Celestica Inc. (CLS)

Executive Summary / Key Takeaways

• The IBM Heritage Pivot Is Paying Off: Celestica's transformation from a commoditized EMS provider into a high-margin AI infrastructure enabler has created a durable competitive moat, with its HPS (Hardware Platform Solutions) business growing 81% in 2025 to $5 billion and expanding segment margins from 6.2% to 8.2% in just two years.

• Hyperscaler Concentration Is a Feature, Not a Bug: While 74% of revenue now comes from CCS (Connectivity and Cloud Solutions) serving a handful of hyperscalers, these relationships generate network effects through co-development and design collaboration that traditional EMS competitors cannot replicate, evidenced by a decade-long Google (GOOGL) TPU partnership and new 1.6T switch wins with three major hyperscalers.

• The CapEx Surge Signals Confidence, Not Concern: Management's decision to increase 2026 capital spending to $1 billion (6% of revenue) from a historical 1.5-2% is based on "booked business" with minimal execution risk, positioning Celestica to capture an estimated $4.5 billion in incremental CCS revenue growth in 2026 and nearly $7 billion in 2027.

• Valuation Multiples Reflect a Structural Re-Rating: Trading at 2.62x sales and 35.97x earnings, Celestica commands a premium to traditional EMS peers but a discount to AI infrastructure plays, reflecting the market's gradual recognition that its ODM model and technology leadership in 800G/1.6T switching create earnings power that traditional contract manufacturers cannot match.

• Two Variables Will Determine the Next 100% Move: The investment thesis hinges on whether Celestica can maintain its technology leadership as the industry transitions from 1.6T to 3.2T switches while managing geopolitical supply chain risks, particularly around U.S.-China tensions and potential tariffs that could disrupt its Thailand-based manufacturing advantage.

Setting the Scene: From Circuit Boards to AI Brains

Celestica Inc., founded in 1996 as a spin-out from IBM (IBM) manufacturing operations in Ontario, Canada, spent its first two decades as a traditional electronics manufacturing services (EMS) provider—building other companies' designs at razor-thin margins. This history matters because it created the manufacturing DNA and global footprint that now underpins its transformation, but it also means many investors still misclassify the company as a commoditized box-builder rather than a technology partner.

The pivotal inflection occurred in late 2015 when Rob Mionis became CEO and initiated a strategic overhaul that began bearing fruit in 2018. Rather than competing on price for low-complexity programs, Celestica deliberately disengaged from margin-dilutive business and ramped investments in design engineering, supply chain orchestration, and technology roadmaps focused on hyperscale data centers. This was a surgical repositioning that shrank the ATS (Advanced Technology Solutions) segment from 42% of revenue in 2023 to 26% in 2025 while CCS exploded from 58% to 74%.

Today, Celestica operates as a hybrid EMS/ODM (Original Design Manufacturer) that makes money through three distinct value layers. First, it provides traditional contract manufacturing for established products. Second, its HPS offering within CCS co-designs custom hardware platforms where Celestica is "solely or jointly involved in the design," earning higher margins for the intellectual property contribution. Third, it captures value through supply chain management and advanced manufacturing capabilities, particularly in complex liquid cooling and high-speed networking assembly.

The industry structure has coalesced around a handful of dominant hyperscalers—Google, Meta (META), Microsoft (MSFT), Amazon (AMZN)—who collectively plan to deploy clusters of up to 1 million accelerators within the next few years, up from today's 200,000-accelerator clusters. This concentration creates both opportunity and risk. These customers have the capital to fund massive infrastructure buildouts, with annual data center CapEx projected to surpass $1 trillion by 2028. However, their purchasing power is immense, and they demand manufacturing partners that can execute at scale while co-innovating on next-generation architectures.

Celestica sits in a sweet spot within this value chain. Unlike pure-play ODMs like Quanta Computer (2382.TW) or Wiwynn (6669.TW) that focus on server assembly, Celestica has built leadership in the networking layer—the switches that connect GPUs in AI clusters. As AI infrastructure bottlenecks shift from compute to networking, this positioning becomes increasingly valuable. The company's 41% cumulative market share across all switch speeds (400G, 800G, 1.6T) as measured by total ports shipped is more than double the next largest competitor's volume, creating a scale advantage that feeds on itself.

