Celestica Inc. (NYSE/TSX: CLS) — A Great Ride Priced as a Destination

An independent fundamental analysis · by Claude (Anthropic) Date: June 4, 2026 Price (06-04-2026): ~$425.36 · Diluted shares: ~116M · Market cap: ~$48.9B · Net debt: ~$186M · EV: ~$49B Sector: Technology — Electronic Manufacturing Services (EMS)/ODM · GICS sub-industry: Electronic Manufacturing Services Segments: CCS — Connectivity & Cloud Solutions (incl. HPS) + ATS — Advanced Technology Solutions · FYE: Dec 31 · Reporting currency: USD · CIK: 0001030894

⚡ Claude’s Take

This block is the author’s own subjective opinion and general information only — not investment advice and not a recommendation to buy or sell any security. Do your own research. The analytical body below deliberately takes no position and sets no price target; the opinion is confined to this fenced block.

Verdict: AVOID at ~$425 / HOLD for existing owners / NOT-A-SHORT. Accumulate only on a deep, cycle-driven pullback into the ~$200–280 zone. Conviction: medium.

Tag: “A great ride priced as a destination.”

Celestica is a genuinely better business than it was three years ago — but it is being valued as something it is not. The HPS/JDM mix shift (design content from 21% to 41% of revenue) is real and has structurally lifted CCS margins ~200 bps; the balance sheet is pristine (net-cash-neutral, undrawn revolver); free cash flow is growing; and management has not done anything dumb with capital. That is the bull’s honest core, and it is why I am not short: the revenue ramp is visible and named ($12.4B → $19B guided → a $25.5B FY27 floor), and shorting a beat-and-raise AI-infrastructure name into a still-running capex cycle is a good way to get carried out. But at ~70x trailing adjusted EPS, ~42x forward, ~36x EV/EBITDA, and a ~0.9% free-cash-flow yield, the market is paying an IP-franchise multiple for what the evidence says is the best corner of a structurally bad, no-moat industry — a contract manufacturer with 58% of revenue in three hyperscaler accounts (32% / 14% / 12%), whose “moat” is a program-specific switching cost re-competed at every silicon generation against rivals 10x its size and against its own customers’ in-housing. This is a momentum/cyclical capital-cycle play dressed in compounder’s clothing, not a durable franchise.

The asymmetry is the whole argument. From ~$425, my scenario work puts the bear zone (a 2027–28 hyperscaler-capex digestion plus any single top-3 re-bid) around ~$85–130 (≈ −70%), the base case around ~$300–415, and the bull case (super-cycle persists, HPS proves durable) around ~$540–770 (+27% to +80%). Spot sits at the top of base / bottom of bull — you are paying for the good outcome and absorbing concentration-plus-cycle risk for the privilege. The internet was real in 2000 and the picks-and-shovels hardware layer still de-rated 70–90%; a 42x-forward multiple has no room for a single digestion year. Add that GAAP EPS is flattered by a $253M non-cash gain on a swap referencing CLS’s own stock (GAAP $7.16 > adjusted $6.05), insiders are net-selling into the run (the CFO sold to zero direct holdings; the only open-market buy in 18 months was a single director’s $2.05M), and the CEO is about to combine the Chair and CEO roles — and the risk/reward at this price is simply poor. Bullish trigger: new 1.6T programs won with new hyperscalers (diversifying the concentration) plus disclosure that HPS segment margin is structurally >10%. Bearish trigger: any top-3 customer re-bids or in-sources a flagship program, or a hyperscaler guides 2027 capex down. Great company, wrong price, late in the cycle.

1. Executive Summary

Celestica is a Toronto-based electronics manufacturing services (EMS) company that has, over the last three years, ridden the artificial-intelligence datacenter build-out from a sleepy, low-margin contract manufacturer into one of the best-performing hardware names on the market. Revenue grew from $7.25B (FY2022) to $12.39B (FY2025), GAAP diluted EPS rose from $1.46 to $7.16, adjusted operating margin expanded from ~5.5% to ~7.5%, and the share count fell from 123.6M to 116.2M. The stock has compounded roughly six-fold and trades near $425, a ~$49B enterprise value.

The engine is entirely the CCS (Connectivity & Cloud Solutions) segment, which grew to 74% of revenue (80% in Q1 FY2026) on hyperscaler networking-switch programs (400G→800G→1.6T) and AI compute. Within CCS, Hardware Platform Solutions (HPS) — a design-led, ODM-style offering where Celestica contributes engineering IP before the bill of materials is frozen — has gone from 21% to 41% of total revenue and is the single best explanation for the margin step-up. The ATS (Advanced Technology Solutions) segment (aerospace & defense, industrial, capital equipment, HealthTech) is roughly one-quarter of revenue and essentially flat.

The investment debate is not about the near-term numbers, which are excellent and visible — management has guided FY2026 to ~$19.0B revenue and $10.15 adjusted EPS and put a ~$25.5B floor under FY2027. The debate is about durability and price. On the evidence assembled here:

• Business quality: there is no durable competitive advantage in the Greenwald sense — EMS is a textbook no-moat industry; CLS’s edge is a real but program-specific, generation-resetting switching cost, not scale, IP ownership, or captive demand. Its customers are larger and more capable than it is and are its most dangerous potential competitors (in-house silicon and direct ODM).
• Concentration: three customers are 32% / 14% / 12% of revenue (58% combined); the top ten are 79%. This is the dominant risk and the antithesis of a stable franchise.
• Financials: the margin/ROIC re-rating (adjusted ROIC ~20%→~37%, ROE 40.5%) is partly structural (HPS mix) and partly cyclical/mechanical (a hyperscaler-capex peak, working-capital financing by powerful customers, and a non-cash own-stock swap gain inflating GAAP). Cash conversion is deteriorating (CFO/NI 1.33x→0.79x).
• Industry: AI-datacenter hardware is late-boom/pre-digestion in capital-cycle terms; the assembly layer CLS occupies is precisely the layer that mean-reverted in the 2000–02 telecom/optical bust even though the end-demand was real.
• Capital allocation: above the EMS base rate — self-funded growth, no empire-building M&A, controlled dilution, owner-aligned incentives — but with two flags: buybacks at escalating prices have become an SBC-offset treadmill, and the own-stock total-return swap is pro-cyclical leverage that flatters GAAP.
• Valuation: priced as a durable AI compounder (~42x forward, ~0.9% FCF yield) at a 1.6–1.9x premium to legacy EMS; the embedded expectations require revenue to roughly double again and ~8%+ margin to persist and a premium multiple to hold — a stacked bet with asymmetric downside.

The body that follows takes no position and sets no price target (the single position in this document is the labeled Claude’s Take above). It lays out the evidence so the reader can weigh durability against price.

2. Business Overview

What Celestica does. Celestica is a global EMS provider that designs, manufactures, and provides supply-chain services for complex electronic hardware. In the traditional EMS model, an original equipment manufacturer (OEM) — say, a networking-equipment vendor or a hyperscale cloud operator — owns the product design and outsources the physical assembly, test, and supply-chain management to a contract manufacturer like Celestica. The contract manufacturer buys components (often the customer dictates the key ones, such as the switch ASIC or GPU), assembles printed circuit boards and full systems, tests them, and ships them. Historically this is a low-value-added, low-margin activity: the rent in the value chain sits in the silicon and the software, not the assembly.

Over the last several years Celestica has deliberately pushed up the value chain from pure EMS toward ODM/JDM (original/joint design manufacturing) through its Hardware Platform Solutions (HPS) offering, in which Celestica contributes its own design IP, new-product-introduction (NPI) engineering, signal-integrity and thermal/cooling design, and increasingly software and integration — before the bill of materials is locked. Management states plainly that HPS “typically has a higher margin profile than our traditional EMS businesses” (FY2025 10-K, Item 1 & MD&A). HPS is the strategic crux of the entire thesis: it is the difference between being a commodity assembler and being a design-and-supply partner.

