CLS

We rate Celestica a Buy at $281 a share against a $246.65 close, 12% below our value. Celestica beat its guide for the ninth quarter running and gave a 2027 revenue outlook of $40.0B, 14% above the $35.0B we had forecast; that alone adds about $47 a share. Two outside reviewers then found that the first draft counted $700.8M of receivables sold to a bank twice, and that it lifted the margin in every year when only 2027 is sourced. We fixed both, and on their check of the margin through a cycle we glide the 2031 margin of Celestica's main segment down to 7.7%. We still expect AI hardware spending to slow after 2028. The stock has fallen 42% since April while the market rose. Our estimate of its earnings rose over the same months, so we read the fall as a lower multiple and sector pressure, not a weaker business (section 6). The previous report rated it a Hold at $286.

Terms used below.

• CCS and ATS. Celestica's two segments. CCS (Connectivity and Cloud Solutions) builds network switches, servers, storage and AI racks for hyperscalers and AI labs: 84% of Q3 revenue. ATS (Advanced Technology Solutions) builds for aerospace, industrial, health and chip-equipment customers. HPS (Hardware Platform Solutions) is the part of CCS where Celestica owns the design.
• Bear, central and bull cases. The central case is our best estimate and sets fair value; the bear is a severe but plausible bad outcome, the bull the matching good one. Each is one consistent business outcome.
• Long fade and outside view. Two estimates of what the company is worth at the end of 2031. The long fade extends the forecast 15 more years while growth slows to 3% and the return on new capital falls to its mature level. The outside view applies an EV/EBIT multiple (enterprise value over operating profit) from the past ten year-ends of Celestica and seven comparable manufacturers. The exit price is their average.
• Mature return. The return on new capital the long fade ends at.
• Adjusted operating margin. Celestica's non-GAAP measure: segment income over revenue, before stock pay, acquired-intangible amortization and restructuring.
• Buy. The price is at or below our central fair value.
• Unblinded. Written after the price was seen (above).

1. Q3 2026 beat our forecast and the guide

Graded in ../2026-10-15-cls/grades.json from the release (8-K accepted 2026-10-26 20:23 UTC). Our revenue miss is the old report's choice to assume half, not all, of the usual beat: Q3 beat the guide midpoint by 8.0%, and the median of the last four quarters is now 6.9% (nine of nine quarters above the midpoint, output.txt).

What the quarter showed (10-Q Q3 2026 and release; reported facts unless marked):

• Compute grew faster than networking. Networking +68% to $3,263M; Compute & Storage +242% to $1,609M (AI compute volumes); HPS about $2.3B, +70%, but its share of revenue fell from 44% a year earlier to 40%. CCS margin 8.9% (Q2 8.7%), from operating leverage.
• Some working capital was financed by selling receivables. Inventory rose $1.1B in the quarter and payables $1.5B. Celestica also sold $1.3B of receivables to a bank in Q3; $700.8M was outstanding at quarter-end (nil at June 30), and the program limit was raised from $450M to $1.0B in September. Counting the sold receivables, working capital was $1,450M, 6.8% of 2026 revenue, not the $750M on the balance sheet. We count them in working capital and deduct what is owed to the bank (section 3).
• Cash. $3,961M at quarter-end after the $3.39B August offering. The 2026 free cash flow outlook was cut from $600M to $460M for "targeted working capital investments", and the 10-Q expects "a net use of cash in 2027" (management claims).
• Customers. 35%, 17% and 14% of Q3 revenue, all CCS (Q2: 32%, 17%, 14%); the top ten were 85%.
• Guidance (management). Q4 revenue $6.275-6.575B at 8.7% margin; 2026 $21.0B; 2027 $40.0B revenue (+90%), 9.0% adjusted operating margin, $21.50 adjusted EPS, $1.5-2.0B capex; "2028 customer demand forecasts that are meaningfully higher than our 2027 revenue outlook."
• Not yet available. The call and Investor Day are on 2026-10-27 at 8:00 ET; there is no transcript or slide deck yet, so this revision rests on the release and the 10-Q. The Investor Day may add multi-year targets (section 9).

