$CLS KEY READ-THROUGHS FROM CELESTICA Q1 2026 EARNINGS CALL
Celestica’s Q1 2026 earnings call was a high-signal update for the AI infrastructure supply chain. The reported quarter was strong, but the broader market read-through was driven by management’s materially raised 2026 outlook, unusually explicit 2027 confidence, and repeated commentary that demand is now being paced by supply availability rather than end-market weakness. Revenue guidance for 2026 was raised from $17.0B to $19.0B, adjusted EPS guidance was raised from $8.75 to $10.15, CCS revenue growth guidance moved to approximately 70%, and management indicated that 2027 revenue growth should be “significantly more” than the more than $6.5B of growth expected in 2026. The call provided high-conviction positive read-throughs for AI Ethernet switching, merchant switch silicon, co-packaged optics, high-speed optical components, memory, advanced PCBs, power components, liquid cooling, wafer fab equipment, and design-capable ODM/EMS platforms. It also provided high-conviction negative or mixed read-throughs for cloud capex intensity, AI hardware free cash flow conversion, branded networking and server OEM share in hyperscale accounts, and any supplier unable to secure constrained memory, silicon, optical, power, or 40+ layer PCB capacity.
AI NETWORKING, ETHERNET SWITCH SILICON AND NETWORKING OEMS
AI ETHERNET SWITCHING IS ACCELERATING FROM 800G TO 1.6T, VALIDATING MERCHANT SILICON AND HIGH-END WHITE-BOX NETWORKING (READ-THROUGH 1)
Supporting call evidence: Celestica reported Communications revenue growth of 69% in Q1, ahead of its low-60s % outlook, driven by “strong demand and ramping programs for 800G networking switches across our largest hyperscaler customers.” Management guided Q2 Communications revenue to grow approximately 50%, driven by “ongoing hyperscaler ramps in multiple 800G programs” and continued 400G strength. More importantly, management said Celestica expects to begin mass production on 1.6T switch programs with 2 hyperscaler customers in 2H 2026 and later added that it has “10 active programs that will be ramping heavily in 2027.”
Affected companies and impact: Broadcom Inc. (AVGO: US) is positive, high magnitude. Arista Networks, Inc. (ANET: US) is positive but mixed, moderate-to-high magnitude. Cisco Systems, Inc. (CSCO: US) is negative-to-mixed, moderate magnitude. NVIDIA Corporation (NVDA: US) is negative-to-mixed for networking attach, moderate magnitude, but not negative for accelerator demand.
Transmission mechanism: Broadcom is the cleanest beneficiary because Celestica’s 800G, 1.6T, and CPO commentary points to accelerating adoption of high-end Ethernet switching silicon in hyperscale AI networks. The transcript specifically referenced Broadcom’s Tomahawk architecture in the CPO discussion, and the broader 800G/1.6T ramp reinforces demand for merchant switch ASICs. Arista benefits from the validation of Ethernet as an AI fabric, but the positive read-through is partially diluted by the fact that Celestica is directly designing and manufacturing advanced switches for hyperscalers, implying that some hyperscale AI switching spend is flowing through ODM/HPS channels rather than branded systems. Cisco faces the same issue more acutely: the call supports the view that hyperscale AI networking growth is increasingly direct, disaggregated, and design-led, which is less favorable for legacy branded routing and switching economics. NVIDIA’s near-term accelerator demand is not impaired by this call, but the long-duration networking read-through is modestly negative because the call strengthens the case that Ethernet can capture more AI scale-out networking share versus proprietary or semi-proprietary alternatives such as InfiniBand.
Near-term trading catalyst: Q2 and 2H 2026 hyperscaler 800G/1.6T ramp updates should support Ethernet exposed names, particularly Broadcom. Celestica’s statement that 1.6T mass production begins in 2H 2026 provides a concrete catalyst window.
Longer-duration fundamental shift: The larger implication is that Ethernet switching is moving from a conventional data center networking cycle into a core AI fabric cycle, with 800G, 1.6T, CPO, and eventually 3.2T creating a multi-year silicon and systems upgrade path.
AI ACCELERATOR ECOSYSTEMS AND SCALE-UP FABRICS
AMD HELIOS VALIDATES A NON-NVIDIA RACK-SCALE AI ECOSYSTEM AND CREATES A NEW SCALE-UP NETWORKING VECTOR (READ-THROUGH 2)
Supporting call evidence: Management highlighted Celestica’s collaboration with AMD on “the design and manufacturing of a scale-up networking switch for the Helios AI architecture.” Management said development is in progress, initial units are expected by year-end, and the broader Helios opportunity is viewed as a “multi-billion dollar market.” Management also stated that 2027 revenue from the program will likely be paced more by silicon availability than by end-market demand.