Two reportable segments.

• CCS — Connectivity & Cloud Solutions (FY25: $9,188.5M, 74% of revenue). Serves communications and enterprise/cloud customers — principally hyperscalers building AI datacenters. Products: datacenter networking switches (the 400G→800G→1.6T generational ladder), routers, AI/ML compute hardware, storage, and full server/rack platforms. CCS splits into two end markets: Communications ($7,126.4M FY25, +81% YoY) and Enterprise ($2,062.1M, −19% YoY). HPS lives inside CCS.
• ATS — Advanced Technology Solutions (FY25: $3,202.4M, 26% of revenue, ~flat). Serves Aerospace & Defense, Industrial, Capital Equipment (semiconductor and display capital equipment), and HealthTech. These are lower-volume, higher-mix, more-qualified programs with longer design-in cycles and (in A&D/medical) regulatory barriers — structurally stickier than CCS, but slower-growing and currently a smaller share.

How it makes money. Celestica earns a margin on the value it adds — assembly, test, engineering, and supply-chain management — typically expressed as a low-double-digit gross margin (12.1% GAAP in FY25) and a high-single-digit operating margin (8.4% GAAP, 7.5% adjusted). In pure EMS, the customer often owns the component pricing and Celestica earns a thin conversion margin; in HPS, Celestica captures more value because it owns more of the design and the customer is buying a solution, not just labor. Revenue is program-based, not recurring/contractual annuity revenue: Celestica wins a program (e.g., a specific hyperscaler’s 800G switch), ramps it, ships for the program’s life, and must re-win the next-generation program. Revenue is sticky within a generation (you do not re-qualify a manufacturer mid-program lightly) but is re-competed at each silicon node.

Customer base and concentration. Celestica’s customers are blue-chip but few. In FY2025, three CCS customers each individually exceeded 10% of total revenue — at 32%, 14%, and 12% — and the top ten customers were 79% of revenue (FY2025 10-K, Item 1A & MD&A). These are hyperscalers and large OEMs; the relationships are deep and design-integrated, but the concentration is extreme and rising (top-10 was 64% in FY23, 73% in FY24, 79% in FY25). This is examined in Sections 4 and 8 as the central risk.

Geographic footprint. Celestica manufactures across North America (including Mexico), Asia (Thailand, Malaysia, China, etc.), and Europe, with a deliberate “China-plus-one” diversification that benefits from supply-chain reshoring and US-China decoupling. It employs ~23,800 people. The company is dual-listed (NYSE and TSX, both ticker CLS), reports in USD, and during the analysis window transitioned from a foreign-private-issuer (20-F/6-K) to a U.S. domestic filer (10-K/10-Q) — the first 10-K was filed March 2025 for FY2024.

Revenue-model economics — why HPS pays more, and why it still isn’t a moat. It is worth being precise about where the incremental margin in HPS comes from, because the bull and bear cases both hinge on it. In classic consignment/turnkey EMS, the customer specifies the design and often the key components; Celestica earns a conversion margin on the value it adds (assembly, test, logistics), and because the BOM is largely pass-through, the gross margin on the bundle looks thin (high-single digits) even when the conversion margin on Celestica’s own labor and overhead is reasonable. In HPS/JDM, Celestica owns more of the design — board layout, signal integrity, power and thermal/liquid-cooling, mechanical integration, sometimes firmware — so it (a) captures engineering margin it previously did not earn, and (b) influences the BOM, which lets it capture procurement value and improves mix. The result is a higher blended margin and faster inventory turns (HPS programs turned at ~6x in FY25 vs ~3.9x for the legacy mix). That is a genuine, demonstrable economic improvement. The catch, developed in Section 4, is that none of this design content is proprietary and defensible the way a switch ASIC or a network OS is: it is engineering labor sold to a customer who can carry the learning to a cheaper supplier at the next node. HPS raises the floor on Celestica’s economics; it does not build a wall around them.

Seasonality and contract structure. Revenue is recognized largely over time / on shipment against firm customer orders and rolling forecasts; there is modest seasonality but the dominant variable is program ramp timing, which is lumpy. Celestica carries the working capital of the ramp (inventory and receivables) and is partially financed by supplier payment terms and, in some hyperscaler relationships, customer deposits — a structurally favorable dynamic while volumes grow and a whipsaw risk if they stall (see Section 5).

3. Industry Dynamics

Where EMS sits in the value chain — and why it has historically been a bad business. The datacenter-hardware value chain runs roughly: silicon/IP owners (Nvidia, Broadcom, Marvell, AMD) → system designers / brand OEMs (Arista, Cisco, Dell, or the hyperscaler’s own design team) → EMS/ODM contract manufacturers (Celestica, Foxconn, Jabil, Flex, Quanta, Wistron/Wiwynn, Sanmina) → component and sub-assembly suppliers. The economic rent is overwhelmingly captured at the silicon and IP layer: Nvidia earns ~60%+ operating margins; Broadcom and Arista earn 40–65% gross margins on proprietary silicon and software. The assembly layer earns 2–4% operating margins on single-digit ROIC in normal times because it owns little IP, faces a dozen-plus credible competitors, uses largely off-the-shelf process technology, and sells to sophisticated customers who hold the bargaining power. It is the classic “race to the bottom.”

Run through Greenwald’s barriers-to-entry tests, EMS is a no-moat industry: you cannot count the credible competitors on one hand (Foxconn, Jabil, Flex, Sanmina, Benchmark, Quanta, Wistron/Wiwynn, plus regional players and the hyperscalers’ own in-house ODM teams); process technology is broadly available; and customers are large, sophisticated, and multi-source by policy. Market shares are not stable — they shift program by program. This is the base rate against which Celestica’s recent results must be judged.

The demand wave is real and large. Hyperscaler capital expenditure (Microsoft, Amazon, Google, Meta, plus Oracle, xAI, and the neoclouds) is the demand driver. Industry tracking puts aggregate hyperscaler capex at roughly ~$256B (2024) → ~$443B (2025) → ~$600–725B (2026), with ~75% of the 2026 spend directed at AI infrastructure, and Street estimates exceeding $1T for 2027. AI back-end networking switch spend alone is projected to exceed $100B by 2030, and 1.6T networking enters volume production in 2026 — the specific generational wave Celestica is riding. Celestica has reportedly won a second 1.6T switching program. The top-line growth underpinning CLS’s results is therefore not fabricated; it is backed by named, funded, multi-year programs.

Competitive intensity — Celestica is a mid-tier player, top-tier specialist. By total revenue, Celestica (~$12–19B) is dwarfed by Hon Hai/Foxconn (~$210B+, and Nvidia’s sole GB200 rack assembler) and is comparable to or smaller than Quanta (~$50B), Jabil (~$39B market cap, large EMS), and Flex (~$58B market cap); Wistron/Wiwynn is the surging hyperscaler rack ODM (Wiwynn revenue reportedly +164% in 2025). Celestica’s distinction is not scale — it is a top-tier specialist in hyperscaler networking switches, the highest-complexity corner of the build (800G→1.6T signal integrity, optics, thermal/liquid cooling). That niche is genuinely demanding and currently undersupplied with qualified design partners — but it is not a structural barrier; it is a head start.