2. What moved the value from $264 to $281

The last restatement of the 2026-10-15 report valued it at $263.97 under today's method (the report was published at $286 under the method of its day). This model rebuilds those inputs to the cent, then changes one group at a time; each step is a full rerun (output.txt, "Bridge").

Of the 2027 step, $37.62 is management's outlook itself and $9.15 is our assumption that Celestica beats it by half its usual margin ($40.0B exactly gives $272.05).

What the reviewers changed. The model pushed before the market pass gave $309.65. Two outside reviewers read it with the market in hand; each change below rests on a filing, and each is a full rerun (output.txt, "Changes after review"). All three were made after the price was seen and all three moved the value toward it, so we name them here.

The mix line is fitted inside an upcycle, so it carries some operating leverage; it is a floor on the argument that design content alone holds the margin, not proof. The 2023 level, 6.2%, came before the AI build. Gliding to it gives $230; the 2025 level, 8.2%, gives $297 (output.txt).

Method changes since 2026-10-15.

• Rule H (2026-10-26). Each case's mature return is now its own 2031 return on the capital it holds, floored at the cost of capital and capped at Celestica's own median five-year history. Celestica's history is low: its windows ending 2019-2022 earned 0-6% on the capital it put in, when it was a slow-growing contract manufacturer, so the median is 6.2%. That is below the 10.14% cost of capital, so the floor sets every case's mature return at 10.14% (the old report judged 10% / 15% / 20%). The fade still starts at the forecast's own return on new capital, 74% in the central, and declines to 10.14%. Using only the two windows reported under US GAAP (2024-25, median 117%) would let each case keep its own held return and add $16. We keep the default because the slow-growth years resemble what our central becomes after 2030.
• No plant-per-revenue floor (ruling 2026-10-26). Our capex rule never had one: from 2028 capex is 1% of revenue plus $0.10 per dollar of the next year's growth, so plant per revenue dollar falls as revenue grows.
• No own-history exit from a different growth era (ruling 2026-10-26). The outside view's default, anchor (d), averages Celestica's own ten-year median EV/EBIT (10.0x) and its peers' (13.7x): 11.9x. Celestica's own leg was earned at a median 7.6% revenue growth over 2016-2025; our central grows 4.9% in 2031 and 5% in 2032. The leg comes from a comparable growth era, so it stays. If anything the two AI-era years (2024 18.8x, 2025 32.1x) are the outliers in it; without them the leg is 9.0x and the anchor 11.4x.
• The judgment range is shown (section 4) for both the exit multiple and the mature return.

3. The forecast keeps the post-2028 slowdown

model.py builds each segment's revenue from growth, applies a segment margin, charges stock pay (0.4% of revenue) and restructuring (0.1%), runs capex and depreciation on a plant schedule, sets working capital as a share of revenue and keeps all cash to 2031. Our assumptions are labelled; reported figures are from the Q3 2026 release and 10-Q.

• 2026-27 (data). Q4 at the guide midpoint plus half the recent beat ($6,648M); 2027 at $41.4B, the outlook plus the same half-beat. CCS margin 9.25% in 2027, the level the 9.0% company outlook implies with ATS at 6.8%. 2027 capex $2.0B, the top of the guide, because our revenue is above the outlook.
• 2028-31 growth (judgment, unchanged). CCS grows 22%, 10%, 3%, 5%; ATS 5%, 5%, 4%, 4%. Management's 2028 claim supports the 22%. Nothing dated says when the AI build digests; in the past Celestica's revenue fell 11% in 2019 and stayed flat to 2021. We do not move the slowdown later on a management claim.
• 2028-31 margin (judgment, changed after review). CCS 9.0% in 2028 (the old path), then a glide to 7.72% in 2031 (section 2). ATS 6.8% in 2027-28, then 7.0%.
• Working capital (data and judgment). 7% of revenue, counting the sold receivables (6.8% at 2026-09-30; 9.3% at the end of 2025), against the old 5%. Bear 9%, bull 6%. The $700.8M the bank paid for them is owed back through the collections it receives, so the bridge deducts it at exit, held at the 2026-09-30 balance. The bank can end the program on three months' notice.
• Tax 21%; 127.4M shares held flat. Stock pay is an expense, so it is not counted again as dilution.
• Where we differ from management's 2027 EPS. Our path gives $23.73 of adjusted EPS in 2027 against the $21.50 outlook. At our share count management's figure implies about $133M of net finance cost; our model earns $97M of net interest instead, on $3.7-3.9B of cash. Either management expects much less cash (more working capital), more shares, or its outlook is cautious, as its record suggests. Interest on that gap is worth about $1 a share; we disclose it and do not fit to it.