Affected companies and impact: Advanced Micro Devices, Inc. (AMD: US) is positive, high magnitude. Celestica Inc. (CLS: Canada) is positive, high magnitude. NVIDIA Corporation (NVDA: US) is negative-to-mixed, moderate magnitude. Broadcom Inc. (AVGO: US) is positive-to-mixed, moderate magnitude, depending on silicon content and scale-up network architecture choices.
Transmission mechanism: AMD benefits because Helios requires a credible rack-scale ecosystem, not just competitive GPUs. Celestica’s involvement in scale-up networking supports AMD’s ability to offer a more complete AI system architecture around Instinct GPUs, narrowing one of NVIDIA’s strongest structural advantages: integrated rack-level compute, networking, and software ecosystems. Celestica benefits because scale-up networking is higher value than commodity assembly and extends its HPS role beyond Ethernet scale-out switching into full AI rack architecture. NVIDIA is not negatively affected on near-term GPU demand, but the call is incrementally negative for the durability of NVIDIA’s proprietary rack architecture advantage if AMD, Celestica, and other ecosystem partners can industrialize an open rack-scale alternative.
Near-term trading catalyst: Initial Helios units by year-end 2026 create a milestone for AMD ecosystem credibility. Supplier commentary around silicon availability will be important because management indicated demand may exceed available silicon.
Longer-duration fundamental shift: The call supports the view that AI accelerator competition is shifting from chip-level benchmarks to rack-level deliverability. If AMD Helios scales, the market will assign more value to companies enabling complete rack-scale systems, not only GPU vendors.
CO-PACKAGED OPTICS, OPTICAL COMPONENTS AND HIGH-SPEED INTERCONNECT
CPO IS MOVING FROM ROADMAP TO PRODUCTION-SCALE HYPERSCALER DEPLOYMENT (READ-THROUGH 3)
Supporting call evidence: Celestica announced a “landmark program award for the design and manufacturing of a 1.6T co-package optics Ethernet switch with a hyperscaler customer.” Management described the win as “not just another switch award” and said it believes the award represents the “first major production scale deployment” of co-packaged optics using Broadcom’s Tomahawk architecture directly integrated into Celestica’s system. Management said production is expected to begin in 2027, specifically in 2H 2027, and added that CPO adoption should become more significant at 3.2T.
Affected companies and impact: Broadcom Inc. (AVGO: US) is positive, high magnitude. Coherent Corp. (COHR: US) is positive, high magnitude. Lumentum Holdings Inc. (LITE: US) is positive, moderate-to-high magnitude. Fabrinet (FN: US) is positive, moderate-to-high magnitude. Zhongji Innolight Co., Ltd. (300308: China) is positive, moderate-to-high magnitude. Eoptolink Technology Inc. (300502: China) is positive, moderate-to-high magnitude. Celestica Inc. (CLS: Canada) is positive, high magnitude.
Transmission mechanism: The CPO win validates the shift of optical interconnect from external pluggable modules toward more deeply integrated switch architectures. Broadcom benefits from Tomahawk ecosystem validation and a potential path to higher-value 1.6T/3.2T silicon platforms. Optical component suppliers benefit because co-packaged optics increases the technical complexity, optical engine value, thermal requirements, test intensity, and manufacturing precision required per switch. Fabrinet benefits through the optical manufacturing and packaging supply chain if CPO-related optical module or engine outsourcing grows. Celestica benefits because CPO design and manufacturing raises its value proposition from manufacturing execution to complex co-design.
Near-term trading catalyst: The 2027 production timeline reduces CPO commercialization uncertainty and should support investor interest in optical suppliers exposed to 1.6T and 3.2T data center networks. The call also cited optical components as constrained, which should support pricing and order visibility.
Longer-duration fundamental shift: CPO can reshape the optics value chain by moving differentiation closer to the switch package and system architecture. This favors suppliers with photonics, packaging, thermal, and high-volume manufacturing capability, while pressuring suppliers overly dependent on less differentiated pluggable cycles.