Technology, Products, and Strategic Differentiation: The Speed-to-Silicon Advantage

Celestica's core technology advantage is a system-level capability to turn silicon into fully functional systems faster than anyone else. When Broadcom (AVGO) releases new Tomahawk switch chips, Celestica can achieve "complete functional power on the systems" in days, while competitors measure the same achievement in weeks. This speed matters because hyperscalers operate on brutal development cycles where being first-to-market with next-generation networking can mean capturing 12-18 months of premium pricing before commoditization sets in.

The HPS business model is the economic engine that captures this value. Unlike traditional EMS where Celestica simply assembles customer-provided designs, HPS engagements involve co-development where Celestica contributes engineering resources and retains intellectual property rights. This transforms the company from a pass-through manufacturer into a technology partner that can command premium pricing. The financial evidence is stark: HPS revenue grew 81% in 2025 to $5 billion, representing 41% of total revenue, while CCS segment margins expanded from 6.2% to 8.2%. The mix shift toward HPS is directly driving margin expansion, with management noting that HPS margins are accretive to ATS.

Product leadership manifests in tangible wins. The 400G switch introduction in 2020 marked Celestica's entry into high-performance Ethernet. The 2025 launches of 1.6TbE Data Center Switches (DS6000 and DS6001) with Broadcom's Tomahawk 6 silicon, including direct-to-chip liquid cooling on the DS6001, position the company for the next upgrade cycle. Management expects mass production for first 1.6T programs to begin ramping in the latter part of 2026, with 10 active programs in the pipeline.

The liquid cooling capability is particularly significant. As AI clusters push beyond 200,000 accelerators toward million-accelerator deployments, thermal management becomes a critical constraint. Celestica's ability to manufacture complex liquid cooling systems at scale in its Thailand facility—what management calls "very unique" capabilities with high automation—creates a switching cost for customers. Once a hyperscaler qualifies Celestica's liquid cooling for its TPU or GPU clusters, ripping it out would require re-qualifying an entirely new manufacturing process, creating a multi-year revenue lock-in.

The AMD (AMD) "Helios" rack-scale AI platform collaboration announced in March 2026 extends this moat further. By partnering with a leading silicon provider, Celestica positions itself not just as a manufacturer but as a system architect for next-generation AI infrastructure. This moves the company up the value chain from component assembly to platform integration, where margins are higher and competition is scarcer.

Financial Performance & Segment Dynamics: Margin Expansion as Proof of Concept

Celestica's financial results read as a case study in strategic execution. Full-year 2025 revenue reached $12.4 billion, up 28% year-over-year, but the composition reveals the real story. CCS grew 42% to $9.19 billion while ATS grew just 1% to $3.20 billion. More importantly, HPS within CCS grew 81% to $5 billion, representing 41% of total revenue, up from 21% in 2023. This mix shift is the primary driver of margin expansion, with consolidated gross margin increasing from 10.7% to 12.1% and operating margin hitting a record 7.7% in Q4 2025.

The segment margin improvement tells the same story. CCS margins expanded from 6.2% in 2023 to 8.2% in 2025, driven by improved mix and operating leverage. ATS margins improved from 4.6% to 5.3%, primarily due to discontinuing a margin-dilutive program in Aerospace & Defense. Celestica is actively managing its portfolio for profitability, shedding low-margin business while scaling high-value HPS programs. This discipline is reflected in the cash cycle, which improved to 61 days in Q4 2025, down eight days year-over-year, demonstrating working capital efficiency even as revenue scales rapidly.