Cases. Bear: 2027 $34.0B (a program slip or shortage, 15% below the outlook), CCS +3% in 2028 then −20% in 2029, margin down to 6.5%; 2031 revenue $30.4B. Bull: 2027 $46.0B, CCS +38%, +20%, +10%, +8%, margin 9.6% then 9.5%; 2031 $87.7B.

4. The value is $281, within a $245-335 judgment range

Cost of equity 10.31% = 5.29% risk-free + 1.195 x 4.20% (Damodaran's 2026-10-01 pair; beta is the median of seven peers, Blume-adjusted). The bridge adds year-end 2031 cash less operating cash, term loans, finance leases and pension ($12,953M) and deducts the sold receivables ($700.8M): $12,252M in the central, $96 a share at exit. The central exit is 11.2x 2031 EBIT. The fade and the outside view are 10% apart.

Judgment range (central operating case fixed, output.txt):

Sensitivities (each alone, output.txt): 2027 revenue ±$1B ±$6.6; CCS growth 2028-31 ±5 points a year −$34 / +$38; CCS 2029 growth at 20% instead of 10% +$18; CCS margin ±1 point ±$37; CCS margin to 6.2% in 2031 $230, to 8.2% $297; working capital ±2 points of revenue ∓$6; 2027 capex +$500M −$2; outside view ±20% ±$24; Celestica's own beta (1.56) −$34.

5. What the market expects and where we differ

At $246.65 the enterprise value is $29.1B (127.4M shares less $2.3B of net cash after debt, pension and the sold receivables). That is 8.3x our 2027 operating profit and 6.9x our 2031 (market.txt). Our bear case exits at 8.7x 2031 profit. So the market prices our forecast as if Celestica were already a shrinking contract manufacturer. Alternatively it doubts the forecast: one of these alone reaches the price (alpha research odds):

• CCS growth 5.1 points a year below ours in 2028-31, about +17%, +5%, −2%, 0%;
• the CCS margin 0.9 point lower, about 6.8% by 2031;
• 2027 revenue of $36.1B, 10% below the outlook a company that has beaten nine straight guides gave the same day;
• an exit at 0.71x the outside-view anchor, about 8.4x EV/EBIT, the bear's multiple on the central's profit.

The Street's targets (average $456) sit above our bull's exit path for 2026-27; its revenue is $36.0B for 2027 (before the outlook), $49.0B for 2028 (ours $49.8B) and $75.4B for 2030 (ours $56.3B). Our value differs from the Street's on 2029-30, where we expect digestion, not on the next two years. Our scenarios put more than half their weight above the price (odds.json: P 58%, simulated median $281, at our central). The spread is wide (10th to 90th $122-645), and most of it is the exit multiple itself.

Variant view. The market and we agree that 2027 is a large year. We part on what follows. The price pays as if profit peaks in 2027-28 and then falls; we expect it to flatten at about $3.3-3.4B of NOPAT, because Celestica now designs a large share of what it builds and the capacity build is funded. What settles it: 2028 guidance (late January 2027 at the earliest), and whether the CCS margin holds near 9% through 2027.

