Celestica

Phase A — Understand the business

Lens 1 · Company Overview

Celestica makes other companies' hardware. It is one of the world's largest electronics manufacturing services (EMS) providers, and over the last five years it has pushed up the value chain into original-design manufacturing (ODM) — co-designing and building the physical guts of AI data centers for the hyperscalers. Incorporated in Ontario, headquartered in Toronto, dual-listed on the NYSE and TSX; it began life as an IBM manufacturing unit (75+ years inside IBM), was carved out by Onex in 1996, and IPO'd in 1998.

Two reportable segments:

• CCS (Connectivity & Cloud Solutions) — the growth engine. Two end markets: Communications (data-center networking switches, optical systems) and Enterprise (servers and storage). Customers are hyperscalers, cloud/AI service providers, digital-natives and IT OEMs. FY2025 CCS revenue $9,188.5M = 74% of total.
• ATS (Advanced Technology Solutions) — the diversified, more-regulated base: Aerospace & Defense, Industrial, HealthTech, Capital Equipment (semicap). FY2025 ATS revenue $3,202.4M = 26% of total, roughly flat YoY.

The strategic crux is the HPS (Hardware Platform Solutions) offering inside CCS — the ODM/JDM business where Celestica co-designs the box and owns IP, not just bends metal. HPS carries a higher margin than traditional EMS and is the entire mix-shift story: HPS = 41% of total revenue in FY2025, up from 29% (2024) and 21% (2023); HPS revenue +81% YoY.

Contract structure is the classic EMS double-edge: master supply agreements frame the relationship but do not guarantee volume or fix price — Celestica bids program-by-program and takes purchase orders. A majority of agreements require the customer to buy unused inventory bought to meet that customer's forecast (a working-capital backstop), but some require Celestica to give contractual price reductions over the term. So it is recurring-ish (sticky programs) but not take-or-pay, and margins are perpetually negotiated down by the same hyperscalers driving the growth.

Named customers: CCS — Amazon, Meta, Google, Ciena, Dell, HPE, IBM; ATS — Applied Materials, Lam Research, Honeywell. ~75% of revenue is produced in Asia (Thailand is the single largest manufacturing geography at 59% of revenue), ~20% in North America.

Lens 2 · Supply Chain

Celestica is the midpoint of the AI-hardware chain — it buys components and sells finished systems. Mapping it end-to-end with named stakeholders:

Upstream (suppliers / inputs Celestica buys):

• Merchant switch silicon — Broadcom is the keystone supplier. Celestica's flagship DS6000-series 1.6T switch is built on Broadcom Tomahawk 6 silicon (102.4 Tbps). This is the single most important upstream dependency: the switch ASIC defines the product.
• Optics / DSPs — co-packaged optics (CPO) and pluggable optical modules; merchant DSP and laser suppliers feed the optical-systems line.
• CPUs/GPUs/accelerators — for the server/rack programs, the customer's chosen compute (NVIDIA GPUs, or the hyperscaler's own silicon like Google TPU) is integrated; often customer-consigned.
• Commodity inputs: PCBs, connectors, power, memory, passives, enclosures/precision machining.
• Component availability is the live chokepoint. Management said Q1 2026 growth is "paced by component availability" and the supply environment is "tighter than 90 days ago" — demand is not the constraint; parts are.

Midstream (Celestica): design (HPS/JDM) → NPI/prototyping → manufacturing & complex assembly (Thailand 59% of revenue, plus Malaysia, China, Mexico, US) → systems integration & test → logistics → after-market (ITAM/ITAD). The US-capacity expansion for Google TPU systems (completion targeted 2027) is a deliberate geographic move closer to a key customer.

Downstream (Celestica's customers → end users): hyperscalers (Amazon, Meta, Google) deploy the gear directly in their own data centers; OEMs (Dell, HPE, Ciena) re-sell to enterprises. End demand = AI training/inference cluster buildout.

Chokepoints / single-source dependencies:

1. Broadcom switch silicon — concentrated merchant-silicon dependency upstream.
2. Three hyperscaler customers — concentrated demand downstream (see Lens 3/13).
3. Thailand — 59% of revenue produced in one country = geographic/geopolitical single point.

