Celestica’s third-quarter revenue rose 83%, and its first outlook for 2027, $40.0bn, came in 12% above analysts’ estimates. Three customers take two-thirds of its sales, and cash still lags profit. On the 2027 outlook and growth that fades from 2028, the shares are worth $305.11 to $456.19, most likely $364.81. At $276.15 the price needs revenue 28% below that path from 2028, which puts 2028 below the 2027 outlook. We rate the shares undervalued.
The outlook moved further than the quarter. Third-quarter revenue was $5.83bn against guidance of $5.25–5.55bn, and adjusted EPS $3.30 against $2.88–3.08. The 2026 outlook rose for the fourth time this year, to $21.0bn of revenue and $11.55 of adjusted EPS. The first outlook for 2027 is $40.0bn of revenue, up 90%, with a 9.0% adjusted operating margin and $21.50 of adjusted EPS; analysts had expected about $35.7bn. Management says customers’ 2028 demand forecasts are “meaningfully higher” than the 2027 outlook and that it is “very confident” in 2028, but gave no figure.
The cash lags the earnings, and three customers hold the keys. Operating cash flow for the nine months was $1,157m, but $701m of it came from selling receivables to a bank under an uncommitted programme that had nothing outstanding in January. Payables stretched to 74 days from 54 a year earlier. The 2026 free cash flow outlook was cut to $460m, and 2027 capital spending will be $1.5–2.0bn, up from $201m in 2025. Three customers took 35%, 17% and 14% of third-quarter revenue. Management expects the company to use cash in 2027 and is targeting positive free cash flow again in 2028; our base case has −$240m after stock pay in 2027 and $1.4bn in 2028.
The price is a 2028 downturn. Our base case takes the company’s 2027 outlook and lets growth fade from 2028: $364.81 a share. At $276.15 the market needs revenue 28% below our path from 2028 onwards, which puts 2028 at $36.1bn, below the 2027 outlook; or a 12.0% discount rate; or a long-run adjusted margin of 6.9%, where Jabil and Flex operate. The price is 9% below the bottom of our $305.11–$456.19 band, closer to that downturn than to our base case.
At $276.15 Celestica’s market value is about $35.2bn on 127.4m shares, counting the 1.3m share awards already granted, and its enterprise value about $32.0bn after $3.1bn of net cash. That is 23.9 times this year’s guided adjusted EPS of $11.55 and 12.8 times next year’s $21.50; Jabil, Flex and Fabrinet trade at 16 to 21 times their own outlooks for the next fiscal year. A manufacturer guiding to 90% growth can earn a multiple above slower peers; the question is whether 2027 is a step or a peak. Our base case answers it explicitly: revenue reaches the company’s $40.0bn in 2027 and $50.0bn in 2028, growth then fades to 3% by 2035, the adjusted operating margin peaks at 9.2% and settles at 8.5%, and stock pay is charged as a cost. That path is worth $364.81 a share, 32% above the price.
Run the model in Section 02 backwards and the price needs one of three things, each on its own with everything else held at the base case. Revenue 28% below our path in every year from 2028: $36.1bn in 2028, 10% below the 2027 outlook, $45.7bn in 2030 and $59.7bn in 2036, against our $50.0bn, $63.25bn and $82.7bn. Or an adjusted operating margin 1.6 points lower every year, 6.9% in the long run, where Jabil and Flex operate, for a mix that has earned 8–9% for two years. Or a discount rate of 12.0% instead of 10.0%. Put plainly, the market is paying for a 2028 digestion year followed by recovery: our bear case in shape, if not in depth. The price sits 9% below the bottom of our $305.11–$456.19 band.
The signature measure for Celestica is how narrow its base is and how its growth is being paid for. The first bar is third-quarter revenue by customer; the second is where nine months of operating cash flow came from; the rows below set the revenue the company has reported and guided against what the share price implies for 2028 and what each of our cases assumes, all on one scale.
Customer shares from the Form 10-Q, note 3; the company does not name the customers. Cash flow from the Form 10-Q cash flow statement and its description of the receivables sales programme, whose limit was raised to $1.0bn from $450m in September. The 2028 figure implied by $276.15 is the revenue our base-case model needs to equal the share price when every year from 2028 is scaled down by the same proportion, with margins, discount rate and terminal growth unchanged; it is derived, not a forecast. Two-thirds of the revenue rests on three buyers, and $701m of the $1,157m of operating cash flow came from a bank that is not committed to keep buying receivables.
The cloud companies’ first signals on 2027 capital spending: Alphabet and Microsoft report on 28 October and Amazon on 29 October. A 2027 framework that grows supports the company’s outlook and our base case; a flat one would make 2028 look like the bear case. The second test is the fourth quarter’s cash: free cash flow against the $460m outlook, and how much of it again rests on receivables sold and 74 days of payables.
The rate backdrop is not friendly to a cyclical growth stock. The 10-year Treasury yield was near 5.0% on 26 October. We use a 10.0% discount rate in every case: Celestica has net cash, and the specific risks of its business, three customers and one spending cycle, are carried in the scenarios and their weights rather than added to the rate. The band runs the rate from 9% to 11%; at 11.0% the base case is worth $314.81, still above the price.
2026–36 revenue CAGR · 2036 operating margin after stock pay · free cash flow in 2027 and in 2036 · terminal value is of enterprise value · shares, after the swap.
This is the model that produced every figure in this report, reduced to its ten moving parts. It is an unlevered discounted cash flow on a ten-year path, 2027 to 2036, with the fourth quarter of 2026 as a stub. Free cash flow to the firm is adjusted operating income less stock pay, taxed at 21%, plus depreciation, less capital spending, less the increase in working capital. Working capital is held at each year end as a share of the revenue run-rate the company exits the year at, because a manufacturer that doubles in a year carries inventory and receivables for next year’s volume, not last year’s. Depreciation runs off the plant, so the spending of 2027–28 is charged back over nine years rather than at a fixed share of revenue. The valuation date is 27 October 2026; each year is discounted at its mid-point, and the Gordon terminal value at 3% on 2037 cash flow is discounted from the end of 2036. On base settings the terminal value is 61% of enterprise value. The scenario cards load whole published paths, and the sliders bend the loaded path rather than replace it, so each card reproduces its published value exactly.
Why these inputs. 2027 revenue of $40.0bn (company outlook). Given on 27 October, with CCS up about 110% and ATS about 10%; analysts had expected about $35.7bn. We add no beat; that belongs to the bull case. 2028 of $50.0bn (assumed). Management gave no 2028 figure, but said customers’ 2028 demand forecasts, counting only programmes already awarded, are “meaningfully higher” than the 2027 outlook, and on the call the Group President said management is “very confident right now in 2028”, with programmes running into 2029 and in some cases 2030. We read that as 20–30% growth and take the middle. 2030 of $63.25bn and 2036 of $82.7bn (assumed). Growth fades from 15% in 2029 to 3% by 2035; between the settings the path keeps the shape of the loaded case. Long-run adjusted operating margin of 8.5% (assumed). 9.0% in 2027 as guided and a 9.2% peak in 2028 at full factories, easing to 8.5% by 2033 as growth slows and second sources arrive; the third quarter was 8.5%, and the company earned roughly 5% in 2022. Stock pay of 0.45% of revenue (assumed), close to this year’s rate, charged as a cost. Working capital of 8.0% of exit run-rate revenue (assumed). Measured before receivable sales it was about 6.6% of annualised third-quarter revenue, held down by 74 days of payables; it steps up to 7.4% at the end of 2027, because revenue is weighted to the second half, and drifts back to 8.0%, the end-2025 level, by 2030. Capital spending at 100% of plan (assumed): $1.76bn in 2027, the middle of the $1.5–2.0bn outlook, $1.75bn in 2028, settling at 1.8% of revenue. Management said the 2027 plan also covers capacity for awarded programmes in 2028; we keep 2028 near the 2027 level to allow for new awards, and cutting it to $1.0bn would add about $3 a share. Discount rate 10.0% and terminal growth 3.0% (assumed), the same in every case.