The capital-cycle read — the decisive industry lens. AI-datacenter hardware is drawing in extraordinary capital. Capex is running at more than 2x the prior three-year base; the gap between AI-infrastructure capex and AI revenue (~46%) is reportedly wider than the gap that preceded the 2001 telecom bust (~32%); financing is shifting toward debt issuance (>$400B); and capex growth is decelerating sharply on the forward path (one estimate: +51% → +13% → +5% across 2026–2028). In a supply-side framework, extraordinary returns in a fundamentally commoditized industry are precisely the precondition for mean reversion: high returns attract capacity (every EMS/ODM peer is chasing AI servers and networking), and margins normalize. The bottlenecks that do protect incumbents — CoWoS advanced packaging, HBM, optics — protect the chip owners, not the assembly layer. The bear’s historical analogue is exact: the internet was real in 1999–2000, and the picks-and-shovels hardware/optical layer still busted 70–90% in 2000–02 when capex digested.

The content layer is migrating in Celestica’s favor — but so is the competition. A second-order point the bulls correctly emphasize: as AI clusters scale, more of the system value is moving into exactly the areas HPS touches — co-packaged optics (CPO), 800G/1.6T pluggable optics, liquid cooling (direct-to-chip and immersion), and high-power rack/power-shelf design. These are real engineering disciplines with rising dollar content per rack, and Celestica has invested R&D ($46M → $118M over three years) to build the capability. This is why CCS margin expanded even before peak volume leverage. The bear’s rejoinder is equally valid: every credible ODM (Foxconn, Quanta, Wiwynn) and several specialists (e.g., power/cooling players) are investing in the same content, and the standards (Ethernet, OCP, UALink/UEC) are open by design precisely so hyperscalers are never captive to one supplier. Rising content raises the prize; it does not assign the prize durably to Celestica.

ATS end markets — the ballast, sized. ATS (~$3.2B, ~26% of revenue) is the structurally stickier, slower-growing half: Aerospace & Defense (multi-year qualified programs, regulatory barriers, long life cycles — genuinely high switching costs), Industrial (factory automation, energy, including a smart-energy/utility exposure), Capital Equipment (semiconductor and display cap-ex — itself cyclical, tied to the semi capex cycle), and HealthTech (regulated medical devices). These markets are each tens of billions in addressable equipment spend, but Celestica’s share is small and growth has been flat-to-low-single-digit. ATS matters to the thesis as diversification and downside ballast — its margins are improving (4.7%→5.3%) and its qualification barriers are realer than CCS’s — but it is too small and slow to anchor a $49B valuation. The investment case lives and dies on CCS.

Regulatory / structural factors. US-China technology decoupling, tariffs, and supply-chain reshoring are, on balance, a net positive for Celestica’s “China-plus-one” footprint (Mexico, SE Asia, US): customers actively seek non-China assembly, and Celestica’s Western footprint is a competitive asset versus China-centric rivals. Export controls on advanced AI chips are a genuine swing factor — they can constrain end-demand (which AI systems can ship to which geographies) and inject policy volatility into the order book, but they also entrench Western-listed, Western-footprint suppliers. Tariff regimes raise landed costs but are largely passed through in EMS pricing. Customer concentration itself is a quasi-regulatory/idiosyncratic risk: a single hyperscaler’s procurement or in-housing decision is a material event with no regulatory recourse.

Verdict — is EMS structurally a good or bad industry? Structurally bad. It is fragmented, low-margin, low-ROIC, customer-captive, and lacks barriers to entry. Celestica occupies the best corner of that bad industry — high-complexity hyperscaler networking — at a moment when that corner is unusually profitable and unusually crowded with incoming capital. The AI-networking sub-segment is better than commodity EMS today, but the durability of that advantage is program-specific and time-limited, and the segment sits late in a capital cycle. This is a demand-cycle-plus-execution story on a real-but-late capital boom, not a structurally attractive industry that protects its incumbents.

4. Competitive Position

This is the section the entire thesis turns on, because Celestica’s ~$49B valuation implicitly asserts a durable competitive advantage. The evidence does not support that assertion.

Name the moat — or its absence. Walking Greenwald’s taxonomy:

• Supply/cost advantage? No. Celestica has no structural cost edge; Foxconn is ~10x larger with greater scale economies, and Asian ODMs operate at lower cost. CLS does not have proprietary process technology that rivals cannot replicate.
• Demand/captivity (customer captivity)? Partial and shallow. There are real switching costs: qualifying a contract manufacturer for a flagship hyperscaler program is expensive and slow (NPI engineering, signal-integrity validation, supply-chain qualification, multi-quarter ramps), so a customer does not switch mid-generation. But the captivity resets at every silicon generation — the program is re-competed when the next switch ASIC or GPU platform arrives — and it sits on an open Ethernet standard, not a proprietary lock-in. This is a head start with sticky innings, not durable captivity.
• Economies of scale + captivity (the strongest moat type)? No. Celestica lacks the scale leg entirely, and the captivity leg is generation-limited. The combination that produces a genuine moat is absent.

The HPS question — does design content change the verdict? This is the bull’s best argument and deserves a fair hearing. By contributing design IP, NPI engineering, and thermal/optical/CPO integration before the BOM freezes, Celestica (a) captures more value (higher margin) and (b) deepens switching costs, because the customer is now buying co-developed design, not just assembly. HPS has gone from 21% to 41% of revenue and carries a higher margin. That is a real move up the value chain and it does raise the floor. But it does not create a Greenwald moat: the design content is program-specific, the customer can take the design knowledge to a rival at the next node, and the most capable design-and-build competitor of all is the customer itself — hyperscalers increasingly design their own silicon (Google TPU, Amazon Trainium/Inferentia, Microsoft Maia) and can route assembly to Foxconn/Quanta/Wiwynn or build in-house. HPS improves the economics of a no-moat business; it does not convert it into a franchise.

The disconfirming evidence — fails the share-stability test. Greenwald’s most practical moat test is market-share stability. Celestica fails it on multiple counts: (1) revenue is re-competed each generation; (2) the FY2024 Enterprise −19% air-pocket — a single hyperscaler’s AI/ML compute-program transition swinging an entire end market down ~$482M — is a live demonstration of share instability; and (3) there is a secondary-sourced report (unverified in filings — treat as interpretation, not fact) that Google is shifting some AI-server assembly away from Celestica. None of this is consistent with a captive, stable customer base.

Customer concentration — the dominant competitive-position risk. Three customers at 32% / 14% / 12% of revenue (58% combined); top-10 at 79% (FY2025 10-K). In a moat business, the supplier has pricing power; here, the bargaining power sits overwhelmingly with the customer. A single concentrated account at 32% of revenue can, by revising a program, swing total company revenue by high-single to low-double digits. Concentration of this magnitude is the financial signature of the absence of a moat, not its presence: a company with genuine pricing power does not depend on three counterparties for 58% of its sales.

Growth — high on origin, low on durability. Revenue grew $7.25B → $12.39B (FY22→FY25) and Q1 FY2026 was +53% YoY — almost entirely organic and CCS/HPS-driven. M&A has been bolt-on only (NCS Global ~$36.1M in April 2024; smaller additions such as PCI/Atrenne) — not an acquired-growth illusion, which is to the company’s credit. So the growth is high-quality on origin (organic, real demand, real programs). But it is low-quality on durability: it is concentrated in three accounts, re-competed each generation, and cyclically tethered to hyperscaler AI capex with no annuity component. It is an excellent stream to own through the up-cycle and a dangerous one to capitalize at a franchise multiple.

Direct peer contrast — where CLS actually stands. Against the relevant set, Celestica is neither the cheapest nor the largest, which is exactly the uncomfortable middle for a no-moat business: Foxconn/Hon Hai is ~17x its revenue and is Nvidia’s primary GB200 rack integrator — it can underprice on scale whenever it chooses to contest a program; Wistron/Wiwynn is the purpose-built hyperscaler rack ODM and grew revenue a reported ~164% in 2025, i.e., it is taking AI share aggressively; Quanta is several times larger and deeply embedded in AI servers. Celestica’s counter is that it is a networking-switch specialist (not primarily a server-rack assembler), and the highest-complexity switching (1.6T signal integrity, optics) has fewer qualified players than server assembly. That is true and is the kernel of the bull case — but “fewer qualified players in one high-end niche, today” is a description of a temporary capacity-and-capability shortage, which is precisely what high returns cure by attracting investment. The Asian ODMs trade at mid-single-digit-to-low-teens P/Es for the same AI build, a market verdict that the assembly function itself does not deserve a franchise multiple. Celestica’s premium rests on the bet that its specific niche stays scarce; the capital cycle says scarcity at a ~37% ROIC is self-correcting.