Lens 3 · Competitive Advantages (moats)

The honest read: Celestica's moat is real but shallow-to-moderate, and built on relationships + execution rather than structural lock-in.

What protects it:

• Co-design switching costs (HPS/JDM). Once Celestica jointly designs a hyperscaler's custom rack or switch and owns part of the IP, ripping it out mid-program is costly and slow — qualification, yield ramp and roadmap continuity all favor the incumbent. This is the genuine moat, and it deepens as HPS grows (41% of revenue). The market explicitly frames this as "hyperscaler lock-in".
• Scale + global footprint + qualified capacity. Hard to replicate a multi-continent, customer-qualified manufacturing network quickly; capacity itself is a gating resource when demand is supply-constrained.
• Climbing the value chain. Moving from build-to-print EMS toward design-led ODM (DS6000 switch sitting alongside Cisco, Arista, Juniper) is a structural margin-mix improvement, not just a volume story.

Bargaining power — who needs whom more: This is the moat's weak seam. The hyperscalers hold the power. Master agreements don't guarantee volume or price; contracts can require Celestica to give price-downs over time; programs can transfer. Customer concentration is extreme and rising — top-10 = 79% of FY2025 revenue (up from 73%/64%), and three CCS customers were each ≥10% in 2025 at 32%, 14%, 12%. When one buyer is ~one-third of your revenue, your pricing power is structurally capped. Upstream, Broadcom likewise has the whip hand on switch silicon.

Net: the moat is "trusted, qualified, co-design incumbent on sticky multi-year programs" — good enough to defend share through a cycle, not good enough to defend margin against a determined hyperscaler or to survive the loss of the 32% customer unscathed.

Lens 4 · Segments

All segment figures `` (FY) and filings/10-q-2026-q1.md, note 3 (Q1).

Revenue by segment / end market (FY, $M):

Segment income & margin (FY, $M):

The story the numbers tell:

• Communications is the whole growth engine — switch program ramps (800G now, 1.6T next) more than doubled it in two years. This is "the why": back-end AI cluster networking.
• Enterprise dipped −19% in FY2025 on a tech transition in an AI/ML compute program at one hyperscaler — and is re-accelerating hard (Q4 2025 +33% YoY, +59% QoQ) as the next-gen program ramps. This volatility is the customer-concentration risk made visible in the numbers.
• CCS margin is the quality signal: 6.2% → 7.4% → 8.2%, driven by mix (HPS) + operating leverage — i.e. profitability is improving as it scales, the opposite of commodity EMS.
• ATS is the stable, lower-growth, slightly-higher-quality-than-it-was ballast (margin up to 5.3% on A&D mix improvement).

Q1 2026 acceleration: total revenue $4,047.0M (+52.8% YoY); CCS $3,241.0M (+76% YoY, 80% of total) — Communications $2,410.6M (+69%), Enterprise $830.4M (+101%); ATS $806.0M (flat). CCS margin 8.6%, ATS 6.0%. The mix is tilting even harder to CCS.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print — Q1 2026, reported 2026-04-27)

Headline (GAAP, from the filing):

• Revenue $4,047.0M, +52.8% YoY (vs $2,648.6M).
• Gross profit $437.2M, GM 10.8% (vs 10.3%).
• Earnings from operations $272.1M, op margin 6.7% (vs 4.9%).
• Net earnings $212.3M (vs $86.2M, +146%).
• Diluted EPS $1.83 (vs $0.74).

Adjusted (non-GAAP, company-reported):

• Adjusted EPS $2.16 (vs $1.20) — the GAAP-to-adjusted gap is mostly SBC + intangible amortization + the TRS fair-value swing.
• Record ~8.0% adjusted operating margin — "favorable product mix and operating leverage from higher volumes in CCS."
• Revenue beat the high end of guidance.