What the controls hold fixed. Tax at 21%, the company’s adjusted rate for 2027; the GAAP rate has been 13–15% this year with deferred-tax benefits, and the Pillar Two minimum tax applies. Plant opens 2027 at $1,598m and depreciates over nine years, with half a year on each year’s spending; in the terminal year capital spending is 1.25 times depreciation, so plant keeps growing with revenue. Receivables sold, $700.8m, are deducted as debt because the bank can stop buying, and working capital is modelled before those sales, so the cash the bank advanced is not counted twice. The total return swap on 1.25m of the company’s own shares is treated as a synthetic repurchase: the shares come out of the count and the $361.1m strike notional is carried as debt. The fair-value band in Section 03 is this model run across the discount rate and terminal growth, and every cell of that grid is reproducible from the two controls.
Derivations, base case. Fourth quarter of 2026: revenue $6,424.7m, the $21.0bn outlook less $14,575.3m in the first nine months; free cash flow $26.5m, the company’s $460m free cash flow outlook less $412.5m in nine months on its own definition, less $21m of stock pay; discounted 0.09 years. Working capital opens 2027 at $1,761m: about $840m of operating working capital at 30 September plus the $700.8m of receivables sold, plus the $220m fourth-quarter build that the cash flow outlook implies (derived). Plant opens 2027 at $1,598.3m: $1,403.3m at 30 September plus about $235m of fourth-quarter spending (the company’s figure of about $1bn for 2026, restated on 27 October, less $767.4m gross in nine months), less $40m of depreciation. Shares 126.16m: 126.11m on the cover of the Form 10-Q (21 October 2026), 1.30m for awards already granted (the third quarter’s diluted less basic weighted average), less the 1.25m shares under the swap. Net cash $1,985.0m: $3,961.3m of cash less $735.6m of term loans at face, $88.0m of finance leases, $700.8m of receivables sold, $90.8m of pension and post-employment obligations and the $361.1m swap notional (1.25m × $288.87).
The number above is not a fact about Celestica. It is what these assumptions produce, and the one that decides it is not in any filing: whether revenue keeps growing in 2028 or falls back below the 2027 outlook. The rating holds in every cell of the sensitivity grid; it changes to fairly valued at a price above $305.11, and a 2028–29 digestion, our bear case, is worth $167.23. Move the 2028 and 2030 revenue settings until they are yours, and use the number that falls out.
Each card loads its published case into the model above and prints its value: $167.23, $364.81, $518.40 and $727.54. All four use the same 10.0% discount rate and 3.0% terminal growth, so they differ only in how much Celestica sells, at what margin and with how much cash tied up in working capital and plant; concentration and cycle risk sit in the cases and the weights, not in the rate. Weighting the four at 25 / 45 / 25 / 5 gives $371.95.
| Input | Bear (25%) | Base (45%) | Bull (25%) | Super-bull (5%) |
|---|---|---|---|---|
| 2027 revenue | $36.0bn | $40.0bn | $42.0bn | $44.0bn |
| 2028 revenue | $30.6bn | $50.0bn | $60.0bn | $68.0bn |
| Growth, 2029 / 2030 | −5% / +6% | 15% / 10% | 20% / 13% | 25% / 18% |
| 2030 revenue | $30.8bn | $63.25bn | $81.4bn | $100.3bn |
| 2036 revenue | $39.7bn | $82.7bn | $112.6bn | $150.2bn |
| Adjusted operating margin | 8.6% → 7.0% → 7.2% | 9.0% → 9.2% → 8.5% | 9.2% → 9.6% → 9.2% | 9.3% → 10.0% |
| Working capital, % of exit run-rate revenue | 7.9% → 8.5% | 7.4% → 8.0% | 7.2% → 7.8% | 7.0% → 7.6% |
| Capital spending, 2027 | $1.62bn | $1.76bn | $1.89bn | $2.02bn |
| Capital spending, % of revenue | 4.5% → 1.7% | 4.4% → 1.8% | 4.5% → 1.9% | 4.6% → 2.0% |
| Free cash flow, 2027 / 2030 | $51m / $1.39bn | −$240m / $2.96bn | −$400m / $3.92bn | −$523m / $4.74bn |
| Common to all | Discount rate 10.0% and terminal growth 3.0%; tax 21%; stock pay 0.45% of revenue charged as a cost; depreciation from plant over nine years; valued at 27 October 2026 from the 30 September 2026 balance sheet; receivables sold and the swap notional deducted as debt | |||
| Value per share | $167.23 | $364.81 | $518.40 | $727.54 |
2027 base revenue is the company’s outlook, given on 27 October; analysts had expected about $35.7bn. Working capital is measured before receivable sales, as a share of the run-rate revenue each year ends at.
| $ millions, base case | 2027 | 2028 | 2029 | 2030 | 2032 | 2034 | 2036 |
|---|---|---|---|---|---|---|---|
| Revenue | 40,000 | 50,000 | 57,500 | 63,250 | 71,400 | 77,597 | 82,723 |
| Growth | 90% | 25% | 15% | 10% | 5.5% | 4% | 3% |
| Adjusted operating income | 3,600 | 4,600 | 5,232 | 5,629 | 6,140 | 6,596 | 7,031 |
| Less stock pay | (180) | (225) | (259) | (285) | (321) | (349) | (372) |
| Less tax at 21% | (718) | (919) | (1,044) | (1,122) | (1,222) | (1,312) | (1,398) |
| Plus depreciation | 275 | 440 | 571 | 668 | 814 | 929 | 1,037 |
| Less capital spending | (1,760) | (1,750) | (1,495) | (1,392) | (1,357) | (1,397) | (1,489) |
| Less increase in working capital | (1,457) | (786) | (644) | (542) | (283) | (231) | (195) |
| Free cash flow to the firm | (240) | 1,360 | 2,362 | 2,957 | 3,771 | 4,236 | 4,614 |
Sonde estimates. The fourth quarter of 2026 is a stub of $26.5m to 31 December. Working capital absorbs $3.4bn of cash between 2027 and 2030, which is why 2027 free cash flow is negative, in line with management’s expectation of a net use of cash.
| Base case | $m | Note |
|---|---|---|
| Present value of cash flows, Q4 2026–2036 | 17,231 | Mid-year convention, the fourth quarter as a stub |
| Present value of terminal value | 26,809 | 61% of enterprise value; 2037 free cash flow $4,951m at 10.0% less 3.0% |
| Enterprise value | 44,040 | |
| + Cash | 3,961 | 30 September 2026 |
| − Term loans, at face | (736) | Term A $246.9m, Term B $488.7m; the $1.74bn revolver is undrawn |
| − Finance leases | (88) | |
| − Receivables sold | (701) | Uncommitted programme; the bank can stop buying |
| − Pension and post-employment obligations | (91) | |
| − Total return swap notional | (361) | 1.25m shares at a $288.87 strike |
| Equity value | 46,025 | Components rounded |
| Shares, millions | 126.16 | Outstanding and granted awards, less the swap shares |
| Value per share | $364.81 |
The grid holds the base case fixed and varies the discount rate against terminal growth. Its corners are the published fair-value band, $305.11–$456.19: a WACC of 9–11% by terminal growth of 2.5–3.5%.