Cash flow generation supports the investment thesis despite the CapEx surge. 2025 free cash flow was $458 million, up $152 million from 2024, and above the $425 million outlook. Management's confidence in funding the $1 billion 2026 CapEx through operating cash flow is rooted in seven years of consistent quarterly free cash flow generation. The balance sheet is healthy: $596 million in cash, $724 million in gross debt (0.7x leverage), and $1.3 billion in available liquidity. The company can self-fund its growth without diluting shareholders or taking on risky debt.

The quarterly progression throughout 2025 shows accelerating momentum. Q1 started with $2.65 billion revenue and 7.1% operating margin. By Q4, revenue reached $3.65 billion (44% YoY growth) with operating margin at 7.7%. HPS revenue grew from $1 billion in Q1 to $1.4 billion in Q4, consistently representing 38-44% of total revenue. This trajectory validates management's guidance methodology—each quarter's results exceeded the high end of guidance, suggesting conservative forecasting that builds credibility with investors.

R&D spending is scaling with the opportunity. 2025 R&D expense was approximately $120 million, with plans to increase it by at least 50% to $200 million in 2026. This funds the 3.2T switch development and other next-generation programs that will determine market leadership in 2028 and beyond. Unlike traditional EMS companies that treat R&D as overhead, Celestica's ODM model allows it to monetize these investments through HPS engagements where it shares in the design IP.

Outlook, Management Guidance, and Execution Risk

Management's 2026 guidance of $17 billion revenue and $8.75 adjusted EPS implies 37% revenue growth and 44% EPS growth at the midpoint. This is a "high-confidence view" based on "booked business" and customer forecasts that are higher than the $17 billion guided. The key phrase is "booked business," which means the CapEx investments have corresponding revenue commitments, mitigating the risk of building idle capacity.

The CCS segment is expected to grow approximately 50% in 2026, driven by strong demand in data center networking and AI/ML compute programs. Management sees 400G as very resilient, 800G as very strong, and 1.6T ramping in the second half of 2026. This suggests no technology generation is being cannibalized—each new speed tier adds to the total addressable market rather than replacing the previous generation. The result is a layering effect where Celestica can capture revenue from multiple generations simultaneously.

The digital native customer program—a rack-scale custom AI system expected to ramp in early 2027—represents a new revenue stream that could contribute "multiple billions of dollars" in its first year. While management has been conservative about including this in formal guidance, the fact that it's on track with samples expected in mid-2026 and mass production in early 2027 provides upside optionality. The gating factor is silicon availability, not Celestica's execution capability.

Execution risk centers on three factors. First, can Celestica scale its engineering talent fast enough to support 10 active 1.6T programs while simultaneously developing 3.2T? The new HPS Design Centers in Austin and Taiwan are direct responses to this challenge. Second, will the Thailand facility's liquid cooling capabilities maintain their unique status as competitors invest? Management notes that very few competitors can produce these products at scale, but Jabil's (JBL) $500 million U.S. AI data center expansion suggests they are attempting to close the gap. Third, can Celestica maintain its 41% switch market share as the market grows and attracts new entrants? The company's track record of being first with a fully functioning 800-gig switch and repeating that feat with 1.6T suggests technical leadership, but pricing pressure from Asian ODMs remains a constant threat.

Risks and Asymmetries: What Could Break the Thesis

Customer concentration is the most visible risk, with the top hyperscalers representing the majority of CCS revenue. If one major customer were to in-source manufacturing or switch to a competitor, Celestica could face a 20-30% revenue hit. The mitigating factor is that these relationships are joint commitments spanning multiple generations, not single-product contracts. Google's TPU partnership is described as stronger and more integrated than ever, and share is largely awarded on performance.

Geopolitical tensions present a binary risk. The Thailand facility is a strategic asset with unique liquid cooling capabilities and high automation, but any disruption from China-Taiwan tensions could impact component availability. Management acknowledges this is uncontrollable but notes that suppliers recognize Celestica's leverage given its scale. The company's geographic diversification—adding capacity in Texas, Mexico, and Japan—reduces but doesn't eliminate this risk. U.S. tariffs could add cost, but management's experience during previous tariff cycles showed customers did not waver in their commitment to CapEx plans, and costs are typically passed through via surcharges.