What drove it: CCS, specifically 800G networking switches and re-accelerating AI/ML compute (Enterprise +101% YoY). Guidance raised at the print: FY2026 to revenue $17.0B (from $16.0B) and adjusted EPS $8.75 (from $8.20). (One later/aggregated web source cites an even higher FY2026 outlook of $19.0B revenue / $10.15 adj EPS — flag as a conflicting figure; I treat the $17.0B/$8.75 Q1-print raise as the primary, most consistently-sourced number and the $19.0B as an unverified upside data point, possibly a subsequent revision.)

Balance-sheet flags — the thing to actually watch:

• Working capital is consuming enormous cash. Q1 used −$529.3M A/R and −$484.9M inventory, offset by +$1,071.4M payables — net operating cash was still +$356.3M, but the gross build is huge. Inventory rose to $2,672.9M (from $2,188.0M), mostly raw materials $2,247.8M. This is the signature of a supply-constrained ramp: pre-buying parts.
• Capex stepping up hard: PP&E purchases $229.5M in Q1 alone (vs $36.7M a year ago) — consistent with the ~$1B/6%-of-revenue FY2026 capex plan.
• Liquidity move: on 2026-04-27 (same day as the 10-Q) Celestica amended its credit facility to expand the Revolver to $1,750M (from $750M) and refinanced/extended its Term A loan to 2031. Read this as pre-funding the working-capital + capex ramp — a growth signal, not distress.
• Customer concentration confirmed in-quarter: three CCS customers = 35% + 15% + 15% = 65% of Q1 revenue.

Market reaction: stock trades ~$382–390 in mid-June 2026, near highs; the AI-infrastructure cohort has rewarded the raise. (Contrast the −6% drop in Jan 2026 on the Google-TPU-shift headline — see Lens 8/10/13: the tape reacts violently to single-customer news.)

Lens 6 · Earnings Calls (sentiment trend)

Transcripts are not on disk (transcripts/ empty); call content is ``.

Tone has been progressively, almost monotonically, more confident across the last several quarters, tracking the AI ramp:

• Recurring management phrases: "durable and cumulative demand," "ramping switch programs," "moving up the value chain," "co-design partner," "record adjusted operating margin," "paced by component availability".
• The tone shift that matters: the constraint narrative flipped from demand to supply. Management now frames growth as gated by parts availability, not order book — the most bullish possible framing for a manufacturer, but also the one that breaks fastest if hyperscaler capex blinks.
• What they lean on now that they didn't two years ago: 1.6T + co-packaged optics, "digital native" rack-scale programs, 10 active 1.6T programs in development for 2H 2026, and US capacity for Google TPU systems.
• What's conspicuously under-discussed: the fragility of 65%-from-three-customers, and the contractual price-down obligations. Bulls and management both skate over concentration until a headline forces it (Jan 2026).

Lens 7 · Comps

Celestica's real peer set is the EMS/ODM cohort (not the robotics index bucket, and not the chipmakers). None of these EMS peers are in the research index, so multiples are pulled live ``.

Reference points: the EMS-peer average trailing P/E is ~51.5 per one source; private EMS businesses change hands at 5–7x EBITDA — a reminder of how far the public AI-levered names have re-rated above the industry's historical private-market value. Dividend yield: CLS pays no dividend; 5-yr average ROE — n/a as a clean figure, but FY2025 net earnings $832.5M on ~$2.22B equity implies a ~37–40% ROE, flattered by a thin, buyback-shrunk equity base.

Read: CLS trades at the top of the EMS cohort (~37x forward vs Flex ~32x), justified by the fastest growth and the richest mix-shift, but it is no longer cheap on any absolute EMS-historical yardstick. The multiple is the bet.

Lens 8 · Stock-Price Catalysts (moves >5%, last ~5 years)

Mostly ``. The pattern is unusually clean and instructive.

• 2019→2025 re-rating: from a single-digit, ~9x-earnings forgotten EMS name to a ~$382 AI-infrastructure darling. The pivot was the AI-datacenter switch ramp + the HPS mix-shift becoming visible in margins.
• Index / sponsor catalysts: Goldman Sachs added CLS to its US Conviction List (December); CIBC raised target $315→$400 "outperform". Analyst conviction + index inclusion fed momentum.
• Earnings beats + guidance raises are the dominant up-catalyst: the Q1 2026 print (beat + raise to $17.0B/$8.75) is the latest.
• The single most revealing down-catalyst — customer concentration: on 2026-01-22, a report that Google may shift TPU-assembly work to other suppliers sent CLS −6.08% to $290.92 and triggered securities-investigation announcements.
• Product catalysts: the DS6000 1.6T switch launch (Broadcom Tomahawk 6) is a watched up-catalyst / "momentum test".