Verdict. (a) No durable competitive advantage — a crowded, no-moat commodity industry in which Celestica currently enjoys a real but temporary, program-specific capability edge in high-complexity hyperscaler networking. The recent 37% adjusted ROIC is a boom-phase capital magnet, not a barrier — it is already attracting rival ODM capacity and hyperscaler in-housing. (b) Growth is high-quality in origin (organic, demand-backed) but low-quality in durability (concentrated, generation-re-competed, cyclical). The ATS segment has more genuine qualification/regulatory stickiness but is the smaller, flat segment and cannot anchor the valuation.

5. Financial Quality

The single most important normalization. Celestica holds a Total Return Swap (TRS) referencing its own common shares. Because the stock rose ~6x, the swap was marked up and Celestica booked non-cash gains of $253.0M (FY25) and $91.0M (FY24) directly into cost of sales ($109.3M FY25) and SG&A ($143.7M FY25), inflating GAAP gross and operating profit (FY2025 10-K, MD&A non-GAAP reconciliation). The tell is unmistakable: GAAP EPS ($7.16) is higher than the company’s own adjusted EPS ($6.05) — the reverse of the usual EMS pattern, where add-backs make non-GAAP higher. Throughout this analysis, adjusted EPS $6.05 is the run-rate anchor, and FY2025 GAAP figures should be read as flattered, near a peak, not a floor.

Multi-year spine (reconciled to filings / XBRL):

Revenue composition — the AI engine, quantified. As shown in Section 2, CCS drove all the growth (to 74% of revenue), Communications +81% to $7.13B, and HPS from 21% to 41% of revenue. Q1 FY2026 confirmed continued acceleration: revenue $4,047.0M (+52.8% YoY), CCS 80% of revenue. The growth is accelerating into a larger base, which is unusual and is the bull’s strongest factual point.

Margins — real expansion, but more modest on the clean line. On the company’s preferred adjusted basis, operating margin expanded 5.5% → 6.5% → 7.5% (≈100 bps/yr) — about half the slope of the GAAP series (4.25% → 6.2% → 8.4%), because the GAAP series is partly the TRS gain. Segment margins: CCS 6.2% → 7.4% → 8.2% (mix toward HPS + operating leverage); ATS 4.7% → 4.6% → 5.3% (A&D profitability plus the discontinuation of a margin-dilutive program). The honest read: ~200 bps of clean, adjusted operating-margin expansion over two years, driven mostly by HPS mix (durable) and volume leverage (semi-durable), with a one-time ATS program-pruning benefit. The key undisclosed item — HPS-only segment margin — is the open question that would settle the structural-vs-cyclical debate; the company does not break it out.

Cash flow — growing, but converting worse. FCF grew every year ($203.8M → $305.9M → $458.3M) and the business self-funds; capex is light at ~1.6% of revenue. But cash conversion is deteriorating: CFO/net income fell from 1.33x (FY23) to 0.79x (FY25); FCF/NI fell to 0.55x. The cause is working-capital absorption as the ramp accelerates — accounts receivable consumed $569M of cash in FY25, partly offset by a $909.5M build in accounts payable, with inventory rebuilding +$427.4M to $2,188.0M. This is the classic signature of a fast-ramping EMS: receivables and inventory build ahead of collections, partly financed by powerful customers and suppliers (favorable terms, customer deposits). It is benign while growth is funded, but it means reported earnings increasingly lead cash, and a growth stall would release working capital (a one-time cash windfall) while the income statement collapses. Encouragingly, working-capital efficiency improved even as balances grew: inventory turns rose from 3.9x to 6.1x and the cash cycle fell from ~68 to ~61 days, reflecting genuinely faster-velocity HPS/switch programs.

Working-capital mechanics (the detail behind the conversion squeeze). The cash flows make the dynamic concrete:

Receivables are the cash sink ($569M use in FY25 as the hyperscaler ramp outran collections), partly offset by a $909.5M build in payables — i.e., Celestica is funding a meaningful slice of the AI ramp on supplier terms (and, in places, customer deposits). This is a favorable financing structure while volumes grow and a sharp reversal risk if they stall: a growth pause would release receivables and inventory into cash (a one-time windfall) even as the income statement deteriorates — the EMS “good cash in a bad year” pattern that can mask the severity of a downturn for a quarter or two. It also means that a portion of Celestica’s headline ROIC is, in substance, a subsidy from powerful counterparties, not an earned return on its own capital.

Returns on capital — re-rated, but partly mechanical. Adjusted ROIC rose from ~20% (FY22) to ~37% (FY25); ROE reached 40.5%. This is a dramatic re-rating from the low-teens ROIC that is the EMS base rate. Three caveats temper it: (1) GAAP ROIC (41.6%) is inflated by the TRS gain — use the ~37% adjusted figure; (2) Celestica’s invested-capital denominator is net of accounts payable and operating working-capital financing, so a high ROIC partly reflects powerful customers/suppliers funding its working capital — if those terms normalize, ROIC compresses with no operating deterioration; (3) the capital-cycle warning applies — a ~37% ROIC in a structurally commoditized industry is the precondition for mean reversion, not evidence of a moat.

Balance sheet — pristine. Cash $595.6M against term loans of ~$723.7M (Term A $250M due 2029; Term B $500M due 2031) plus ~$58M finance leases — net debt ≈ $186M, roughly net-cash-neutral (~0.1x adjusted EBITDA). The $750M revolver was undrawn at year-end (and was subsequently expanded to $1.75B with maturity extended to 2031 in April 2026). Liquidity is a non-issue; the company is arguably under-levered given its FCF. SBC is low and stable at ~0.56% of revenue ($69.8M FY25), and buybacks more than offset it — diluted shares fell from 123.6M to 116.2M.

Verdict — do economics improve with scale? Yes, demonstrably, on every operating metric — margin, ROIC, turns, FCF, share count all moved the right way, and the HPS mix shift is a genuine move up the value chain. But the improvement is amplified by three cyclical/mechanical factors: (a) a hyperscaler-capex up-cycle, (b) extreme customer concentration that doubles as a working-capital subsidy from powerful counterparties, and © a non-cash own-stock TRS gain inflating GAAP. The structural core (HPS) supports a permanently higher baseline than the old Celestica; the cyclical overlay means FY2025 GAAP figures are near a peak, not a floor. Durability rests on retaining a handful of hyperscaler design wins — which is precisely the unresolved question.

6. Capital Allocation

Deployment priorities and philosophy. Celestica funds organic growth first (capex, light at ~1.6–2% of revenue historically, guided to ~6% of revenue in FY2026 to support AI capacity — still self-funded by a CFO that scaled from $326M to $660M over FY23–25), then settles SBC, then opportunistically repurchases stock, with minimal debt amortization. There is no dividend (confirmed), and there was no acquisition in FY2025. This is a disciplined, returns-focused posture.

M&A — disciplined bolt-ons, no empire-building. Over the 36-month window the only acquisition of note was NCS Global (~$36.1M, April 2024) — the primary, XBRL-confirmed figure (a $59.6M number circulating in secondary press could not be verified and should not be used). Goodwill is trivial and stable (~$333M). At the top of a capital cycle, with the stock at all-time highs and a currency to spend, management did not go on an acquisition spree — a meaningful positive on the asset-growth screen, which flags capital-cycle tops as the most dangerous time for acquisitive growth.

