Revenue grew 62% through a documented component shortage and gross margin still went down. That is not pricing power.
Thesis: Celestica is executing a genuine Beat and Raise Machine on named AI networking programs, but the constraint they celebrate sits with TSMC and the silicon vendors, not with them. Gross margin fell 20 bps to 11.5% while customers, per the CEO, 'want more than the supply chain could supply,' and inventory of $3.4B grew 79% YoY against 62% revenue growth under NCNR commitments they cannot unwind. Operating margin expansion came from SG&A leverage on a doubling revenue base. Hold until Q3 proves the inventory converts to revenue at stable gross margin.
Verdict: HOLD — Conviction: MEDIUM
Catalyst: Q3 revenue at the $5.25-5.55B midpoint (69% growth) with 1.6T mass production starting at two hyperscaler customers, plus first OpenAI custom rack deliveries in H2 2026, would show the $3.4B inventory build converting.
Key Risk: Three customers at 32%, 17%, and 14% fund 63% of revenue and the entire capex program. If any one revises CapEx, $3.4B of inventory bought on 52-week lead times and NCNR silicon orders lands on the balance sheet with no demand behind it.
The Tell: CFO Chawla, unprompted: 'we have the capacity in place... However, the constraint is really around material... the demand does exceed the revenue figures that we've been sharing.' The guide is capped by someone else's bottleneck, not by Celestica's.
Friction Level: MODERATE_FRICTION — Both sides accept the numbers. The fight is over what 11.5% gross margin during a 52-week lead time shortage means: mix-driven timing versus structural absence of pricing power at EMS assembly.
2026Q2 · Jul 28, 2026
3 of 5 readings agreed3 HOLD · 2 LONGRead the full call
Beat Every Line, Yet Gross Margin Fell 20 Basis Points
Celestica is executing a genuine Beat and Raise Machine on named AI networking programs, but the constraint they celebrate sits with TSMC and the silicon vendors, not with them. Gross margin fell 20 bps to 11.5% while customers, per the CEO, 'want more than the supply chain could supply,' and inventory of $3.4B grew 79% YoY against 62% revenue growth under NCNR commitments they cannot unwind. Operating margin expansion came from SG&A leverage on a doubling revenue base. Hold until Q3 proves the inventory converts to revenue at stable gross margin.
CFO Chawla, unprompted: 'we have the capacity in place... However, the constraint is really around material... the demand does exceed the revenue figures that we've been sharing.' The guide is capped by someone else's bottleneck, not by Celestica's.
Q2 revenue $4.7B and EPS $2.54 both exceeded the high end of guidance. FY2026 revenue went $19B to $20.5B, EPS $10.15 to $11.30, FCF $500M to $600M. Every line moved up together.
HPS at 41% of revenue and CCS margin at 8.7%, up 40 bps. CFO stated HPS design content 'allows strong pricing because of the value add that we are bringing,' with accelerated HPS growth in 2027.
Binding customer forecasts stretching 12 to 18 months, non-cancellable non-refundable orders on long-lead silicon, and supply agreements on memory and PWBs. Lead times above 52 weeks lock the pipeline.
Gross margin fell 20 bps to 11.5% in a quarter where management says customers want more than supply can deliver. Operating leverage came from SG&A on a doubling revenue base, not from price.
Inventory $3.4B, up $0.7B sequentially and $1.5B YoY, roughly 79% growth against 62% revenue growth, financed against NCNR orders that do not cancel when hyperscaler CapEx normalizes.
$1B CapEx in 2026 with $1.5B as the 2027 placeholder, against $600M FCF guidance. Management is building capacity ahead of demand and states demand exceeds the guide.
+72
CONFIDENT"Every guided line was exceeded and raised. Management answered OpenAI, Helios, CPO timing, and 2027 assumptions with dates and dollar figures. CFO volunteered that demand exceeds the revenue guide. Hedging appeared only on material supply."
Full transcript with segment detail, three named customer concentration percentages, inventory and cash cycle days, capex against FCF, and specific 2027 program timing from both CEO and CFO.
Q3 revenue at the $5.25-5.55B midpoint (69% growth) with 1.6T mass production starting at two hyperscaler customers, plus first OpenAI custom rack deliveries in H2 2026, would show the $3.4B inventory build converting.