What the tape says the market actually reacts to for this name: (1) hyperscaler capex sentiment, (2) any news about a single key customer (Google/Amazon/Meta), (3) earnings beats-and-raises, (4) switch-generation product milestones. It does not react much to ATS. This is a high-beta, concentration-sensitive AI-capex proxy.

---

Phase C — Judge people & books

Lens 9 · Management

CEO: Robert (Rob) Mionis — President & CEO since August 1, 2015 (~11 years); became Chair as well in 2026.

1. Track record (quantified): Mionis is the architect of one of the cleaner industrial transformations of the cycle. Under his tenure Celestica went from a low-margin, commodity EMS to a design-led AI-infrastructure ODM: revenue $7.96B (2023) → $12.39B (2025), diluted EPS $2.03 → $7.16, adjusted EPS to $6.05, CCS margin 6.2%→8.2%, and a stock re-rating from single digits to ~$382. Prior: CEO of StandardAero (aerospace MRO, led through significant growth), and head of Honeywell Aerospace's $11B integrated supply chain — a supply-chain operator by background, exactly the right DNA for an EMS pivoting to ODM.
2. Tenure & skin in the game: 11 years is long for the industry and signals continuity. Precise insider ownership % — n/a — not cleanly sourced; outstanding share data shows the equity is widely held with modest option/RSU/PSU/DSU awards outstanding (38.6k options, 611.6k RSUs, 782.9k PSUs at target, 444.6k DSUs). He is now also a director at Textron (since 2025).
3. Capital-allocation history: disciplined and shareholder-friendly. Share count is shrinking (diluted 120.3M → 116.2M, 2023→2025) via NCIB buybacks; the 2025 NCIB authorizes up to 5.7M shares, and Celestica repurchased 0.2M at a $271.76 avg through Feb 2026. Net debt is modest (~$181M; see Lens 10). The notable shift: capex is being stepped up to ~$1B / 6% of revenue in 2026 (vs a historical 1.5–2.0% target) to fund the hyperscaler ramp — a deliberate, demand-driven reinvestment. No dividend; capital returns are buyback-only and "opportunistic." ROE/ROIC trend is strongly positive (~37–40% ROE estimate, Lens 7), though flattered by the thin equity base.
4. Red flags: light. Comp appears performance-weighted (PSUs 0–200% of target). The TRS (total return swap) on its own stock is an unusual instrument — it generated large fair-value gains ($253.0M of FVAs in FY2025) that inflate non-GAAP-adjusted-out line items and add equity-price risk; it's disclosed and economically a hedge of SBC obligations, but it makes the GAAP-vs-adjusted bridge noisier than peers (Lens 10). No related-party deals of note; Onex is no longer the controlling shareholder (MVS fully converted Aug 2023).
5. Archetype: professional manager / operator-turned-transformer, not founder. For this stage (scaling a capital-intensive ramp with demanding customers) that is the right archetype — execution and supply-chain discipline matter more than visionary founding.

Lens 10 · Forensic Red Flags

Forensic lens. Every figure labeled. The accounting is clean for an EMS, but the business model embeds specific watch-items.

Income statement:

• Non-GAAP bridge is wide and TRS-driven. FY2025 GAAP net earnings $832.5M / diluted EPS $7.16, but adjusted EPS $6.05 — adjustments strip SBC ($69.8M segment-level), intangible amortization, restructuring, and crucially TRS fair-value gains ($253.0M in FY2025, $91.0M in 2024). The TRS gains flatter GAAP gross profit and SG&A; management adjusts them out of non-GAAP. So here, GAAP > adjusted — the opposite of the usual "SBC flatters non-GAAP" pattern. Watch the TRS both ways: it can swing negative ($1 share-price drop = −$1.3M TRS value ).
• Margins are real and improving (CCS 8.2%, operating leverage), not an accounting artifact.