Buybacks — disciplined timing, but a float treadmill. Repurchases were $35.6M (FY23, 2.6M shares @ ~$13.83), $152.0M (FY24, 3.2M @ ~$47.15), and ~$150.7M for cancellation (FY25, 1.3M @ ~$111.27). Two observations: (1) management bought meaningfully when the stock was cheap (2023–24) and effectively stopped buying for cancellation as the stock ran in 2025 — disciplined, value-aware timing; but (2) the net float reduction stalled in 2025 — shares held roughly flat after Q1 despite the spend, because a large portion of repurchases (~$221M plus ~$225M of cash SBC settlement) effectively funds stock-based-comp delivery rather than shrinking the float. The buyback is part genuine return of capital, part SBC-offset treadmill. The escalating repurchase price ($13.83 → $111.27) also means recent buybacks are far less accretive per dollar — a timing observation, not a red flag.

R&D — funding the value-chain shift. R&D rose from $46.3M (FY22) to $118.2M (FY25), consistent with funding the HPS/JDM design-led shift. This is productive, strategy-aligned spending, not financial engineering.

The own-stock Total Return Swap — flag as risk, not strength. Celestica holds a synthetic levered long on its own shares (notional ~1.25M shares at a ~$288.87 strike, no hedge accounting), sized to its SBC-settlement needs. It threw off ~$345M of cash in 2025 (re-struck twice) that funded the SBC program, and booked the $253M non-cash gain that flattered FY25 GAAP. It is internally coherent (it economically hedges the cash cost of settling appreciating SBC), but it is pro-cyclical own-stock leverage with asymmetric downside: if the stock reverses, the swap becomes a P&L drag exactly when the business is under pressure. Treat it as a monitorable risk and a reason to ignore GAAP in favor of adjusted earnings — not as a capital-allocation virtue.

Management incentives — well-aligned. CEO Rob Mionis’s FY2025 compensation was ~$17.4M (a ~1,257:1 CEO-to-median-worker ratio). The corporate incentive plan (annual) keys on revenue, non-GAAP operating margin, and non-GAAP free cash flow (it paid out at 185% of target, against a 300% cap); the long-term PSUs key on three-year adjusted EPS and relative TSR (a 2023 grant vested at the 200% cap). Crucially, Mionis owns ~1.01M shares (~$412M, ~358x his salary) — substantial skin in the game — and the company has anti-hedging and clawback policies. The metrics are sensible and shareholder-aligned (all on the adjusted basis, which excludes the TRS and SBC). The one critique: incentives are levered to the very up-cycle metrics (revenue, margin, TSR) that are currently peaking, which rewards riding the cycle.

Governance flags. Two negatives to weigh: (1) Rob Mionis is set to become Chair and CEO at the 2026 AGM, combining the roles — a governance step backward (mitigated by Laurette Koellner becoming Lead Independent Director); and (2) audit-committee churn — three changes in twelve months — worth monitoring for any accounting/controls signal, though nothing in the filings indicates a problem.

Insider behavior — net selling into strength. Because Celestica only became a domestic filer in late 2024, the Form 4 record begins December 2024 (prior activity sat in Canada’s SEDI). In that window, insiders were net sellers into the ~6x run: the CFO sold ~$11.1M, taking direct holdings to zero (Feb-2025); the President sold 120,000 shares for ~$36.9M (Feb-2026); the CEO sold ~$10.6M (Feb-2025) but held in Feb-2026 (only exercising/withholding, keeping ~1.0M shares). The sole open-market purchase (code P) in the entire 18-month corpus was a single director (Koellner) buying 6,000 shares at ~$341 (~$2.05M, Oct-2025). Notably, no Form 4 carried a 10b5-1 footnote — all sales were discretionary. Interpretation: routine diversification after a massive run is normal, and the CEO holding is reassuring, but discretionary selling into strength by the CFO (to zero) and President is not a vote of conviction at these prices.

Verdict — has management allocated capital intelligently? Yes — materially better than the EMS/capital-cycle base rate. No empire-building, self-funded growth, near-zero net leverage, productive R&D, controlled dilution, and genuine owner alignment. Two disconfirming flags to watch: forward buybacks risk being a frothy-price SBC mop-up, and the own-stock TRS is pro-cyclical leverage that inflates GAAP and reverses badly in a drawdown. Capital allocation is a strength of the story — but it does not create a moat, and it cannot justify the valuation by itself.

7. Changes and Headwinds — Last Two Years

The transformation (FY2024–FY2026). The dominant change is the AI-driven re-rating already described: revenue +28.5% in FY25 and +53% in Q1 FY26; CCS to 74–80% of revenue; HPS to 41%; adjusted operating margin to ~8%; the stock up ~6x. Specific events on the timeline:

• Filer transition (late 2024): Celestica converted from foreign-private-issuer (20-F/6-K) to U.S. domestic filer (10-K/10-Q), increasing disclosure cadence and transparency — a structural positive for investors.
• Beat-and-raise cadence: consecutive quarterly beats with guidance raises through FY2025 and into FY2026. FY26 revenue guidance was raised from ~$16.0B → ~$17.0B → ultimately ~$19.0B, with adjusted EPS from ~$8.20 → ~$8.75 → $10.15, and a ~$25.5B FY2027 revenue floor introduced on the Q1 FY26 call (April 2026). [GlobeNewswire 2026-04-27; earnings transcripts 2026-04-28.]
• Program dynamics: the FY2024 Enterprise −19% air-pocket (one hyperscaler’s AI/ML compute-program technology transition) and its FY2026 recovery (Enterprise +101% YoY in Q1 FY26) — a vivid illustration of program-level lumpiness in a concentrated book.
• 1.6T design wins: Celestica reportedly secured a second 1.6T networking switching program — the key forward growth driver and a real competitive marker.
• Capital structure: a new ~5% NCIB (normal-course issuer bid, Oct-2025); the revolver doubled from $750M to $1.75B with maturity extended to 2031 (April 2026), pre-funding capacity for the ramp.
• Governance: CEO/Chair role combination set for the 2026 AGM; audit-committee churn; Koellner to Lead Independent Director.
• Bolt-on M&A: NCS Global (~$36.1M, April 2024).

News-flow / sentiment note. The recent-events timeline above is built from primary filings (8-Ks, 10-Qs, the proxy) and dated company releases/transcripts. The qualitative tape is overwhelmingly positive (a “Strong Buy” Street, beat-and-raise momentum), with one notable nuance: the Q1 FY26 beat-and-raise was reportedly initially sold by the market — a possible “good news priced in” exhaustion signal at these levels.

Headwinds — do these changes strengthen or weaken the thesis? The operational changes strengthen the business (real margin step-up, design wins, better disclosure, pre-funded liquidity). The risk changes weaken the investment case at this price: concentration rose (top-10 64%→79%), the governance posture stepped backward (combined CEO/Chair), insiders sold into strength, and the entire story became more levered to a single, late-stage capital cycle. Net: the business got better and the bar got higher. The changes do not weaken the company; they raise the stakes on the price.

8. Risk Analysis

Risk of catastrophic loss / total loss. Low at the company level: Celestica is profitable, cash-generative, net-cash-neutral, and diversified across geographies and (within ATS) end markets. The risk here is not bankruptcy; it is valuation drawdown — a 50–70% equity decline driven by a capex digestion plus multiple de-rating, not insolvency. The asymmetry is in the stock, not the enterprise.

9. Valuation Discussion — Embedded Expectations

No price target, no recommendation. The figures below are scenario outputs and embedded-expectations math.