Three customers at 32%, 17%, and 14% fund 63% of revenue and the entire capex program. If any one revises CapEx, $3.4B of inventory bought on 52-week lead times and NCNR silicon orders lands on the balance sheet with no demand behind it.
“Second quarter revenue is projected to be between $4.15 billion and $4.45 billion, representing growth of 49% at the midpoint.”
“Revenue in the second quarter was $4.70 billion, up 62% and above the high end of our guidance range as strong demand and solid operational execution led to higher than expected revenue in our ATS segment and both of our CCS end markets.”
Companies tend to guide low: the median company we cover reports +1.7% against its guidance midpoint, so a small beat is the norm. A transcript can drop part of a guided range, so read both quotes.
“Third quarter revenue is projected to be between $5.25 billion-$5.55 billion, representing growth of 69% at the midpoint.”
Changed less than 99% of calls
Wording compared with the Apr 28, 2026 call, ranked among 98 calls reporting on Q2 2026. A measure of how much the story moved, not a signal about the stock.
Figures are extracted from the transcripts and checked in code: every quote appears word for word. Language change is computed from word frequencies, with no model involved. Descriptive only: none of this feeds the verdict.
Beat and raise on every line: $4.7B revenue, $2.54 EPS, 8.2% record operating margin, FY guide raised $1.5B on revenue and $1.15 on EPS in ninety days. Enterprise grew 167% against a 130% outlook and communications 62% against 50%, so they beat aggressive forecasts, not soft ones. Structural drivers are named and dated: 1.6T mass production with two hyperscalers in Q3, 10 active 1.6T programs, OpenAI racks with Broadcom on the Jalapeno accelerator for 2027, and design partnership on AMD Helios across all programs as a multi-billion pipeline. HPS mix shift gives real pricing, CCS margin 8.7% up 40 bps, ATS at a record 6.3%. Incremental margins on the raise run near 12%, cash cycle days improved 19, leverage at 0.5x. The street still prices an assembler.
The constraint they celebrate is one they do not own. TSMC and silicon vendors hold margin and schedule; Celestica sits at EMS assembly with no advanced packaging or foundry exposure. Gross margin fell 20 bps during a shortage they describe as customers wanting more than supply can provide. A constrained player with leverage raises price. This one discounted. Inventory of $3.4B rose 79% YoY against 62% revenue growth, hedged only by binding 12 to 18 month forecasts and NCNR silicon orders that do not cancel when hyperscaler CapEx normalizes. Capex intensity jumped from 1.1% to 5.6% of revenue, $264M against $147M of quarterly FCF, full-year $1B against $600M FCF. Three customers at 32%, 17%, and 14% are 63% of revenue and 81% of CCS, the same cohort funding the buildout. OpenAI racks are consigned, so that revenue never touches the gross margin line. 2027's 65% acceleration rests on samples shipping end of 2026 and CPO mass production in H2 2027. None shipped.
Zero Pricing Power Despite Constraints
Gross margin down 20 bps to 11.5% while management states customers want more than supply can deliver. Capacity-constrained players with leverage raise price; Celestica discounted on mix.
Inventory Divergence
Inventory $3.4B up 79% YoY vs revenue up 62%, against 52-week lead times and NCNR orders. Canceled hyperscaler orders do not cancel component POs.
Hyperscaler Dependency
Three customers at 32%, 17%, and 14% are 63% of revenue, 81% of CCS. No consumer endpoint to absorb a downturn in AI CapEx.
High Utilization as Ceiling
CFO called high utilization intentional productivity management. Factories running near full tilt with $1.5B of 2027 capex planned says capacity ceiling, not achievement.
Backlog Quality
The 2027 acceleration of roughly $13B incremental rests on Helios samples shipping end of 2026, OpenAI racks 2027-2028, and CPO mass production H2 2027. Announcements, not recognized revenue.
Unsustainable Trend Confidence
CEO said 'unprecedented demand visibility' and 'we're not placing a ceiling on the upside' while pre-announcing 2027 acceleration past 65%.
HOLD. Do not chase a 62% grower with falling gross margin and a 79% inventory build. The numbers are real so a short here is shorting execution. Long only after Q3 shows $5.4B revenue with gross margin holding mid-11s and inventory converting rather than compounding. Short trigger is clean: any of the 32/17/14 customers revises CapEx, or inventory passes $4B while growth decelerates. Position size at half until one of those prints.