CLS
Celestica
Summary
What they do:
Design and manufacture AI servers, 800G/1.6T networking switches, and liquid-cooled rack systems for hyperscalers through an ODM-Direct model — selling custom-engineered hardware directly to cloud providers rather than through intermediaries, capturing higher margins and deeper architectural integration.
Why they matter:
Celestica fills the critical gap between chip and rack. Hyperscalers need someone to integrate NVIDIA GPUs, Broadcom networking ASICs, and HBM memory into production-ready server and switch platforms at massive scale. CLS's CCS segment grew 64% YoY in Q4 2025 and is expected to grow ~50% in 2026. HPS (hardware platform solutions) -- the higher-value, design-led portion of CCS -- hit $1.4B in Q4 alone (38% of total revenue), up 72% YoY, driven by 800G switch programs with multiple hyperscalers. They are one of a handful of companies that can build AI infrastructure at hyperscaler volume.
Recent performance:
FY2025 revenue $12.4B, up 28% YoY; adjusted EPS $6.05, up 56% YoY. Q4 revenue $3.65B, up 44% YoY (above high end of guidance). Adjusted EPS $1.89 in Q4, up 70% YoY, beating guidance range. Non-GAAP operating margin 7.7%, a company record (+90bps YoY). Adjusted gross margin 11.3% (+30bps). Adjusted ROIC 43% (+14pp YoY). Guided FY2026 revenue to $17B (+37% YoY) with adjusted EPS of $8.75 (+45% YoY). Q1 2026 guided: revenue $3.85-4.15B (+51% at midpoint), EPS $1.95-2.15 (+71% at midpoint). Stock ~$400, market cap ~$46B. Reports Q1 2026 on April 27.
Our Verdict
The AI server ODM riding the hyperscaler capex wave with 37% revenue growth guided for 2026 and deepening engineering moat in 800G/1.6T networking — but at ~45x forward P/E the stock prices in flawless execution, and the ODM model inherently limits margin upside.
Structural trends
Structural
60
/ 100
Moat
6/10
ODM-Direct model, 800G/1.6T networking design wins, hyperscaler engineering depth
AI Exp.AI Exposure
High~50% AI
Play Type
EstablishedAI Growth
~40%
Rel. Value
38
FAIRWhen a hyperscaler like Google, Meta, or Microsoft decides to deploy 100,000 AI GPUs, they need someone to build the servers those GPUs go into. Not just assemble them — design the power delivery, thermal management, board layout, and rack-level integration from scratch, then manufacture them at a scale measured in tens of thousands of units per quarter. This is what Celestica does.
Celestica operates in two segments. Connectivity & Cloud Solutions (CCS, ~78% of Q4 revenue) designs and manufactures AI/ML compute servers, 800G and 1.6T networking switches, storage platforms, and liquid-cooled rack systems for hyperscalers. Within CCS, the communications end market (networking switches) grew 79% YoY in Q4, while the enterprise end market (AI/ML compute programs) grew 33% YoY. The company's HPS (Hardware Platform Solutions) business -- its design-led, higher-margin engagement model -- generated $1.4B in Q4 (38% of total revenue, +72% YoY), driven by 800G switch programs. HPS is increasingly the default engagement model at 1.6T and above due to engineering complexity. The remaining CCS revenue is EMS (electronics manufacturing services) -- lower engineering content, cost-plus work. Advanced Technology Solutions (ATS, ~22%) serves industrial, aerospace and defense, and healthtech markets; ATS was flat/slightly down in Q4 and is guided flat to mid-single-digit growth for FY2026.
The company's ODM-Direct model is the strategic differentiator. Rather than selling through intermediaries or licensing designs from chipmakers, Celestica engineers custom platforms in partnership with the hyperscaler, manufactures them in its own facilities, and ships directly. This creates deeper engineering relationships, better margins than contract manufacturing, and visibility into the customer's next-generation roadmap. Management explicitly stated on the Q4 call that customers view Celestica "less as a supply chain partner and more as a technology leader," with technology roadmap alignment now informing capacity investment decisions.
Customer concentration is significant. In Q4, three customers each exceeded 10% of revenue at 36%, 15%, and 12% respectively (63% combined). For FY2025, the same three at 32%, 14%, and 12% (58% combined). Google is the largest -- management named Google explicitly as a "decade-long partnership" and confirmed Celestica is Google's "preferred manufacturing partner" for TPU systems (though not sole-sourced; a second source exists for BCP).
Celestica has 10 active 1.6T programs in its pipeline, with ~5 starting mass production ramps in H2 2026 and the rest in 2027-2028. The company secured a third hyperscaler customer for 1.6T networking (announced Q4 call), building on an existing relationship that progressed from 400G to 800G to 1.6T. Co-packaged optics conversations are increasing but mass adoption is expected at 3.2T, with R&D already underway. Celestica is expanding manufacturing capacity globally with ~$1B in 2026 capex: 700K+ sq ft in Texas (Richardson campus + new Fort Worth site, online 2027), 1M+ sq ft in Thailand (liquid cooling manufacturing, online late 2026/2027), plus retooled lines in Mexico and Japan. New HPS Design Centers are planned in Austin and Taiwan. Maintenance capex is ~$70-80M; the rest is growth capex tied to booked business.
FY2025 revenue was $12.4B with the CCS segment driving the majority of growth. Management raised the 2026 outlook to $17B -- a 37% YoY increase -- calling it their "high-confidence view" with customer forecasts actually running higher than the $17B guide. CCS is expected to grow ~$4.5B in 2026 (up from the ~$3.5B growth implied 90 days prior). Management further signaled CCS could grow ~$7B in 2027, implying CCS revenue approaching ~$17-18B on its own. Free cash flow guided at $500M+ for 2026 despite the $1B capex, funded entirely from operating cash flow with no balance sheet draw needed.
Supply Chain Dependencies
Upstream Suppliers
The Catch
Celestica is a $46B market cap ODM — a contract manufacturer with engineering capabilities. The business model has a structural margin ceiling: gross margins top out at 10–12% because hyperscalers use competitive bidding across multiple ODMs to control pricing. Revenue growth requires sustained hyperscaler capex at current extraordinary levels; if AI capex moderates by even 20%, CLS's growth rate halves. The stock has run 350%+ in a year and trades at 45x forward P/E — pricing in flawless execution in a business with limited recurring revenue and high customer concentration.
If They Win
If hyperscaler AI capex sustains at $300B+ annually through 2028 and Celestica's 1.6T networking and liquid cooling capabilities drive higher-value programs, CLS becomes a $25B revenue company by 2028 with adjusted EPS of $15+. The ODM-Direct model matures into something more like an engineering partnership — where Celestica co-designs the compute architecture, not just assembles it. At that scale, CLS trades at 25–30x forward earnings ($375–450), validated by 3+ years of consistent execution rather than one year of hypergrowth.
Others in Assemble the Server
Not financial advice. All scores generated via AI algorithms using public data.