6. Why it keeps getting cheaper

The price fell 41.6% from 2026-04-27 to $246.65 while XLK, the technology sector fund, rose 16.0% (research-check, market.txt). It fell 11.7% on 2026-10-22 alone, then 3.3% and 5.1% on the two days before the release. The four questions research-check flags:

1. Is our value stale or wrong? Not stale: it is built on the release issued after today's close. It may still be wrong in the direction the price points. Two outside reviewers cut it from $310 to $281 on filing evidence (section 2). The margin through a cycle is the open question, and 6.2% by 2031 gives $230, below the price.
2. Are forward owner earnings falling? No. 2026 and 2027 outlooks rose (2026 revenue $16.0B in October 2025 to $21.0B; 2027 now $40.0B), and our 2027-28 NOPAT is 17-21% above the old report's. Free cash flow is weaker: the 2026 outlook was cut to $460M and 2027 is a net use of cash (working capital), which our 7% charge carries.
3. Is the market paying a lower multiple for this kind of business? Yes, across AI hardware, not Celestica alone. Since 2026-04-27 Fabrinet fell 52.5%, Vertiv 38.0%, Jabil 19.1%, Credo 16.3% and Flex 14.5% (market.txt); Arista (+2.8%) and Nvidia (−0.3%) held. At 8.3x our 2027 profit the market pays below our bear's 8.7x exit. We tested our exit against it: at 0.71x the anchor (8.4x) the value reaches the price (section 5). We keep the anchor because its legs are ten-year medians that include past downturns, but this is the de-rating path in section 7.
4. Sector, factor or positioning pressure? On 2026-10-22 the whole group fell together: Celestica −11.7%, Flex −11.1%, Fabrinet −10.8%, Credo −10.7%, Vertiv −7.8%, Arista −7.2%, XLK −3.3% (market.txt). That is a sector move, not news about Celestica. We found no Celestica announcement that day (TradingView news for NYSE:CLS, read 2026-10-26). The cause of the group's fall that day is not established; press accounts of October cite worries about debt-funded AI data-centre building and higher interest rates, and a nine-session fall in the semiconductor fund to 2026-10-16 (Benzinga and Reuters via search summaries, secondhand). We did not verify those for 2026-10-22, so we treat it as speculation.

Supported: 3 and 4 (a group de-rating). Open: 1 (the margin through a cycle). Not supported: 2. What would change the call: a CCS margin below 8.5% in 2027 quarters (about −$29), or 2028 guidance that implies a fall from 2027. A lower price alone does not change fair value. Chart read (timing only, not an input): the close is at its 52-week low ($244.51 low on market data), far below its 50- and 200-day averages ($329, $333), a falling trend with no base yet.

7. Bear case and why the other side could be right

The bear ($111) is an AI spending pause: 2027 lands 15% below the outlook, 2029 CCS revenue falls 20%, and the margin returns to its 2023 level. Three customers are two-thirds of revenue, and in Q3 inventory grew $1.1B in a quarter, so a pause would hit revenue and cash together. The bear's exit is 8.7x 2031 EBIT.

De-rating path. If our central forecast holds but the market keeps paying the bear's 8.7x at the end of 2031, the stock is worth $232 today (output.txt), 6% below the price. Celestica fetched 8.4-14.7x at its 2016-2021 year-ends, when it grew 0-8% with concentrated customers. Two pieces of evidence would keep the multiple there: hyperscaler capex guides flattening for 2028, or a top-three customer moving a program to a Taiwanese design manufacturer or in-house.

The other side. The case for Avoid is the margin and the multiple together: if the CCS margin returns toward 2023's 6.2% and the market keeps paying 8-9x, value is near $200. The bull's strongest evidence is dated and specific: management says customers' 2028 forecasts are "meaningfully higher" than $40.0B, and its record backs it, with nine straight beats and four raises of the 2026 outlook. 20% CCS growth in 2029 instead of 10% alone adds $18. What decides it is whether the AI build is a cycle (our view) or a long build-out at slowing growth. That will not be dated before 2028 guidance.