The multiples. At ~$425 (EV ~$49B), on FY25 adjusted EPS $6.05: trailing P/E ~70x, EV/EBITDA ~36x, EV/Sales ~3.9x, P/B ~22x, and FCF yield ~0.9%. On management’s raised FY26 guide (revenue $19.0B, adjusted EPS $10.15), the forward P/E is ~42x and EV/Sales ~2.6x; on the ~$25.5B FY27 revenue floor, EV/Sales falls toward ~1.9x. The ~0.9% trailing FCF yield — paying ~110x trailing free cash flow for an asset-light but working-capital-hungry contract manufacturer with deteriorating cash conversion — is the cleanest “priced for perfection” tell. The forward multiple compresses quickly if you believe the ramp: 42x FY26 → ~25–28x on a plausible FY27 adjusted EPS (~$14–16) → ~18–22x FY28. The bull case is literally “insane on trailing, reasonable on forward, cheap on out-year — because the growth is that fast”; every clause depends on the capital cycle the industry work dates as late.

Comp set — the premium is the bet.

*Sanmina growth/ROIC step-up is acquisition-driven (ZT Systems), not organic. Source: public market-data aggregators, 2026-06-04; peer margins/ROIC from peer filings, accessed 2026-06-04.

Celestica at ~42x forward trades at a ~1.6–1.9x premium to legacy EMS (JBL/FLEX/SANM ~22–26x). Yet on EV/Sales (3.9x) the market does not price CLS like Arista (~21x EV/Sales, ~64% gross margin, real silicon/OS IP) — it recognizes a ~12%-gross-margin assembler. The Asian ODMs that are CLS’s true functional peers on the AI build (Foxconn, Quanta, Wiwynn) trade at a fraction of CLS’s multiple despite larger AI footprints. So the market is paying EMS-plus (well above JBL/FLEX) but not IP-owner multiples. The EMS-plus premium is the precise, quantified measure of the bet that HPS/JDM design-incumbency is a durable advantage rather than late-cycle inflation that mean-reverts toward 22–26x as growth decelerates.

Embedded expectations — what ~$49B EV requires. To “justify” ~$49B at a generous terminal ~20x P/E (a quality-industrial multiple, rich for EMS), the market needs roughly ~$21 of sustainable adjusted EPS at some out-year — ~3.5x FY25’s $6.05 and ~2x the FY26 guide. Translated to operations, that implies revenue roughly doubling again from $19B toward ~$26–30B by ~FY28–29, AND adjusted operating margin pushing from 8% toward 9–10% and holding, AND a premium multiple persisting. The company’s own $25.5B FY27 floor gets most of the way on revenue — which is why the bulls are not obviously wrong on the top line. The fight is margin durability and multiple persistence, not near-term growth.

Scenario zones (3-year, to ~FY28; ranges, not targets):

Bear: capex digestion + a top-3 re-bid/in-source; CCS rolls over; HPS premium competes away; multiple re-rates toward JBL/FLEX. Base: capex decelerates but stays positive; CLS retains its design wins; margin flat; share count flat-to-down. Bull: super-cycle persists; new 1.6T programs ramp; share gains vs Foxconn/Quanta; margin pushes up on mix + leverage.

Spot (~$425) sits at the top of Base / bottom of Bull. The market is underwriting close to the base-case revenue doubling and a persistent premium multiple, with bull-case optionality — and it is not pricing the bear. The distance to the bear zone (~−70%) is far larger than the distance to the bull zone (~+27–80%), so the risk/reward at this price is asymmetric to the downside: you are paying for the good outcome and carrying concentration-plus-cycle risk with little margin of safety.

What the market is pricing correctly vs incorrectly. Correctly: the visible, named revenue ramp and the real (if cyclical) HPS-driven margin step-up — some of the re-rating is earned. Incorrectly (or as a high-conviction bet): it is paying a franchise multiple for a no-moat, 58%-concentrated, late-cycle contract manufacturer, under-weighting the concentration as a valuation risk and extrapolating a late-stage capital cycle that has no room for a digestion year at 42x forward.

10. Variant Perception

What consensus believes. Overwhelmingly bullish: ~21 analysts, a “Strong Buy” consensus, average price target ~$441 (range ~$360–$550) — the stock has run up to the targets, not below them. The Street narrative is a secular AI-networking winner with multi-year visibility (the $25.5B FY27 floor), expanding margins, and a clean balance sheet — i.e., a quality compounder riding the AI build.

The strongest bull case. HPS/JDM is a genuine design-incumbency barrier in the highest-complexity corner of the build (800G→1.6T signal integrity, optics, thermal/liquid cooling); hyperscalers can qualify only 2–3 design partners per program, so a won 1.6T program is durable for its life. Revenue compounds $19B → $25.5B+ → $30B+ while margin pushes 8% → 9–10% on mix and leverage; on FY27–28 adjusted EPS of ~$16–20, today’s price is ~20–26x out-year, not 70x — cheap for that growth. China-plus-one reshoring and a US listing add structural/sentiment tailwinds versus Asian ODMs. If durable, the stock grows into the multiple.

The strongest bear case. EMS is a structurally bad, no-moat industry; CLS’s edge is program-specific and re-competed each silicon generation against larger, cheaper rivals (Foxconn, Quanta, Wiwynn) and against its customers’ own in-housing. 58% of revenue sits in three hyperscaler accounts late in a debt-financed capital boom whose second derivative is already turning (+51%→+13%→+5%). A 2027–28 digestion air-pocket plus any single-account re-bid stalls revenue; working capital releases cash while the income statement collapses; and the ~42x-forward / ~70x-trailing multiple de-rates toward 12–16x at the same time earnings fall — a double compression to the ~$85–130 zone (~−70%). The own-stock TRS that flattered GAAP becomes a P&L drag. This is the 2000–02 optical/telecom picks-and-shovels bust, re-run.

The 3–5 assumptions that matter most. (1) Hyperscaler capex durability through 2027–28 — the single dominant variable. (2) HPS/JDM margin durability — structural floor or cyclical peak? (the undisclosed HPS-only margin is the key open question). (3) Customer-concentration stability — do the top-3 hold their programs? (4) Terminal multiple — does a premium-to-EMS multiple persist? (5) Share count — does the buyback shrink the float or just offset SBC?

What would falsify each side. Falsifies the bull: a top-3 customer publicly re-bids or in-sources a 1.6T program; a second Enterprise-style air-pocket; adjusted op margin rolls below ~7%; a hyperscaler guides 2027 capex down. Falsifies the bear: CLS wins additional 1.6T programs with new hyperscalers (diversifying the concentration); HPS margin is disclosed and demonstrably structural (>10% segment); FY27 actuals beat the $25.5B floor with margin expansion; the capex super-cycle extends into 2028 with no digestion.

11. Fact vs. Interpretation Table

12. Open Questions

1. What is HPS-only segment margin? Undisclosed. This single number would largely settle the structural-vs-cyclical margin debate.
2. How many qualified 1.6T design-partner slots exist per hyperscaler program? Determines the durability of CLS’s switching-cost advantage.
3. Is the reported “Google volume shift” real and material? Secondary-sourced, unverified in filings; load-bearing to the moat argument if true.
4. Do the top-3 customers’ programs renew at the next silicon node? The concentration whipsaw hinges on this.
5. Does the buyback actually shrink the float going forward, or remain an SBC-offset treadmill?
6. What is the run-rate effective tax rate once valuation-allowance releases lapse? FY25 GAAP tax (~11%) flatters EPS versus the ~19% adjusted rate.
7. Reconcile the two 10-Ks’ ROIC quarterly columns (FY24 GAAP ROIC shown as 34.0% vs 26.1% FY-basis) — likely an annualization difference; immaterial to the conclusion.