| WACC ↓ / terminal growth → | 2.5% | 3.0% | 3.5% |
|---|---|---|---|
| 9.0% | $411.47 | $431.97 | $456.19 |
| 9.5% | $379.00 | $395.74 | $415.28 |
| 10.0% base | $350.98 | $364.81 | $380.76 |
| 10.5% | $326.56 | $338.10 | $351.28 |
| 11.0% | $305.11 | $314.81 | $325.81 |
The $276.15 price lies below every cell, 9% under the lowest: undervalued. The rating would change to fairly valued above $305.11. Run across the same grid, the probability-weighted value spans $310.51–$466.09. The second table holds the rates at the base case and moves one operating input at a time:
| Change to the base case, all else held | Value | Change |
|---|---|---|
| Revenue 10% higher from 2028 | $396.74 | +9% |
| Revenue 10% lower from 2028 | $332.88 | −9% |
| Revenue 20% lower from 2028 | $300.96 | −18% |
| Adjusted margin 1 point higher every year | $418.68 | +15% |
| Adjusted margin 1 point lower every year | $310.94 | −15% |
| Working capital 2 points of revenue higher | $351.94 | −4% |
| Working capital 2 points of revenue lower | $377.67 | +4% |
| Capital spending 25% higher throughout | $350.11 | −4% |
| Tax at 23% | $353.66 | −3% |
| Stock pay treated as free | $389.05 | +7% |
Modal base case $364.81, 32% above the price. Fair-value band $305.11–$456.19, the WACC 9–11% by terminal growth 2.5–3.5% grid, which anchors the rating; the price is below it: undervalued. Probability-weighted value $371.95 across the four cases at our own weights, 35% above the price and close to the base case. There is no material skew to name: the bear sits $197.58 below the base and the bull $153.59 above it, and the super-bull’s small weight adds a long tail.
| Single input changed, all else at base | Price implies | Our base |
|---|---|---|
| Revenue from 2028, every year | 28% lower | The outlook, then a fade |
| 2028 revenue on that path | $36.1bn | $50.0bn |
| 2030 revenue on that path | $45.7bn | $63.25bn |
| Adjusted operating margin, every year | −1.6 points; 6.9% long-run | 8.5% long-run |
| Discount rate | 12.0% | 10.0% |
The revenue reading is the most natural way to see the price: the market is paying for 2027 and very little after it, with 2028 revenue 10% below the 2027 outlook and recovery from that lower base. A long-run margin of 6.9% would put Celestica back where Jabil and Flex operate, despite a design-led mix that has earned 8–9% for two years. Our disagreement with the market rests on one judgement: that a 2028 decline is less likely than continued, slower growth, given what customers have told the company about 2028, management’s stated confidence in it and the scale of 2026 spending already committed. If the cloud companies’ first 2027 spending plans, due on 28 and 29 October, come in flat, the market’s reading gains ground and so does our bear case.
| Scenario | Value | Against price | Weight | Contribution |
|---|---|---|---|---|
| Bear | $167.23 | −39% | 25% | $41.81 |
| Base | $364.81 | +32% | 45% | $164.16 |
| Bull | $518.40 | +88% | 25% | $129.60 |
| Super-bull | $727.54 | +163% | 5% | $36.38 |
| Probability-weighted | $371.95 | +35% | 100% | $371.95 |
We weight the bear at 25% because two-thirds of revenue comes from three customers whose 2027 budgets are not yet public, hyperscaler spending paused in 2019 and 2023, and the largest spenders are now raising outside capital to keep building. We weight it no higher because the company has published an outlook for the whole of 2027 built from customer demand, and says customers’ 2028 forecasts are higher still. Weights are our judgement, not a model output.
| Risk | Priced in | Not also in |
|---|---|---|
| AI spending digestion | The bear revenue path (2028 down 15%, 2029 down 5%) and its 25% weight | The base case, the discount rate or terminal growth |
| Customer concentration | Bear revenue and margin | The discount rate |
| Payables and receivable sales reversing | Working capital rising to 8.0% of run-rate revenue in the base case and 8.5% in the bear; receivables sold deducted as debt | The margin |
| Second sourcing and pricing | Base margin fading from 9.2% to 8.5%; the bear at 7.0–7.2% | Base-case revenue |
| Capital intensity | Capital spending in every case, peaking in 2027–28 and depreciated from plant | A terminal multiple (there is none) |
| Dilution from share awards | Stock pay charged as a cost; granted awards in the share count | An add-back |
| Equity-market and rate risk | The 10.0% discount rate and the 9–11% grid | The cash flows |
| $ millions | Value | Note |
|---|---|---|
| Market value | 34,827 | 126.11m shares (10-Q cover, 21 Oct) × $276.15 |
| + Term loans, at face | 736 | Term A $246.9m, Term B $488.7m |
| + Finance leases | 88 | |
| − Cash | (3,961) | 30 Sep 2026 |
| Enterprise value | 31,689 | Receivables sold excluded, as for peers |
Source: Form 10-Q at 30 September 2026; reference price the NYSE close on 27 October 2026. Peers’ own receivable sales are also off balance sheet, so Celestica’s $700.8m is left out here for comparability; the valuation deducts it as debt. With the 1.3m granted awards the market value is $35.2bn and the enterprise value $32.0bn.
| Measure | CLS | Basis |
|---|---|---|
| EV / 2026E revenue | 1.5× | $21.0bn outlook |
| EV / 2027E revenue | 0.8× | $40.0bn outlook |
| EV / trailing EBITDA | 18.8× | Twelve months to 30 September, GAAP operating income plus depreciation and amortisation |
| P/E, 2026E adjusted | 23.9× | $11.55 outlook |
| P/E, 2027E adjusted | 12.8× | $21.50 outlook |
| P/E, trailing GAAP | 28.0× | Twelve months to 30 September |
| Price / trailing free cash flow | 61× | Operating cash flow less capital spending, including receivables sold |
| Company | Mkt value | EV | Growth, latest quarter | Op. margin, TTM | Fwd P/E | PEG | EV / revenue | EV / EBITDA | P / FCF |
|---|---|---|---|---|---|---|---|---|---|
| Celestica $276.15 | $34.8bn | $31.7bn | +83% | 8.3% | 12.8× | 0.15 | 1.74× | 18.8× | 61× |
| Jabil $283.36 | $29.2bn | $30.8bn | +29% | 4.7% | 16.1× | 0.47 | 0.86× | 12.9× | 21× |
| Flex $83.44 | $30.8bn | $33.2bn | +21% | 5.0% | 18.2× | 0.47 | 1.13× | 16.5× | 37× |
| Sanmina $169.55 | $9.1bn | $9.4bn | +70% | 4.2% | 13.2× | n/m | 0.74× | 13.6× | 15× |
| Fabrinet $343.57 | $12.3bn | $11.4bn | +45% | 10.0% | 20.6× | 1.11 | 2.46× | 21.5× | n/m |
All five at their NYSE closing prices on 27 October 2026 (regular-session daily bars, Interactive Brokers). Market value is price × the share count on the cover of each company’s latest filing; enterprise value adds net debt, or deducts net cash, from the latest balance sheet. Trailing figures are GAAP: growth is the latest quarter against a year earlier; operating margin, EV/revenue, EV/EBITDA (operating income plus depreciation and amortisation) and P/FCF (operating cash flow less capital spending) use the latest twelve months. Forward P/E uses each company’s own adjusted EPS outlook; PEG divides it by the growth that outlook represents. Celestica: twelve months to 30 Sep 2026; 126.11m shares (Q3 Form 10-Q cover, 21 Oct 2026); 2027 EPS outlook. Jabil: fiscal year to 31 Aug 2026; 102.97m shares (Form 10-K cover, 20 Oct 2026); fiscal 2027 outlook. Flex: twelve months to 26 Jun 2026; 369.40m shares (Form 10-Q cover, 24 Jul 2026); fiscal 2027 outlook midpoint. Sanmina: twelve months to 27 Jun 2026; 53.60m shares (Form 10-Q cover, 20 Jul 2026); fourth-quarter fiscal 2026 outlook midpoint × 4. Fabrinet: fiscal year to 26 Jun 2026; 35.83m shares (Form 10-K cover, 7 Aug 2026); first-quarter fiscal 2027 outlook midpoint × 4. Flex’s depreciation is its latest quarter annualised; Fabrinet’s free cash flow is close to nil after a year of heavy capital spending, so its P/FCF is not meaningful. Sanmina’s growth includes ZT Systems.