Balance sheet:

• Inventory and receivables are outrunning revenue — by design, but watch it. Q1 2026: inventory +$484.9M to $2,672.9M (raw materials $2,247.8M); A/R consumed $529.3M of cash. Justified by the supply-constrained ramp and partly de-risked by customer deposits ($388.7M) and master-agreement clauses requiring customers to buy unused inventory — but a demand air-pocket would expose this. Inventory write-downs $23.4M in Q1 2026 (vs $16.5M) — rising but small.
• Contract assets $382.5M (up from $296.3M) from over-time revenue recognition on custom work — a judgment-heavy line; reasonable for ODM but inherently estimate-laden.
• Net debt is modest: cash $595.6M vs total borrowings (incl. leases) ~$776.5M at Dec 31, 2025 → net debt ~$181M. Leverage is not a concern; the Revolver was expanded to $1.75B for ramp flexibility. ~80% of cash sits in foreign subsidiaries (repatriation tax friction).

Cash flow: operating cash positive (+$356.3M Q1) despite the working-capital build — quality is fine. Free cash flow guidance is the soft spot: FY2026 FCF outlook only ~$500M against $832M+ net earnings and ~$1B capex — i.e. net earnings are converting to modest FCF because the ramp eats working capital and capex. This is the legitimate "growth costs cash" tension, not fraud.

Revenue recognition: over-time (cost-to-cost) for custom products where the customer must pay for work-to-date — standard ODM treatment, but estimate-sensitive.

Controls: large accelerated filer; auditor attestation present; no material weakness; no restatement; no clawback-triggering error disclosed.

Regulatory findings (required sub-section):

• SEC EDGAR EFTS (LR + AAER): searched 2021-06-20 → 2026-06-20 — zero Litigation Releases and zero AAERs naming Celestica.
• 10-K Item 3 / contingencies (company's own disclosure): management states the ultimate resolution of all currently pending matters will not be material; the only specifics named are an ongoing Romanian income/VAT matter and Thailand tax matters — tax disputes, not fraud or product liability.
• Non-SEC / web check — material finding the EFTS scan could not surface: in February 2026, Pomerantz LLP (and Bleichmar Fonti & Auld, "In re Celestica Inc. Securities Litigation") announced securities-fraud investigations of Celestica and its officers, triggered by the Jan 22, 2026 report that Google may shift TPU-assembly work away from Celestica (stock −6%). These are plaintiff-firm investigations / early-stage class-action activity, not an SEC action or a finding of wrongdoing — but they are a live legal overhang and a flashing sign that customer-concentration disclosure is where the litigation risk lives.
• Net: No accounting/SEC enforcement findings; clean audit. One live securities-litigation overhang tied to customer-concentration headline risk — monitor, don't ignore.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 adjusted EPS)

Built bottom-up from the latest actuals + company guidance. All outputs ``, inputs labeled. Anchor: FY2025 adjusted EPS $6.05; company FY2026 guide revenue $17.0B / adjusted EPS $8.75 (raised at Q1 print).

Base case — accept management's FY2026 guide; decelerate thereafter as 1.6T programs mature and the law of large numbers bites; assume continued modest buyback (−1%/yr share count) and flat-to-slightly-up margin:

• FY2026 adj EPS ≈ $8.75.
• FY2027 adj EPS ≈ $11.4.
• FY2028 adj EPS ≈ $13.7.

Bull case — the higher web-cited FY2026 outlook ($19.0B / $10.15) proves real and 1.6T/CPO over-delivers:

• FY2026 ~$10.15, FY2027 ~$14, FY2028 ~$18.

Bear case — a single large customer (the 32–35% account) cuts or transfers a program, or hyperscaler capex digests in 2027:

• FY2026 still ~$8.0–8.5 (backlog covers near term), but FY2027 flat-to-down to ~$7–8 as Communications decelerates and the working-capital build reverses painfully.