13. What Must Be True (Bull and Bear, with falsification tests)

For the BULL (durable AI compounder) to be right:

1. Hyperscaler AI capex must stay on a positive trajectory through 2027–28 (no digestion air-pocket). Falsification: any major hyperscaler guides 2027 capex flat-to-down.
2. HPS/JDM design-incumbency must prove durable — CLS retains and re-wins its flagship 1.6T programs at the next node. Falsification: a top-3 customer re-bids or in-sources a flagship program.
3. Adjusted operating margin must hold ~8%+ (ideally rise toward 9–10%) through a competitive re-bid cycle. Falsification: adjusted op margin rolls below ~7% on pricing pressure.
4. The premium-to-EMS multiple must persist as growth decelerates. Falsification: the multiple compresses toward JBL/FLEX (22–26x) even on in-line results.

For the BEAR (late-cycle no-moat de-rating) to be right:

1. The AI-datacenter capital cycle must digest in 2027–28 (the base rate for a debt-financed boom with a turning second derivative). Falsification: capex super-cycle extends into 2028 with no digestion.
2. The concentration must bite — at least one top-3 account must revise/re-bid/in-source. Falsification: CLS diversifies by winning new hyperscalers, dropping the top-3 below ~50% of revenue.
3. The HPS margin premium must compete away as peers (Foxconn, Quanta, Wiwynn) scale into AI networking. Falsification: HPS segment margin is disclosed and demonstrably structural (>10%).
4. The multiple must de-rate toward EMS norms as the growth narrative fades. Falsification: FY27 beats the $25.5B floor with margin expansion, validating the premium.

The two cases share their pivot: the durability of a handful of hyperscaler design wins through one capital-cycle digestion. That is the entire investment question.

14. Conclusion

Celestica is a genuinely improved business — a contract manufacturer that has pushed up the value chain into design-led HPS/ODM work, lifted margins ~200 bps on a clean basis, generated growing free cash flow, kept a pristine balance sheet, and allocated capital sensibly. None of that is in dispute. What is in dispute is durability and price. The evidence says Celestica occupies the best corner of a structurally bad, no-moat industry, with 58% of revenue in three hyperscaler accounts, late in a capital cycle, while the market prices it as a durable AI-infrastructure compounder at ~42x forward earnings and a ~0.9% free-cash-flow yield. The embedded expectations require a stacked set of conditions — revenue roughly doubling again, ~8%+ margins persisting, and a premium multiple holding — and offer asymmetric downside if the capital cycle digests or the concentration breaks. A real, well-run, accelerating business must be weighed against a price that has already underwritten the good outcome. (The single position taken in this document is the labeled Claude’s Take at the top; this body carries none.)

Appendix A — Diligence Questionnaire

Supplemental. Labels: FACT / INTERPRETATION / ASSUMPTION where material. No price target, no recommendation (the one position in this piece is the labeled Claude’s Take in the body above).

General

What thoughtful questions have other investors asked about this company? The recurring debates: (1) Is the margin/ROIC re-rating structural (HPS design content) or a cyclical AI-capex sugar high? (2) How dangerous is the customer concentration (three accounts = 58% of revenue), and what happens when a flagship program re-bids at the next silicon node? (3) Is HPS a real moat or just a better-paid version of commodity assembly? (4) Where are we in the hyperscaler capex cycle, and what does a 2027–28 digestion do to a 42x-forward stock? (5) Why is GAAP EPS above adjusted EPS (the own-stock TRS), and what does that say about earnings quality? (6) Can a contract manufacturer sustain an 8%+ operating margin, or does it compete away? These map precisely to the body’s competitive-position, financial-quality, valuation, and variant-perception sections.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? INTERPRETATION: a cyclical high. FY25 GAAP earnings are flattered by a $253M non-cash own-stock TRS gain and favorable tax items, and the underlying business is riding a hyperscaler AI-capex peak. Adjusted EPS ($6.05) is the cleaner anchor, but even that reflects peak-cycle volumes and margins.

Are earnings driven primarily by the external environment or internal company actions? Both, but the external environment dominates the level: the AI-capex super-cycle drives volume and mix. Internal actions (the deliberate HPS/JDM shift, working-capital discipline, capital allocation) drive the quality and explain why CLS captured more of the wave than a passive assembler would. INTERPRETATION: internal execution is real and creditable; the macro is the bigger swing factor.

How stable are revenues? Low stability / high lumpiness. Revenue is program-based, not contractual annuity revenue, and 58% sits in three accounts. The FY24 Enterprise −19% air-pocket (one program transition, −$482M) is the clearest evidence of instability.

Outlook for the company’s products and services? FACT: strong near-term — FY26 guided to ~$19.0B revenue / $10.15 adjusted EPS, with a ~$25.5B FY27 revenue floor, on named 800G/1.6T networking programs. INTERPRETATION: visibility is genuine for 1–2 years; durability beyond the current silicon generation is the open question.

How big will this market be — growing, shrinking, domestic or international? Growing fast near-term, internationally: hyperscaler capex ~$443B (2025) → ~$600–725B (2026) → $1T+ (2027 est.); AI back-end switch spend >$100B by 2030; 1.6T volume production in 2026. INTERPRETATION: large and growing, but late-cycle with a decelerating second derivative (+51%→+13%→+5%, 2026–28).

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More. AI-server/networking returns are pulling in the entire peer set (Foxconn, Quanta, Wiwynn, Jabil, Flex, Sanmina) plus hyperscaler in-housing — the classic capital-cycle signature.

How profitable is this business (ROIC, ROE)? FACT: adjusted ROIC ~37%, ROE 40.5% (FY25) — far above the EMS base rate of low-teens. INTERPRETATION: partly structural (HPS mix), partly mechanical (asset-light denominator net of A/P; powerful customers/suppliers finance working capital) and cyclical.

How profitable is the industry — competitors, barriers to entry? Structurally low: 2–4% operating margins and single-digit ROIC in normal times; a dozen-plus credible competitors; no meaningful barriers to entry. CLS sits in the best (networking) corner.

Can the business be easily understood? Yes — it assembles and increasingly co-designs electronic hardware for a few large customers. The accounting wrinkle (own-stock TRS distorting GAAP) requires care, but the operating model is comprehensible.

Can it be undermined by foreign, low-cost labor? Yes — this is the core structural risk. Asian ODMs (Foxconn, Quanta, Wiwynn) operate at greater scale and lower cost. CLS’s defense is design content (HPS) and a China-plus-one footprint, not a labor-cost edge.

Do brands matter? No consumer brand. “Brand” here = reputation as a qualified, reliable design-and-build partner — relevant for winning programs but not a pricing-power moat.

What is the nature of competition? Program-by-program competitive bidding at each silicon generation, on capability, capacity, reliability, footprint, and price. Won programs are sticky within a generation; the contest resets at the next node.

Customer switching costs? Real but shallow/generation-limited: qualification, NPI engineering, and signal-integrity validation make mid-program switching costly, but the customer can move at the next node, and the design sits on an open Ethernet standard.

Barriers to entry? Low for commodity EMS; moderate and temporary for high-complexity 1.6T networking (a limited number of qualified design partners per program) — a head start, not a structural barrier.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? INTERPRETATION: the design-engineering/NPI capability and hyperscaler design-win relationships are intangible and largely unrecognized — but they are also non-transferable and program-specific, so their off-balance-sheet “value” is contingent on re-winning programs.

Off-balance-sheet liabilities? Operating commitments (purchase obligations, leases) typical of EMS; the own-stock TRS is a derivative exposure (notional ~1.25M shares) that is on-balance-sheet but pro-cyclical. No unusual hidden liabilities surfaced.

How conservative is the accounting? Mixed. The non-GAAP discipline is reasonable, but GAAP is flattered (not conservative) by the TRS gain and tax-allowance releases — GAAP EPS exceeds adjusted EPS. Use adjusted figures. Audit-committee churn (3 changes/12 months) is a monitorable, though nothing indicates a problem.