Cheap, fair or expensive? Expensive on trailing figures, cheap on the company’s own outlook. On trailing figures Celestica sits at the expensive end: 19 times trailing EBITDA against 13 to 22 for the peers, with only Fabrinet higher, and 1.7 times revenue against 0.7 to 2.5. Trailing figures miss most of the growth. On next year’s earnings it trades at 12.8 times against 13.2 to 20.6, and relative to growth, a PEG of 0.15, it is the cheapest in the group by a wide margin. The discount is the market’s answer to one question: are 2027 earnings a step on the way up, or a peak? Peers with slower growth earn a higher multiple because their earnings are judged more durable. The discount is compensation for concentration and cyclicality; the DCF carries those risks in the bear case and its weight, and in no other place, so they are not charged twice.
| What would expand the multiple | What would compress it |
|---|---|
| Hyperscalers’ 2027 spending guidance, due from 28 October, confirming the outlook | Any sign of a pause at the largest customer |
| 2028 visibility firming: a figure, or a raise to the 2027 outlook in January | A decline in adjusted gross margin |
| The customer list broadening beyond the three that take two-thirds of revenue | A second capital raise |
2028 is the valuation. The base case’s 2028 revenue is our reading of the company’s statement that customers’ 2028 forecasts are “meaningfully higher” than the 2027 outlook; no figure has been disclosed. Working capital is the largest single assumption after revenue: we let it return from 6.6% to 8.0% of revenue, measured before receivable sales; two points either way moves the base case by about $12.86 a share. Receivables sold ($701m) are deducted as debt because the programme is uncommitted; if the bank keeps buying indefinitely, the base case rises by about $5.55 a share. The fourth-quarter 2026 cash flow is derived from the company’s annual free cash flow outlook, and fourth-quarter capital spending from its figure of about $1bn for 2026, restated on the call. The 10.0% discount rate is below a strict capital-asset-pricing estimate for a stock with a beta well above 1 and a ten-year yield near 5.0%; at 11.0% the base case is worth $314.81. Terminal value is 61% of enterprise value. Peer forward earnings use each company’s own outlook for different fiscal years, two of them a single quarter multiplied by four. The reference price was set on a day the shares moved sharply on the results and the Investor Day; the rating is sensitive to the price chosen. The customers are not named in filings, and identities in market commentary are not used.
Celestica reported its third quarter, the three months to 30 September 2026, after the close on 26 October and filed its Form 10-Q the same day. The call and the 2026 Investor and Analyst Day followed on 27 October. The shares closed at $246.65 on 26 October and at $276.15 on 27 October, up 12.0%, after trading between $261.91 and $282.64.
| Q3 2026 | Value | Q3 2025 | Change | Why it matters |
|---|---|---|---|---|
| Revenue | $5,830m | $3,194m | +83% | Above the $5.25–5.55bn guide by $280m at the top |
| CCS segment revenue | $4,872m | $2,413m | +102% | 84% of the total, from 76% |
| Networking | $3,263m | $1,943m | +68% | Switches; the outlook was about +60% |
| Compute & Storage | $1,609m | $470m | +242% | Custom AI racks; the outlook was about +190% |
| ATS segment revenue | $957m | $781m | +23% | Capital equipment recovery; the outlook was mid-teens |
| HPS (designed products) | ≈$2.3bn | ≈$1.4bn | +70% | 40% of revenue, from 44%: compute grew faster |
| Adjusted gross margin | 11.2% | 11.7% | −0.5 pts | Mix shift to compute |
| Adjusted operating margin | 8.5% | 7.6% | +0.9 pts | A record; overheads 1.8% of revenue |
| CCS / ATS segment margin | 8.9% / 6.6% | 8.3% / 5.5% | +0.6 / +1.1 pts | Operating leverage in both |
| Adjusted EPS | $3.30 | $1.58 | +109% | Above the $2.88–3.08 guide |
| Free cash flow, company definition | $128m | $89m | +43% | Includes receivables sold; $701m outstanding |
| Cash cycle, days | 42 | 65 | −23 | Payables 74 days against 54 |
| Three largest customers | 66% | 59% | +7 pts | 35%, 17%, 14% |
| Adjusted return on invested capital | 64.7% | 37.5% | +27 pts | Working capital largely supplier-funded |
Sources: Q3 2026 results release and Form 10-Q, 26 October 2026; call prepared remarks, 27 October 2026. HPS revenue is the company’s approximate figure.
One: a larger beat than usual. Revenue of $5.83bn was 5% above the top of the guidance range and adjusted EPS of $3.30 was 7% above it. Networking grew 68% against an outlook of about 60%, Compute & Storage 242% against about 190%.
Two: 2027 at $40.0bn. Revenue up about 90%, CCS up about 110% and ATS about 10%, a 9.0% adjusted operating margin and $21.50 of adjusted EPS on about 127m shares. Revenue is expected to be meaningfully higher in the second half than the first.
Three: 2028 described, not quantified. Management said customers’ 2028 demand forecasts, counting only programmes already awarded, are “meaningfully higher” than the 2027 outlook. In the question-and-answer session the Group President said management is “very confident right now in 2028”, with some programmes running into 2029 and 2030. No figure and no long-term margin target were given.
Four: growth will consume cash. The 2026 free cash flow outlook fell to $460m from $600m to fund working capital ahead of 2027, the company will be a net user of cash in 2027, and 2027 capital spending is $1.5–2.0bn after about $1bn in 2026. The chief financial officer gave no range for the 2027 outflow and said the company is targeting a return to positive free cash flow in 2028.
Five: new programmes named. OpenAI’s “Jalapeno” platform (128 accelerators per dual rack) ramps in 2027; AMD’s Helios rack uses Celestica’s DS6010 switch from around year-end; a multi-generation relationship on Google’s TPU platform; and a co-packaged-optics 1.6T rack award from a hyperscaler, ramping in late 2027. Management expects 1.6T revenue to overtake 800G in 2027.
| 2026 revenue outlook | Oct 2025 | Jan 2026 | Apr 2026 | Jul 2026 | Oct 2026 |
|---|---|---|---|---|---|
| Revenue | $16.0bn | $17.0bn | $19.0bn | $20.5bn | $21.0bn |
Revenue rose $1.13bn from the second quarter, of which Networking added $610m and Compute & Storage $452m. The fourth quarter is guided to $6.275–6.575bn, up 76% at the midpoint, with Networking up about 65%, Compute & Storage about 160% and ATS in the low 20s, and an adjusted operating margin of 8.7%. That puts 2026 at $21.0bn: CCS about $17.4bn and ATS about $3.6bn at a record 6.4% margin. Margin expanded on overheads, not on price: adjusted gross margin fell half a point on mix, while adjusted SG&A fell to 1.8% of revenue from 3.0%. The 2027 outlook adds another 0.6 points of operating margin while research and development spending is set to exceed $300m, more than three times 2023, and HPS engineers rise above 2,000 from 1,600. The chief financial officer expects adjusted gross margin to stay in the low 11% range in 2027, depending on the mix of networking and compute, with no material difference between the halves; networking remains the higher-margin work, and the step to a 9% operating margin comes from operating leverage. The $1.5–2.0bn capital plan covers capacity for awarded programmes in both 2027 and 2028.
The tone was confident and expansive. The presenters were Rob Mionis, chair and chief executive; Todd Ankenmann, chief financial officer; Mandeep Chawla, Group President; Steven Dorwart and Todd Cooper, presidents of CCS and ATS; and Gavin Cato of CCS engineering. The chief executive called this an “AI investment super cycle” and described the momentum as unprecedented; the 2027 outlook was framed as the company’s “best assessment” of supply constraints and the timing of capacity ramps. Capital discipline was a recurring theme: returns on invested capital, payback periods and an investment-grade balance sheet. Organic growth is the first use of cash, acquisitions are opportunistic, and the buyback is to be renewed, with no repurchases made in the third quarter.