8. Next results preview: Q4 2026, expected late January 2027 (date not announced)

Each $1B of 2027 revenue above our $41.4B is worth about $6.6 a share (through the model, later years growing from it). The old report's 2027 outlook forecast ($35.0B, range $33.8-38.0B) falls due on 2026-10-27 and is graded then; the release puts it at $40.0B, above the range.

9. Catalysts and falsifiers

• 2026-10-27, call and Investor Day (company-announced). Any 2028 or multi-year revenue target: one at or above $55B for 2028 (10% above our $49.8B) would move our 2028-29 path up; one below $45B moves toward the bear. Also 2027 capex, working capital, customer concentration and any multi-year margin target.
• Late January 2027, Q4 2026 results (expected; earlier Q4 releases were 2025-01-29 and 2026-01-28): section 8.
• Falsifiers. 2027 outlook cut below $38B: weight the bear. CCS margin at 8.5% or lower through 2027 (our 9.25%): about $29 lower if it lasts. Sold receivables above $1B, or customer deposits falling while inventory rises: working capital is worse than our 7%. A top-three customer below 10% of revenue in a 2027 filing: toward the bear. A 2028 target at or above $55B with the margin held near 9%: our slowdown and margin glide are too low.

10. What the reviewers found and what we did

Two outside reviewers (Codex, read-only, and an Opus reviewer) read the pushed model with the market and flags. Codex advised against landing; the Opus reviewer advised landing with fixes.

11. Sources

• Celestica Q3 2026 press release, Exhibit 99.1 to the 8-K accepted 2026-10-26 20:23:49 UTC, https://www.sec.gov/Archives/edgar/data/1030894/000103089426000049/cls-20260930prcondensedfs.htm (primary; data/q3-2026-release.htm).
• Celestica 10-Q for the quarter ended 2026-09-30, accepted 2026-10-26 20:26:17 UTC, https://www.sec.gov/Archives/edgar/data/1030894/000103089426000050/cls-20260930.htm (primary; data/10q-q3-2026.htm), with its Exhibit 10.1 amending the receivables sale agreement.
• The 2026-10-15 report's sources: 10-K FY2025 (segment margins and HPS shares 2023-2025), 10-Qs Q1-Q2 2026, releases Q2 2024-Q2 2026, 8-Ks of 2026-08-07 and 2026-09-11, Q2 2026 call notes.
• Market data annual statements (CLS and peers) and fiscal year-end enterprise values, fetched 2026-10-26 (data/peer-income.json, data/exit-anchor.json); peer weekly returns saved 2026-10-15.
• Market data daily closes for CLS, peers, XLK and SPY through 2026-10-26 (data/market-closes.json); market data Street targets and estimates via scripts/research-check.py, 2026-10-26.
• TradingView news for NYSE:CLS, read 2026-10-26 (no Celestica story on 2026-10-22); web search summaries of October 2026 AI-hardware and semiconductor declines (Benzinga, Reuters; secondhand, not verified).
• Damodaran implied equity risk premium and risk-free rate, 2026-10-01 (the shared method's dated pair).

Revision history

• 2026-10-15, $286, Hold (bear $100, bull $530). First report: 2027 at management's +65% floor ($35.0B), CCS slowing to 3-5% by 2030-31, exit the median of a judged 15x, a fade and an outside view. Partly unblinded (offering and award values seen before the model).
• Restated 2026-10-16 to 2026-10-23, $318, then $260-264 (method). The same forecast under the shared method's exit rules: the mean of fade and outside view (2026-10-16), anchor (d) (2026-10-18) and the case spread (2026-10-24). Last value $263.97, bear $99, bull $539.
• 2026-10-26, $281, Buy (+$17 from $264). Data: Q3 2026 results and a $40.0B 2027 outlook (+$38 for the outlook, +$9 for our half-beat on it, −$3 tax, −$6 working capital, the sold receivables now a claim). Judgment: the 2028-31 growth path unchanged; the CCS margin back on the old path after 2027 and gliding to 7.72% by 2031 (−$16 net of the 2027 margin). Method: rule H's mature return, −$6. The blind draft was $310; the reviewers' three fixes took $29 (section 2). Made after the price was seen.