The CapEx surge itself creates an asymmetry. If demand materializes as forecast, the $1 billion investment will generate returns through 2027 and 2028 as new capacity comes online. If demand falters—due to AI investment pullbacks, data center power constraints, or macroeconomic slowdown—Celestica could face underutilized capacity and margin pressure. Management insists the buildout is based on firm commitments, not speculative demand. However, the history of tech infrastructure cycles suggests that even firm forecasts can evaporate quickly if end-customer economics shift.

Competitive intensity is rising. Jabil's $500 million U.S. AI expansion, Flex's (FLEX) enhanced rack-scale solutions, and Sanmina's (SANM) defense diversification all target pieces of Celestica's market. The moat is execution-based: very few competitors can produce these products at scale. But if competitors close the execution gap, pricing could compress. Management notes that customers currently prioritize certainty of supply at scale and best-in-class designs over pricing. This dynamic could reverse if supply constraints ease.

Valuation Context: Between EMS and AI Infrastructure

At $281.68 per share, Celestica trades at 2.62x sales, 35.97x earnings, and 70.80x free cash flow. These multiples place it in a valuation twilight zone—premium to traditional EMS peers but discount to pure-play AI infrastructure names. The PEG ratio of 0.7 (calculated using 50% earnings growth) suggests the market hasn't fully priced the growth trajectory, while the EV/EBITDA of 26.37x reflects expectations of sustained high margins.

Comparing to direct EMS competitors reveals the re-rating story. Jabil trades at 0.90x sales with 4.67% operating margin and 35.80x earnings. Flex trades at 0.90x sales with 5.67% operating margin and 27.16x earnings. Sanmina trades at 0.76x sales with 3.69% operating margin. Celestica's 2.62x sales multiple and 8.71% operating margin show the market is paying up for superior profitability and growth, but the gap isn't as wide as the operational differences suggest.

The free cash flow multiple of 70.80x appears extreme, but context matters. 2025 free cash flow of $458 million was suppressed by the $201 million CapEx base; 2026 CapEx of $1 billion will temporarily depress free cash flow further. Management's guidance of at least $500 million free cash flow in 2026 despite the CapEx surge implies underlying cash generation power of $1.5 billion when normalized. On that basis, the free cash flow multiple falls to roughly 23x, much more reasonable for a 37% grower.

The balance sheet strength supports the valuation. Net debt of $128 million is negligible relative to $32.77 billion enterprise value. The 0.7x leverage ratio provides flexibility to weather downturns or fund opportunistic acquisitions. Return on equity of 40.49% demonstrates efficient capital deployment, while return on assets of 10.10% shows strong asset utilization despite the capital-intensive nature of manufacturing.

Conclusion: The Manufacturing Company That Thinks Like a Tech Platform

Celestica has engineered a rare transformation: a 75-year-old manufacturer that now competes on technology leadership rather than cost. The 81% HPS growth, expanding margins, and hyperscaler partnerships demonstrate that the 2015 strategic pivot has created a durable moat in AI infrastructure. The company is no longer a commoditized EMS provider but an essential partner in building the physical layer of AI.

The investment thesis hinges on two variables. First, can Celestica maintain its technology leadership as the industry moves from 1.6T to 3.2T switches and beyond? The 50% increase in R&D spending and new design centers suggest management is betting aggressively that it can. Second, can the company navigate geopolitical and supply chain risks while scaling capacity? The geographic diversification and booked business approach to CapEx provide confidence, but execution risk remains.

The valuation reflects a market still grappling with how to price a manufacturer with software-like economics. At 2.62x sales and 35.97x earnings, Celestica trades at a premium to EMS peers but a discount to AI infrastructure plays—a gap that should narrow as HPS becomes an even larger portion of revenue. The $1 billion CapEx investment is a statement of conviction: management sees demand extending well beyond 2026 and is building capacity to capture it. For investors, the question isn't whether the AI infrastructure buildout is real—it's whether Celestica can maintain its execution edge as the market scales. The financial evidence suggests it can.