Ten analysts asked questions, and they concentrated on what the market has priced: whether supply can support $40bn, how much cash the growth will consume and when it comes back, and how much of 2028 is already awarded. TD Cowen, keeping its rating after the release, said investor concerns about free cash flow may persist.
| Firm | Question | Answer, in short |
|---|---|---|
| UBS | Can supply of silicon, optics, substrates and boards support $40bn? | The outlook already allows for ramp, component and execution risk; “high confidence”, with upside. Capacity is in place; the constraints are land, power and geopolitics. |
| Scotiabank | Where is the engineering build-out, and is it aimed at 2028? | 500–600 engineers added a year across nine design centres; about 27 programmes for 2027, mostly in test; 60–70 programmes active, many for 2028 and later. |
| J.P. Morgan | How much cash will 2027 use, and when does free cash flow turn positive? | No range: 2027 is “a year of investment” for about $20bn of extra revenue, funded by the August equity and about $4bn of cash. Positive free cash flow targeted for 2028. |
| BNP Paribas | Does switch and rack complexity lift margins? | Programmes are chosen partly on margin. Gross margin in the low 11% range in 2027, depending on mix; networking is margin-accretive; operating leverage takes operating margin to 9%. |
| Susquehanna | Pluggable or co-packaged optics at 1.6T? Why does compute outgrow networking in 2027? | Mostly pluggable at 1.6T, co-packaged more at 3.2T. Compute growth comes from the TPU platform and OpenAI, both ramping in the second half of 2027. |
| TD Cowen | Do the two halves of 2027 differ in growth or margin? | No material gross-margin difference between the halves; about 11% for the year, with quarterly mix aside. |
| Bernstein | Is the OpenAI scale-up work winning other customers? | A proof point that has opened other opportunities; the company is being selective. |
| Goldman Sachs | How is capital spending balanced against returns and 2028 demand? | $1.5–2.0bn covers awarded programmes and capacity for 2027 and 2028; each investment is tested for return and payback, and reassessed as new awards arrive. |
| Stifel | Does 2028 confidence rest on follow-on generations or unawarded work? | “Very confident right now in 2028”; visibility into 2029 and some of 2030; some programmes are awarded before the silicon exists. Capacity is planned for awarded work only; the funnel is upside. |
| BMO | Is 2027 scale-up revenue mainly AMD’s Helios and OpenAI? What about 2028? | Mostly awarded programmes; scale-up is a widening opportunity. |
Paraphrased from a published transcript of the 27 October call (Investing.com), whose record ends during the last answer; every figure checked against the release and the Form 10-Q.
| Positive surprises | Negative surprises |
|---|---|
| The size of the 2027 outlook: $40.0bn against about $35.7bn expected | The 2026 free cash flow outlook cut by $140m, and a net use of cash expected in 2027 |
| 1.6T overtaking 800G in 2027, and a co-packaged-optics award | $701m of receivables sold, from nil at the start of the year; inventory up $1.1bn in the quarter |
| Disclosure of the Google TPU relationship and the OpenAI and AMD programme details | Adjusted gross margin down half a point; the three largest customers up to 66% of revenue |
| ATS growth of 23%, with double-digit growth targeted for 2027 | No 2028 figure and no long-term margin target, after an Investor Day |
| A target to return to positive free cash flow in 2028, and visibility described into 2029–30 | No range for how much cash 2027 will use |
What to watch next. Fourth-quarter free cash flow against the $460m outlook and the year-end receivables-sold balance; the phasing of 2027 between halves; first revenue from AMD’s Helios switch around year-end and from OpenAI’s racks in 2027; the cloud companies’ 2027 capital-spending signals on 28–29 October; and whether the January results move the 2027 outlook, as every 2026 report moved the 2026 outlook.
Celestica is a Toronto-based electronics manufacturer. It began as IBM’s Canadian manufacturing arm and has been listed since 1998. For most of its history it built other companies’ products to their designs, a low-margin business known as electronics manufacturing services. Over the past five years it has moved towards designing the hardware itself, mainly the high-speed network switches that connect servers inside data centres, and towards building complete racks of AI servers for the largest cloud companies.
It reports two segments. Connectivity & Cloud Solutions (CCS) sells network switches (the Networking end market, called Communications until this quarter) and servers and storage (Compute & Storage, formerly Enterprise), mostly to the large cloud companies known as hyperscalers and, increasingly, to AI laboratories. Advanced Technology Solutions (ATS) makes equipment for aerospace and defence, industrial, health technology and semiconductor capital equipment customers. Inside CCS sits Hardware Platform Solutions (HPS), the products Celestica designs alone or with the customer: about $2.3bn in the third quarter, up 70% and 40% of the total, against 21% of revenue in 2023 and 41% in 2025. The company says HPS carries a higher margin than build-to-print work because the design is priced in. Production is concentrated in Malaysia, Thailand, Mexico, China and Japan, with a growing footprint in Texas.
| End market | 2025 revenue | Q3 2026 | Q3 change y/y | What drives it |
|---|---|---|---|---|
| NetworkingCCS | $7,126m | $3,263m | +68% | 800G Ethernet switches for hyperscalers; 1.6T switches ramping; switches inside customers’ AI racks |
| Compute & StorageCCS | $2,062m | $1,609m | +242% | AI servers and racks built around customers’ own accelerator chips; storage |
| ATS | $3,202m | $957m | +23% | Chip-making equipment, aerospace and defence, industrial, health technology |
| Total | $12,391m | $5,830m | +83% | CCS segment margin 8.9%, ATS 6.6% in Q3 2026 |
Source: 2025 Form 10-K; Q3 2026 Form 10-Q, note 3. Segment margin is segment income as a share of segment revenue, before stock pay and acquired-intangible amortisation.
The company buys components, assembles and tests products, and ships them. In most switches it buys the main chip itself, so the chip’s cost passes through revenue. In most AI servers the customer supplies the processor, so revenue covers only Celestica’s content. A consigned programme therefore adds less revenue than its physical volume suggests and more margin per dollar of revenue. The quarter showed the reverse mix effect as well: adjusted gross margin fell to 11.2% from 11.7% as lower-margin compute grew fastest, while adjusted operating margin rose to 8.5% from 7.6% because overheads fell from 3.0% to 1.8% of revenue.
Who is paying. Three CCS customers took 35%, 17% and 14% of third-quarter revenue, against 30%, 15% and 14% a year earlier. The company does not name them, and its master supply agreements set terms, not volumes. Named programmes are the evidence of who is buying: OpenAI’s rack programme with Broadcom silicon, disclosed in July, AMD’s Helios rack, and a multi-generation relationship on Google’s TPU platform.
The driver is spending on AI data centres. The four largest US cloud spenders have guided to roughly $725–745bn of capital spending for 2026: Amazon about $220bn, Alphabet $195–205bn, Microsoft about $175bn for the calendar year and Meta $135–145bn. None has yet given a 2027 figure. The financing of that spending is starting to show: Meta’s free cash flow was negative in its third quarter, and on 26 October it launched about $35bn of share and mandatory convertible offerings plus an at-the-market programme of up to $60bn. Each generation of AI chips needs faster links between servers, which is moving switch speeds from 800 gigabits to 1.6 terabits, and large cloud companies and AI laboratories increasingly design their own chips and racks and hire a manufacturer to engineer and build them. That is the work Celestica has positioned for; it also means its customers are few and very large.
In designed switches and servers the main rivals are the Taiwanese original design manufacturers: Accton, Quanta, Wiwynn and Hon Hai (Foxconn). In contract manufacturing they are Jabil, Flex and Sanmina, each with its own AI data-centre business; Sanmina added server-rack capacity by buying ZT Systems’ manufacturing operations, and Flex plans to spin off its cloud and power infrastructure segment. Branded network vendors such as Arista and Cisco sell finished switches to the same customers, and Nvidia sells complete racks with its own networking. The company’s filings describe pricing as aggressive and note that customers commonly add a second source once a programme matures.
| $ millions unless stated | 2024 | 2025 | 9M 2026 | 2026 outlook | 2027 outlook |
|---|---|---|---|---|---|
| Revenue | 9,646 | 12,391 | 14,575 | 21,000 | 40,000 |
| Growth | +21% | +28% | +67% | +69% | +90% |
| Adjusted operating margin | 6.5% | 7.5% | 8.3% | 8.4% | 9.0% |
| Adjusted EPS | $3.88 | $6.05 | $8.03 | $11.55 | $21.50 |
| Free cash flow, company definition | 306 | 458 | 413 | 460 | Net use of cash |
| Capital spending, net | 168 | 201 | 744 | ≈1,000 | 1,500–2,000 |
Sources: Q4 2025 results release (28 January 2026); Q3 2026 results release and Form 10-Q (26 October 2026). The 2026 capital-spending figure is the January 2026 plan of about $1bn; nine-month free cash flow includes $701m of receivables sold.