The spread between bull ($14 in FY2027) and bear ($7–8 in FY2027) is ~2x — that fork is entirely about hyperscaler capex durability + customer retention, which is exactly what the ~37x multiple is exposed to.

Per --watchlist rules, no forecast.ts tracked forecast is logged in this loop. If promoted to a position, log: "CLS FY2026 (Dec-2026) non-GAAP adjusted EPS ≥ $8.75, p≈0.62, resolves 2027-02, tags celestica,deep-dive."

Lens 12 · Bull vs Bear

Bull case. Celestica is the cleanest large-cap pick-and-shovel on AI back-end networking — the part of the AI buildout (1.6T Ethernet switching, co-packaged optics, custom racks) that is arguably more supply-constrained and less commoditized than GPUs. The HPS mix-shift (21%→41% of revenue in two years) is a structural margin re-rating, not a one-off; CCS margin has marched 6.2%→8.2% while tripling revenue — operating leverage is proven, not promised. Demand is "durable and cumulative" and growth is paced by component availability, not orders — a manufacturer's dream framing. Management is a proven operator with the right supply-chain DNA, the balance sheet is clean (net debt ~$181M), capital returns shrink the share count, and the credit facility was just expanded to pre-fund the ramp. Backlog + 10 active 1.6T programs for 2H 2026 give multi-quarter visibility. Analysts are Strong Buy, avg PT $444.

Bear case (risks that could permanently impair or de-rate):

1. Customer concentration — the existential one. Three customers = 65% of Q1 2026 revenue; one is 32–35%. The Jan 2026 Google-TPU-shift scare (−6%, securities investigations) is a live preview of what a real program loss/transfer does. EMS contracts don't guarantee volume — a single hyperscaler insourcing or re-sourcing a flagship program would gut a segment and the multiple simultaneously.
2. Margin is rented, not owned. Master agreements can require contractual price-downs; the same hyperscalers driving growth have the bargaining power to compress the very margin expansion the bull case capitalizes. CCS at 8.2% is a good EMS margin but a thin absolute one — a few points of price-down erases a lot of EPS.
3. Multiple + capex-cycle risk. ~37x forward prices in years of sustained hyperscaler capex. If AI-datacenter capex digests in 2027 (the perennial overbuild fear), CLS de-rates and growth decelerates — the classic double-hit on a high-multiple cyclical.

Pre-mortem (it's late 2027, the thesis broke — what happened?): Most likely story — a major hyperscaler insourced or dual-sourced a flagship switch/compute program (or hyperscaler capex paused), Communications revenue rolled over, the $2.7B inventory build reversed into write-downs, FCF stayed weak, and a ~37x multiple compressed to ~20x on a now-decelerating EPS line — a 40–50% drawdown from a single-customer event the market had been told to ignore.

Is the multiple too high? At ~37x forward / ~38x 2026 adj EPS it is full — priced for execution and durability, with little margin for a concentration accident. Not absurd given ~45% EPS growth (PEG <1 on the guide), but the asymmetry has narrowed sharply versus 12 months ago.

Contrarian view (what the market is refusing to see): the bull consensus treats "demand is supply-constrained" as unambiguously positive. The contrarian read is that supply-constrained demand + extreme customer concentration is a fragility, not a strength — it means a single customer's decision (to insource, to dual-source, to pause) moves a third of the revenue, and the market is paying a growth multiple for what is structurally a levered bet on three buyers' capex budgets. The Jan 2026 episode wasn't noise; it was the model showing its seam.

Lens 13 · Devil's Advocate (short-seller)

Skeptical short-seller dismantling the bull case:

