021-01-04 → 2026-07-29. Daily closes from Alpaca (adjustment=split). Denominator = trailing-four-quarter revenue and gross profit, lagged 45 days to approximate reporting availability, so no observation uses data the market did not have. Share count from the balance-sheet series for the corresponding period. Price/gross-profit and price/revenue are used as proxies for EV/gross-profit and EV/Sales; net debt has been small and the share count fell only 4.4% (121.6m → 116.2m), so the proxy is close and if anything overstates the multiple's rise.

Q3 2025 and Q4 2025 revenue and gross profit were overridden with the filed values before computing the series, because AV's figures for those two quarters are wrong (§7).

P/Sales for reference: current 2.756x = 87.8th percentile of the full window (median 0.383x), 32.1st percentile of the trailing twelve months (median 2.967x).

The full-window figure is not used. A median of 4.19x describes a business that no longer exists — CLS at a 6.5% gross margin selling to enterprise OEMs. The 100th-percentile-to-twelve-months-ago reading is the formal regime-change signature and it is declared.

Trailing 252-day realised volatility: 76.31%, from log returns on the last 253 closes.

---

7. Alpha Vantage figures that disagreed with the filings

AV's four 2025 quarters of operating income sum to $939.6m against a filed FY2025 $1,040.7m — a $101.1m / 9.7% shortfall. AV's 2025-12-31 value of $325.7m is within $0.7m of the true Q3 figure of $325.0m, which is consistent with a one-quarter shift rather than random noise.

Agreed exactly: FY2023 and FY2024 annual revenue and gross profit; Q1 2026 and Q2 2026 revenue, gross profit and operating income; both balance sheets at 2026-06-30 and 2025-12-31; OVERVIEW.RevenueTTM ($15,594,500,000) and OVERVIEW.GrossProfitTTM ($1,863,900,000), both correct to the dollar while AV's own statement detail is not.

SPLITS: queried, one record — 2.0000 effective 1999-12-22, twenty-two years before the price window opens. The split-basis defect class does not apply.

av_vs_edgar.py was not used. A full manual line-by-line reconciliation of FY2025, Q3 2025, Q1 2026 and Q2 2026 was done instead, which is a stronger test than the script's revenue/margin comparison and is what surfaced all of the above.

---

8. Fields deliberately not used, and why

---

9. Open items

1. Transcript mention-frequency — NOT RUN. A required core metric, outstanding. Terms to count, split prepared remarks vs Q&A on the first speaker title containing "Analyst": AI/ML compute, switch, hyperscaler, TRS, A/R sales program, payment terms, price reductions, insourcing.
2. NCS Global revenue contribution — not disclosed by CLS; recorded as not disclosed rather than estimated. Immaterial on any plausible attribution ($46.2m of consideration against a $5,007m revenue increase).
3. Peer EBIT figures in the comparator set are un-normalised. JBL, FLEX, SANM and FN are carried forward from FN_analysis.json on a reported basis. The CLS work shows a reported EBIT in this class can be 23.6% too favourable; the peer EV/EBIT column should be treated as a soft anchor.
4. Customer-level receivable concentration — structurally unavailable. CLS gives only a count of customers above a 10% A/R threshold. The AAOI test cannot be run at customer level on this name.
5. Options-implied volatility term structure — measured at one tenor only (387 days, three strikes). A full surface was not built; the single reading (IV 83.56% vs RV 76.31%) is sufficient to reject the LEAP and is recorded as MEASURED rather than extrapolated.