The 2026 revenue outlook has been raised at every report since it was first given, from $16.0bn in October 2025 to $21.0bn. Adjusted EPS has come in above the top of the company’s own guidance range in each of the last four quarters. Adjusted return on invested capital, as the company measures it, was 64.7% in the third quarter, because suppliers and customers’ deposits fund most of the working capital.
Disclosed drivers: 1.6T switching, ramping through 2027, where Celestica designs and builds white-box switches for several hyperscalers and AI laboratories; custom AI racks built around customers’ own accelerators, including the OpenAI programme; design-led share gains through HPS, with AMD’s Helios rack using a Celestica-designed switch; operating leverage, with overheads at 1.8% of revenue; and an ATS recovery led by capital equipment after two flat years. Assumed in our base case: the 2027 outlook, 25% growth in 2028 and a fade to 3% by 2035; no beat.
Risks, in order of weight: concentration, with three customers 66% of the quarter and no guaranteed volumes; a digestion period, as in 2019 and 2023, made likelier by long lead times that encourage over-ordering and by customers now funding capital spending with debt and equity; cash conversion, with earnings well ahead of cash and the gap bridged by stretched payables and receivable sales; second sourcing and pricing, against larger Taiwanese rivals; capital intensity, with $1.5–2.0bn of spending in 2027 raising fixed costs ahead of a possible slowdown; and trade policy, with most production in Asia and Mexico, where the company says customers bear tariffs.
| Source | Score / 5 | Assessment |
|---|---|---|
| Brand | 2 | Known to hyperscaler engineering teams, not to end users; customers buy capability, not the name |
| Network effects | 1 | None |
| Switching costs | 3 | A switch or rack programme is designed in and qualified over many months; changing builder mid-generation is costly, but at each new generation the customer can re-bid |
| Cost advantage | 3 | Low-cost sites in Thailand, Malaysia and Mexico and purchasing scale in switch silicon; the Taiwanese manufacturers are larger |
| Scale | 3 | $21bn of revenue in 2026, against Foxconn’s and Quanta’s far larger server businesses; the largest North American-listed white-box switch maker |
| Intellectual property | 3 | HPS designs for switch hardware and rack-level integration; useful but not exclusive, and customers often co-own the design |
| Distribution | 1 | Direct relationships with a few dozen customers; no channel |
| Regulation and location | 2 | Capacity outside China in Thailand, Malaysia, Mexico and the US is valued under current trade policy; an advantage that depends on policy |
| Data | 1 | None of note |
| Customer loyalty | 3 | Long relationships with the top customers and repeated programme wins across generations; no contracted volumes |
Scores are our judgement. Total 22 out of 50, about 2.2 out of 5: narrow.
Against the competitors. Against Jabil, Flex and Sanmina, Celestica has the higher share of designed product and the higher margin: an 8.5% adjusted operating margin in the third quarter against Jabil’s 5.8% core margin for its last fiscal year and Flex’s 6.7% adjusted margin in its last quarter. Against the Taiwanese design manufacturers it is smaller, but has a stronger position in switches and a footprint outside China and Taiwan; at the Investor Day the company cited independent research naming it the leading provider of 800G-and-faster switching, with 58% of custom Ethernet switch revenue. Against Arista and Cisco it competes where hyperscalers prefer to write their own switch software and buy the hardware unbranded. Against Nvidia it competes only where customers choose their own accelerators and networking rather than Nvidia’s complete systems.
Strongest advantages. Qualification and design ownership. Once a hyperscaler has qualified a Celestica-designed switch for a generation, the switching cost lasts for that generation. Winning the next generation is never guaranteed, but the company has carried its switch programmes from 400G to 800G and into 1.6T.
Biggest threats. Second sourcing as programmes mature, the scale of the Taiwanese rivals in racks, and customers designing more themselves. The rapid growth of Compute & Storage also means more of the business sits where the moat is thinnest: rack assembly around a customer’s own chips.
Direction: stable to expanding. HPS has doubled as a share of revenue since 2023, and new design wins, the OpenAI racks and AMD’s Helios switch, broaden it. The offset is that the fastest-growing work is the least defensible. Our base case reflects that by easing the margin from a 9.2% peak to 8.5% as second sources arrive.
| Concern | Severity | Evidence and why it matters |
|---|---|---|
| Customer concentration | HIGH | Three CCS customers were 35%, 17% and 14% of third-quarter revenue, up from 30%, 15% and 14% a year earlier. The company does not name them; supply agreements set terms, not volumes |
| Cash conversion and receivable sales | HIGH | Nine-month operating cash flow of $1,157m includes $701m of receivables sold to a bank under an uncommitted programme whose limit was raised to $1.0bn from $450m in September; nothing was outstanding at 31 December. Without the sales, nine-month free cash flow would have been about −$288m. The 2026 free cash flow outlook was cut to $460m from $600m |
| Stretched payables | MED-HIGH | Days in payables rose to 74 from 54 a year earlier and 50 at the end of 2025; payables grew $3.2bn in nine months against $2.3bn of inventory. Suppliers are financing part of the growth. If payables returned to 60 days, about $1.4bn of cash would reverse at 2027 revenue levels (our estimate) |
| Capital intensity | MEDIUM | Capital spending goes from $201m in 2025 to about $1bn in 2026 and $1.5–2.0bn in 2027. In January the company said it could fund the expansion from operating cash flow; in August it sold $3.45bn of new shares |
| Dilution | MEDIUM | 11.1m shares were sold at $310 in August, lifting the count to 126.1m from 115.0m a year earlier (+9.7%). The shares now trade below the offering price. A new buyback of up to 2.5% of the public float is planned |
| Cash cost of share awards | MEDIUM | Stock compensation expense was $71m in nine months, but the company paid $374m in cash for withholding taxes on vesting awards, booked in financing, not in free cash flow. A total return swap on 1.25m shares (strike $288.87) offsets part of this cost |
| Insider selling | MEDIUM | $217m of open-market sales by officers and one director in twelve months, all under 10b5-1 plans, $131m of it by the chief executive; one director bought $2.05m in October 2025. Directors and officers owned 1.1% in March |
| Margin mix | LOW | Adjusted gross margin fell to 11.2% from 11.7%; the margin gain came from overhead leverage, not pricing. Compute work grows fastest and earns less per dollar than designed switches |
| Leadership changes | LOW | A new chief financial officer from 1 October (an internal promotion; the predecessor became Group President) and a new head of CCS from July, with the incumbent retiring at year-end |
| Accounting and legal | LOW | No material litigation is disclosed. Tax disputes in Romania (about $7m, paid) and Thailand (about $12m) are small. Adjusted results exclude stock compensation and swap revaluations, both disclosed and reconciled |
Severity is our judgement. Sources: Form 10-Q, notes 3, 9, 15 and 17, liquidity and performance indicators; Q4 2025 and Q3 2026 releases; prospectus supplement (424B5), 6 August 2026; Forms 4; 2026 proxy statement; Forms 8-K of 6 July and 11 September 2026.