• What structurally breaks the way it makes money: Celestica doesn't own the customer or the silicon. It is a qualified middleman between Broadcom (switch ASICs) and three hyperscalers. Either end can squeeze it: Broadcom on component allocation/price, hyperscalers on program price-downs or insourcing (the entire history of hyperscaler hardware is "build it yourself once it's strategic"). The Google-TPU-shift report is the template.
• Where revenue is concentrated / what happens if it shifts: 65% from three CCS customers; 32–35% from one. Lose or halve the top account and an entire segment's growth — and the multiple — evaporate in a quarter. ATS (26%) cannot backfill it.
• Why the moat is weaker than bulls think: "co-design switching costs" are real but time-limited — every program has a next generation, and at each transition the hyperscaler can re-bid, dual-source, or insource. The moat must be re-won every cycle; it is not an annuity.
• Most dangerous competitor bulls underestimate: not another EMS — it's the customer itself (hyperscaler vertical integration) and, secondarily, Foxconn/Hon Hai and Quanta/Wistron, the Taiwanese ODM giants with deeper scale who are also chasing AI-rack share. Flex is the visible public comp; the silent threat is insourcing + the Asian ODMs.
• Worst capital-allocation / accounting optics: the TRS on its own stock generating $253M of FY2025 fair-value gains is an unusual structure that flatters GAAP and complicates the bridge; the $2.7B inventory build is a bet that demand holds; FCF conversion is weak (~$500M guide on $800M+ earnings). None are fraud, but all are places a short pokes.
• Assumptions that must hold for ~$382 / ~37x: hyperscaler AI capex keeps compounding through 2027–28; the 32% customer stays and grows; price-downs stay gentle; 1.6T/CPO ramps on schedule; inventory clears without write-downs. Break any one and the stock is materially lower.
• If growth disappoints 20–30%: FY2027 EPS to ~$7–8 (Lens 11 bear) and multiple compression to ~20–25x → a plausible 40–50% drawdown. The valuation gives back fast because it's a high-multiple cyclical, not a defensive compounder.
• Single scenario that permanently impairs: a flagship hyperscaler insources its switch/compute program (strategically the most likely event in this industry's history). Plausibility: moderate and rising — exactly what the Jan 2026 Google headline gestured at.

Lens 14 · Management Questions (ordered by information value)

1. Of FY2026's revenue, how much sits with your single largest customer, and what is the contractual/qualification switching cost that would slow them insourcing or dual-sourcing your flagship switch and compute programs?
2. The Jan 2026 report on Google potentially shifting TPU-assembly work — what actually changed in that relationship, and how should we size your TPU/Google exposure into 2027 given the US-capacity expansion completing then?
3. What share of CCS programs carry contractual price-down obligations, and what is the typical annual price-down — i.e., how much gross-margin headwind must volume/mix overcome each year?
4. Walk us through the path from ~$800M+ net earnings to only ~$500M FCF guidance: how much is structural working-capital intensity vs the current ramp, and where does steady-state FCF conversion settle?
5. The $2.7B inventory (mostly raw materials) — how much is customer-committed or deposit-backed, and what is your write-down exposure if a major program slips or cancels?
6. How do you defend CCS margin (8%+) as hyperscalers consolidate spend and bargaining power — what structurally stops this from re-commoditizing toward historical EMS margins?
7. On the 1.6T/CPO transition and 10 active 2H-2026 programs: what's the revenue and margin step-up vs 800G, and what's the realistic ramp risk on timing?
8. How concentrated is your switch-silicon supply on Broadcom, and what is your exposure if their allocation or pricing tightens during the 1.6T cycle?
9. Reconcile the $17.0B/$8.75 Q1 guide with the higher $19.0B/$10.15 figure circulating — what's the current authoritative FY2026 outlook and what moved it?
10. ~59% of revenue is produced in Thailand — what is the concentration/geopolitical/tariff contingency if that footprint is disrupted?
11. Capital allocation: with capex stepping to ~$1B (6% of revenue), what ROIC hurdle governs the ramp, and at what point does buyback yield to reinvestment or an acquisition?
12. Explain the TRS on your own shares — why hold it, how do we think about the GAAP-vs-adjusted impact, and what's the downside if the share price falls sharply?
13. ATS has been flat for three years — is it strategic ballast, a future growth vector, or a divestiture candidate, and what would change that?
14. What does a hyperscaler-capex digestion in 2027 do to your revenue and margin, and how much of your book is "cumulative/durable" vs re-bid every generation?
15. As you become Chair and CEO, how is the board thinking about key-person risk and the next layer of operating leadership for a company that has fundamentally changed shape since 2015?

---