How CapEx-hungry is the business? Historically light (~1.6–2% of revenue) — asset-light EMS pushes intensity into working capital, not PP&E. FY26 capex is guided to ~6% of revenue to fund AI capacity, still self-funded.

Capital Allocation & Management

How much FCF, how is it used, and what is the philosophy? FCF $458.3M (FY25), growing. Priority: self-fund organic growth → settle SBC → opportunistic buyback → minimal debt paydown. No dividend. Philosophy is returns-focused and disciplined.

Significant acquisitions recently? No. Only a bolt-on, NCS Global (~$36.1M, April 2024). No empire-building at the cycle top — a positive on the asset-growth screen.

Buying back shares? Yes — $35.6M / $152.0M / $150.7M (FY23/24/25). Disciplined timing (bought cheap, paused as the stock ran), but the buyback increasingly funds SBC delivery rather than shrinking the float (net shares ~flat in 2025).

Issuing large amounts of new shares to insiders? No — SBC is low (~0.56% of revenue) and more than offset by buybacks; diluted shares fell 123.6M → 116.2M.

Compensation policy of directors and management? CEO comp ~$17.4M (FY25). Annual incentive on revenue + non-GAAP operating margin + non-GAAP FCF (paid 185%, capped 300%); LTI on 3-yr adjusted EPS + relative TSR (vested at 200% cap). Anti-hedging and clawback in place. Sensible and shareholder-aligned, though levered to peaking up-cycle metrics.

Motivations of management? INTERPRETATION: well-aligned by ownership — CEO Mionis holds ~$412M of stock (~358x salary). Governance caveat: he is set to become combined CEO + Chair at the 2026 AGM (a step backward, partly mitigated by a Lead Independent Director). Insiders (CFO, President) net-sold into the run.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? No. CLS is a Canadian-domiciled common share dual-listed on the NYSE and TSX, reporting in USD as a U.S. domestic filer (10-K/10-Q since FY2024). No K-1; ordinary common-stock tax treatment.

Dividend policy? None — no dividend; capital return is via buybacks.

How profitable is the business? Adjusted operating margin ~7.5–8%; adjusted net margin ~5.7%; adjusted ROIC ~37%; ROE 40.5% (FY25) — high now, with the structural-vs-cyclical caveat.

Is net income diverging from cash from operations? YES — a flag. CFO/net income fell from 1.33x (FY23) to 0.79x (FY25) as working capital absorbs the ramp. Net income increasingly leads cash; a growth stall would release working capital (one-time cash) while earnings fall.

Risks & Downside

What factors would cause the stock to decline? A hyperscaler capex digestion (2027–28); any top-3 customer revising, re-bidding, or in-sourcing a flagship program; margin mean-reversion as peers scale; a multiple de-rating toward EMS norms; a TRS reversal becoming a P&L drag. See the risk matrix in Section 8.

Risk of a catastrophic loss? Low at the enterprise level (profitable, net-cash-neutral, positive FCF, diversified footprint). The real risk is a valuation drawdown (50–70%), not insolvency.

Chance of a total loss? Very low. Celestica is a financially sound, cash-generative business; the bear case is a severe equity de-rating, not a zero.

Recent News & Events

Has the business environment changed recently? Yes, materially and favorably on the operating side: the AI-capex super-cycle drove FY25 revenue +28.5% and Q1 FY26 +53%; CCS to 74–80% of revenue; HPS to 41%; FY26 guidance raised to ~$19.0B / $10.15 adjusted EPS; a ~$25.5B FY27 floor introduced; a second 1.6T design win; the revolver doubled to $1.75B (Apr-2026); a new ~5% NCIB (Oct-2025). The risk environment also intensified (concentration rose to 79% top-10; combined CEO/Chair set for 2026 AGM; insiders net-sold).

Any significant acquisitions? Only the NCS Global bolt-on (~$36.1M, April 2024).

Any recent change in accounting policies? The FPI→domestic-filer transition (20-F → 10-K, late 2024) increased disclosure cadence/transparency. The own-stock TRS gains inflated GAAP (a presentation/earnings-quality issue, not a policy change). Audit-committee churn is a monitorable.

Recent changes — new markets, facilities, management? Capacity expansion to support AI programs (capex guided up to ~6% of revenue); continued China-plus-one footprint diversification; CFO unchanged (Chawla); CEO/Chair role combination pending; Koellner to Lead Independent Director.

Note: the recent-events record above is built from primary filings and dated company releases/transcripts.

Appendix B — Source Appendix

Public primary sources prioritized over secondary; every entry carries a URL and date. CIK = 0001030894. Fiscal year = calendar year. All $ in USD.

A. SEC Filings (primary — authoritative)

These are the spine of the analysis.

A.10 Material 8-Ks (domestic era; recent-events timeline)

(The FY25 results condensed-financials exhibit cls-20251231prcondensedfs.htm is the FY25 results 8-K exhibit, filed 2026-02-27: https://www.sec.gov/Archives/edgar/data/0001030894/000103089426000006/cls-20251231prcondensedfs.htm)

A.11 Form 4 insider-transaction filings (load-bearing subset)

A.12 SEC EDGAR XBRL company-facts (machine-readable primary data)

• Title: Celestica Inc. — EDGAR company-facts / company-concept API (US-GAAP XBRL). Publisher: U.S. SEC. URL: https://data.sec.gov/api/xbrl/companyfacts/CIK0001030894.json Accessed: 2026-06-04.
• Tags relied upon: RevenueFromContractWithCustomerExcludingAssessedTax, GrossProfit, OperatingIncomeLoss, NetIncomeLoss, NetCashProvidedByUsedInOperatingActivities, PaymentsToAcquireProductiveAssets, CashAndCashEquivalentsAtCarryingValue, StockholdersEquity, InventoryNet, Assets, ShareBasedCompensation, LongTermDebtAndCapitalLeaseObligations, PaymentsForRepurchaseOfCommonStock, PaymentsToAcquireBusinessesNetOfCashAcquired, ResearchAndDevelopmentExpense, Goodwill.

B. Company Earnings Releases & Investor Materials (primary/company-issued)

C. Industry / Market Data (secondary — sized claims)

D. Trade Press / Financial Media (secondary — qualitative)

Note on verification and flagged claims

The most load-bearing quantitative claims were verified directly against the FY2025 10-K (accession 0001030894-26-000011) and the 2026 DEF 14A (accession 0001030894-26-000018), located by literal-string match in the primary documents. Verified items include: FY25 revenue $12,390.9M; GAAP diluted EPS $7.16; adjusted EPS $6.05; TRS fair-value gain $253.0M; GAAP net earnings $832.5M; customer concentration 32%/14%/12%; CCS $9,188.5M / ATS $3,202.4M; Communications $7,126.4M (+81%); buybacks-for-cancellation $150.7M (WAP $111.27); gross profit $1,494.1M / GAAP operating income $1,040.7M; cash $595.6M / term loans $723.7M; TRS strike $288.87 / re-strike $91.58 / cash proceeds $345.2M; NCIB authorization 5,722,527 shares; shares outstanding 114,967,854 (as of 2026-02-19); CEO Mionis total comp ~$17.4M / pay ratio 1,257:1; Mionis ownership 1,010,091 shares / 358x salary; 2025 corporate performance factor 185%; Koellner open-market buy (Form 4 0001193125-25-258705).

Three substantive open questions / flags survive: (1) the NCS Global price — use $36.1M (net of cash, the primary 10-K/XBRL figure), not the $59.6M secondary headline; (2) the “Google shifting assembly” claim is secondary-only and uncorroborated in filings — kept labeled as interpretation; (3) 2026 guidance figures diverge across reporting dates — the most recent dated guide (Q1-FY26, 2026-04-27) is the current number. Two minor reconciliations (net-debt $186M vs $181M depending on debt definition; the cross-10-K ROIC comparative) are immaterial to the conclusions.