Areas reviewed with nothing material to report: revenue recognition (no change in policy; no bill-and-hold disclosed), inventory write-downs ($61.1m in nine months on a $4.5bn balance, 1.4%), debt ($736m of term loans, covenants not at issue), and management language (the outlook statements are specific and dated). Overall red-flag score: 5 out of 10. The two high-severity items are related. A business with two-thirds of its sales to three buyers is funding its fastest growth with supplier credit and receivable sales. That is normal in a boom and becomes a liability in a slowdown, when payables shorten and receivables are harder to sell at the same time as volumes fall. It is a cyclical risk, not an accounting one.
| Strength | Weight | Evidence |
|---|---|---|
| Visibility | HIGH | An October outlook for the whole of 2027, $40.0bn, is unusually specific for a manufacturer, and management says customers’ 2028 forecasts are higher still. The 2026 outlook has been raised at every report |
| Returns on capital | HIGH | Adjusted return on invested capital was 64.7% in the quarter; net invested capital was only $3.2bn against $23bn of annualised revenue |
| Balance sheet | MED-HIGH | Net cash of $3.1bn before the receivables sold ($2.4bn after) funds 2027 capital spending without new shares; the $1.74bn revolver is undrawn |
| Design ownership | MED-HIGH | HPS revenue of about $2.3bn a quarter, up 70%, is work where Celestica owns the design, which is harder to second-source than build-to-print |
| Customer deposits | MEDIUM | Customer cash deposits for inventory rose to $561m from $407m at the start of the year, so customers are funding part of the inventory build |
| Valuation against growth | MEDIUM | The shares trade on a lower multiple of next year’s earnings than Jabil, Flex or Fabrinet, while guiding to grow several times faster (Section 03) |
The balance of the evidence. Overall green-flag score: 6 out of 10. The strongest point in Celestica’s favour is that the 2027 outlook was given at all, a year ahead, at a level 12% above where analysts were, by a management team that has raised its 2026 outlook at every report this year. The weakest point is that the outlook depends on the same three customers whose own spending plans for 2027 have not yet been published.
| Dimension | Score / 5 | Assessment |
|---|---|---|
| CEO track record | 5 | Rob Mionis, chief executive since 2015 and now also chair, led the shift from build-to-print to design-led data-centre hardware; revenue has gone from $7.25bn in 2022 to a $21bn outlook for 2026 |
| CFO credibility | 4 | Mandeep Chawla, chief financial officer since 2017, became Group President on 1 October; Todd Ankenmann, with the company since 2018, is the new chief financial officer. Continuity is good; on his first call as CFO he set a target of positive free cash flow in 2028 but gave no range for 2027’s cash use |
| Guidance accuracy | 5 | Adjusted EPS above the top of the guided range in each of the last four quarters; the 2026 outlook raised at every report since it was first given |
| Transparency | 3 | Detailed working-capital metrics and a clear disclosure of receivable sales; but customers are not named, and the January statement that the expansion would be funded from cash flow was overtaken within seven months |
| Capital allocation | 3 | Capital spending is being raised into strong demand, with customer deposits and high returns on invested capital. The $3.45bn share sale at $310 was well timed in hindsight |
| Acquisitions | 3 | Small, bolt-on purchases in recent years; no large deal risk |
| Buybacks | 3 | Buybacks resumed only once the equity was raised; a new programme of up to 2.5% of the float is planned, and buying below the offering price is accretive |
| Dilution | 2 | Share count up 9.7% in a year from the offering; share awards add about 1% more |
| Insider ownership and dealing | 2 | Directors and officers owned 1.1% in March, the chief executive about 1.0m shares. Officers sold $217m in twelve months under trading plans; one director bought |
| Compensation | 3 | Chief executive’s total pay $17.4m in 2025, 92% at risk; pay ratio 1,257 to 1. The annual bonus weights revenue 40%, adjusted operating margin 40% and free cash flow 20% on the company’s definition, which receivable sales and supplier credit can lift; performance shares vest on three-year adjusted EPS, adjusted for relative shareholder return. The 2025 bonus paid at 185% of target |
| Board quality | 3 | All nominees other than the chief executive are independent; recent additions include the chief executive of Entegris and a former Rackspace chief executive. The chief executive now also chairs the board, and most non-executive directors hold deferred share units rather than shares |
| Communication | 4 | Specific, dated outlooks; consistent metrics; willing to give a full-year outlook a year ahead |
Composite about 3.3 out of 5. Sources: 2026 proxy statement; Forms 8-K of 6 July and 11 September 2026; Forms 4, 27 October 2025 to 27 October 2026; Q3 2026 release and call.
Do they act like long-term owners? In operating decisions, yes: the company has invested ahead of demand, reinvested the profits of the boom into design capability and capacity, and kept the balance sheet in net cash. In ownership, less so: insiders hold about 1% and have been steady sellers, and only a fifth of the annual bonus depends on free cash flow, measured in a way that receivable sales can flatter. The test of the next year is whether management lets working capital normalise, at the cost of reported cash flow, or keeps stretching payables to show cash generation.
| Window | Event | Upside | Downside | Date · source |
|---|---|---|---|---|
| 28 Oct 2026 | Alphabet and Microsoft results | First 2027 capital-spending signals from two of the largest buyers | A flat or lower 2027 framework | Firm · company dates |
| 28 Oct 2026 | US Federal Reserve rate decision | Lower rates support long-duration valuations | Higher for longer, with the ten-year near 5% | Firm · Fed calendar |
| 29 Oct 2026 | Amazon results | As above | As above | Firm · company |
| Late Oct–Nov 2026 | Meta’s $35bn offering and $60bn at-the-market programme | Funding secured for Meta’s build-out | A sign that the largest spenders need outside capital to keep spending | Likely · Meta, 26 Oct |
| From 2 Nov 2026 | Current buyback expires; a new one, up to 2.5% of the float, filed with the TSX | Repurchases below the $310 offering price | None material | Firm · Q3 release |
| Early Nov 2026 | Sanmina and Fabrinet results | Read-across on AI rack and optics demand | Margin pressure at peers | Likely · peer releases |
| 10 Nov 2026 | Flex Investor Day; spin-off of its data-centre segment | A separately listed data-centre peer sets a valuation marker | A larger, focused competitor | Likely · Flex, 29 Jul |
| ≈18 Nov 2026 | Nvidia results | Data-centre demand into 2027 | Any digestion language | Likely · estimate |
| 31 Dec 2026 | Head of CCS retires, successor in place since July | — | Execution risk in the largest segment | Firm · 8-K, 6 Jul |
| Late Jan 2027 | Q4 2026 results; 2027 outlook reaffirmed or raised | A raise of the $40.0bn; the 2027 capital-spending plan firming | A cut, or free cash flow below $460m | Firm · reporting pattern |
| Q1–Q2 2027 | 1.6T switch and custom-rack ramps at volume | A revenue run-rate above $10bn a quarter | Component or power shortages delaying ramps | Likely · company commentary |
| Late Apr 2027 | Q1 2027 results | Tracking ahead of a $10bn quarterly run-rate | Working capital absorbing more cash | Firm · reporting pattern |
| May 2027 | Annual meeting | — | Say-on-pay vote | Firm · proxy |
| Late Jul 2027 | Q2 2027 results; first 2028 commentary in numbers | A 2028 outlook above consensus | 2028 guided flat | Firm · reporting pattern |
| H2 2027 | Hyperscalers’ 2028 capital-spending plans | Continued growth | Digestion: the bear case | Likely · — |
| Ongoing | US tariff and Section 232 decisions on electronics | Capacity outside China favoured | Costs that customers refuse to absorb | Uncertain · — |
Dated from company announcements where stated; “reporting pattern” means the date follows the company’s usual calendar, and “estimate” is a third-party estimate. The single most informative items are the cloud companies’ first 2027 spending signals this week and the January results, because the base case already assumes the 2027 outlook. No dividend is paid.
BULLGrowth. The company has just told investors what 2027 looks like: $40.0bn, up 90%, built from customer demand, and it says customers’ 2028 forecasts are higher still; its Group President is “very confident right now in 2028”, with visibility into 2029. It has raised its 2026 outlook at every report this year, and the 2027 figure came in 12% above the analysts. The four largest cloud buyers are spending roughly $735bn this year, and 1.6T switching and custom racks are only starting to ramp.
BEARTwo-thirds of revenue comes from three customers, and their 2027 budgets are not yet public. Meta’s free cash flow turned negative last quarter and it is raising up to $95bn of equity to keep building. When the buyers need outside money to keep buying, the suppliers’ growth rate is near its peak. Long lead times make customers order early, and early orders are what get cancelled in a digestion year. And confidence is not a number: after a full Investor Day there is still no 2028 figure.
BULLValuation. At $276.15 the shares trade at 12.8 times the company’s 2027 EPS outlook, below Jabil, Flex and Fabrinet, which grow a fraction as fast. The reverse DCF shows the price needs revenue to fall in 2028. Our base case, which only fades growth, is worth $364.81.
BEARCyclicals always look cheap at the peak. On trailing figures the shares are on 19 times EBITDA, and on free cash flow, which is negative in 2027 even in the base case, there is nothing to put a multiple on. A low multiple of peak earnings is the market doing its job.
BULLBusiness quality. Design-led revenue is 40% of the total and growing 70%. Adjusted return on invested capital is 65%. The company designs switches that customers qualify for a whole generation, and it has won OpenAI and AMD as design customers this year.
BEARThe fastest-growing work is rack assembly around customers’ own chips, where the moat is thinnest and the Taiwanese manufacturers are larger. Adjusted gross margin fell this quarter; the margin gain came from spreading overheads, which reverses when volume does.
BULLRisks and financials. Net cash of $3.1bn, an undrawn $1.74bn revolver and rising customer deposits mean the build-out is funded. The share sale removed the financing risk, receivable sales are an ordinary treasury tool across the industry, and the chief financial officer is targeting positive free cash flow again in 2028.
BEARIn January management said the expansion would be funded from operating cash flow; seven months later it sold $3.45bn of shares. Nine-month operating cash flow would have been $456m without $701m of receivable sales, and payables are at 74 days. In a slowdown all of that reverses at once, and the chief financial officer gave no range for how much cash 2027 will use.
BULLManagement and catalysts. The best operating record among North American manufacturers, consistent beats, specific and early outlooks, and an orderly succession in finance and in CCS. Alphabet, Microsoft and Amazon give their first 2027 signals this week, and a buyback below the $310 offering price is accretive.
BEAROfficers sold $217m of shares in twelve months, the chief executive $131m, while insiders own about 1%, and four-fifths of the bonus depends on revenue and margin. The same three reports can take the multiple lower: the shares fell 36% between 2 and 26 October, before the results, on fear alone.
On the evidence available today the bull has the better of it. The bear’s strongest points, concentration and cash conversion, are real, but they are arguments about how bad a 2028 downturn would be, and at $276.15 the price already assumes one. The bull’s strongest point is the one the bear cannot rebut with evidence: the company has published a 2027 outlook built from customer demand, says it is “very confident” in 2028, and the price implies revenue falling below the outlook the following year. Undervalued is the evidence-based conclusion. Most uncertain: 2028, where management’s confidence comes without a figure and no one outside the three customers knows the answer. What would change it: a price above $305.11, or flat 2027 spending plans from the cloud companies that put the bear path in reach. Verify next: the 2027 capital-spending signals from Alphabet, Microsoft and Amazon on 28–29 October; fourth-quarter free cash flow against the $460m outlook; the receivables-sold balance and payables days at year-end; and whether the 35%, 17% and 14% customer shares move further apart.
Celestica builds the hardware that sits inside AI data centres: the network switches that connect thousands of chips, and racks of servers built around chips its customers design. It started as a factory that built other companies’ products; today it designs many of the products itself.
It is paid to design and build. It buys components, assembles them, tests the finished product and ships it. It keeps about 8 to 9 cents of operating profit from each dollar of sales.
Its revenue is growing as fast as almost any large technology company, because the biggest cloud companies and AI laboratories are spending hundreds of billions of dollars on data centres. It expects $40 billion of sales next year, almost double this year.
Three customers buy about two-thirds of what Celestica makes, and the company does not say who they are. If they slow down at the same time, sales could fall sharply, and the cash the company has effectively borrowed from suppliers would have to be paid back just as sales fell.
Assuming the company hits its $40 billion forecast next year and growth slows after that, we think the shares are worth about $305 to $456 each, most likely $364.81. At $276.15 we call them undervalued: the price only makes sense if sales fall back in 2028. Cheap shares can stay cheap for a long time, and if the AI spending boom does pause, they could fall a long way. The shares fell by about a third in October before these results.
Right: the spending boom lasts into 2028 and beyond, Celestica wins more designs, and next year’s $40 billion becomes a stepping stone rather than a peak.
Wrong: a pause in AI spending; losing or sharing a big customer; cash flow falling short of profits.
The big cloud companies’ spending plans for 2027, this week. Whether the December quarter turns profit into cash without more invoice sales. And Celestica’s January results, which show whether the $40 billion forecast goes up or down.
| Status | What it covers |
|---|---|
| Verified from Celestica filings and releases | Revenue by segment and end market, margins, adjusted EPS, cash flow, capital spending, working-capital metrics, the receivables sales programme, cash, term loans, finance leases, pension obligations, the total return swap, the share count, customer concentration, stock compensation and withholding taxes, inventory write-downs, tax disputes, the August share sale, leadership changes, insider transactions (Forms 4), executive pay and the board (2026 proxy statement), and the 2026 and 2027 outlooks |
| Verified from market data and peer filings | Celestica’s NYSE close and range on 27 October 2026 and close on 26 October; Jabil, Flex, Sanmina and Fabrinet at their 27 October closes (regular-session daily bars, Interactive Brokers); peer share counts, debt, cash and outlooks from their latest filings and results releases |
| Third-party, attributed | Prepared remarks from an automated transcript (Benzinga) and the question-and-answer session from a published transcript (Investing.com) of the 27 October call, whose record ends during the last answer; figures checked against the release and Form 10-Q; TD Cowen’s note as reported by Investing.com; consensus 2027 revenue of $35.71bn before the release (TipRanks); hyperscaler 2026 capital-spending guidance and reporting dates (SimianX tracker); Meta’s offerings and third-quarter free cash flow (Global Banking & Finance Review, Benzinga); the ten-year Treasury yield (Briefs) |
| Our estimate or judgement | Revenue after 2027 and the margin, stock-pay, working-capital and capital-spending paths in every scenario; the fourth-quarter stub; the opening working capital and plant; the payables reversal arithmetic; the scenario weights; the moat, flag and management scores |
| Still open | The identities of the three largest customers; any 2028 figure; how much cash 2027 will use; the size and timing of repurchases under the new buyback |
Method. Unlevered discounted cash flow on a ten-year explicit path for a cash-generative manufacturer, methodology version 1.3. The fourth quarter of 2026 is a stub to 31 December and each later year is discounted at its mid-point; the Gordon terminal value at 3.0% is discounted from the end of 2036; everything is valued at 27 October 2026 from the 30 September 2026 balance sheet. One discount rate (10.0%) and one terminal growth rate are used across all scenarios, and each material risk is priced once. Stock-based compensation is charged as a cost and awards already granted are in the share count; working capital is measured before receivable sales and held against exit run-rate revenue; depreciation runs off plant; receivables sold and the swap notional are deducted as debt. Scenario weights are our judgement, the fair-value band is the corners of the WACC 9–11% by terminal growth 2.5–3.5% grid, and the reverse valuation is run against the market price. The rating changes to fairly valued if the price rises above $305.11 with no change in the evidence, and to overvalued above $456.19; the conclusion is most sensitive to revenue after 2027, the long-run margin and the discount rate, in that order. Ratings are undervalued, fairly valued or overvalued according to where the price sits relative to the band; Sonde does not issue buy, sell or hold recommendations or price targets. The reference price used throughout is $276.15, the NYSE close on 27 October 2026, and every multiple in this report scales directly with it. Ratings are reviewed each quarter on results and immediately on a material contract, a financing event or a share-price move of more than 25%.