MEMORY, CUSTOM SILICON AND AI COMPUTE COMPONENTS
MEMORY AND CUSTOM SILICON ARE NOW THE PRIMARY AI COMPUTE BOTTLENECKS, SUPPORTING MEMORY PRICING POWER BUT RAISING SYSTEM COST PRESSURE (READ-THROUGH 4)
Supporting call evidence: Management stated that Celestica is experiencing “more component shortages now than 90 days ago.” The constrained commodities were identified as “custom silicon and memory,” with additional constraints in PCBs, power, and optical components. Management added that “input costs are going up materially, whether it be memory or whether it be silicon” and later clarified that “on compute, it’s all about memory.”
Affected companies and impact: SK hynix Inc. (000660: Korea) is positive, high magnitude. Micron Technology, Inc. (MU: US) is positive, high magnitude. Samsung Electronics Co., Ltd. (005930: Korea) is positive, high magnitude. Broadcom Inc. (AVGO: US) is positive, high magnitude for custom silicon and networking silicon. Marvell Technology, Inc. (MRVL: US) is positive, moderate magnitude for custom silicon read-through. Dell Technologies Inc. (DELL: US), Super Micro Computer, Inc. (SMCI: US), Celestica Inc. (CLS: Canada), and Quanta Computer Inc. (2382: Taiwan) are mixed-to-negative, moderate magnitude, due to BOM inflation and working capital burden. Alphabet Inc. (GOOGL: US), Microsoft Corporation (MSFT: US), Meta Platforms, Inc. (META: US), https://t.co/SpqvHNUxpK, Inc. (AMZN: US), and Oracle Corporation (ORCL: US) are mixed-to-negative near term due to higher AI infrastructure capex per deployed unit.
Transmission mechanism: Memory suppliers benefit directly from allocation conditions, stronger ASPs, and durable AI server demand. The call’s memory commentary is particularly important because the AI/ML compute ramp was partially gated by component constraints in Q1, yet management indicated the demand issue was resolved operationally and the program remains on track. Custom silicon suppliers benefit from the same dynamic: silicon availability, not demand, is now the constraint. The negative offset is that system vendors and hyperscalers absorb rising component costs, higher inventory requirements, and potentially lower gross margin if pricing recovery lags input-cost inflation.
Near-term trading catalyst: Memory earnings, ASP commentary, and HBM/DRAM supply allocation updates should remain positive. Hardware integrators may trade more cautiously if gross margin commentary fails to offset rising memory costs.
Longer-duration fundamental shift: AI infrastructure economics are increasingly controlled by scarce memory and custom silicon capacity. The durable winners are component suppliers with structural bottleneck positions; the weaker economic position sits with assemblers and customers unless they can pass through cost inflation or secure preferential allocation.
ADVANCED PCBS AND HIGH-DENSITY INTERCONNECT
40+ LAYER PCBS ARE AN UNDERAPPRECIATED BOTTLENECK IN AI NETWORKING AND COMPUTE HARDWARE (READ-THROUGH 5)
Supporting call evidence: Management specifically identified “PCBs, the 40 plus layer ones” as a supply challenge. This comment was made alongside custom silicon, memory, power components, and optical components, indicating that advanced board capacity is a meaningful constraint, not a minor procurement issue.
Affected companies and impact: TTM Technologies, Inc. (TTMI: US) is positive, high magnitude. Unimicron Technology Corp. (3037: Taiwan) is positive, high magnitude. Ibiden Co., Ltd. (4062: Japan) is positive, moderate-to-high magnitude. Compeq Manufacturing Co., Ltd. (2313: Taiwan) is positive, moderate-to-high magnitude. Tripod Technology Corp. (3044: Taiwan) is positive, moderate magnitude.
Transmission mechanism: 800G, 1.6T, CPO, scale-up networking, and rack-scale AI compute systems require extremely dense, high-layer-count boards with demanding signal integrity, thermal, and reliability specifications. As switch speeds rise, PCB complexity increases materially. If 40+ layer PCBs are constrained at Celestica, similar constraints are likely affecting other AI hardware programs. Suppliers with high-layer-count capability should see stronger utilization, mix, pricing, and backlog visibility.
Near-term trading catalyst: Any PCB supplier commentary around AI networking, high-layer-count capacity, or price/mix improvement should be viewed as more credible after Celestica’s explicit bottleneck disclosure.
Longer-duration fundamental shift: Advanced PCB capability is becoming a structural differentiator in AI infrastructure. The market may need to value high-end PCB suppliers more like scarce enabling infrastructure rather than generic electronics component vendors.
POWER COMPONENTS, LIQUID COOLING AND DATA CENTER PHYSICAL INFRASTRUCTURE
POWER AND THERMAL MANAGEMENT ARE BECOMING SYSTEM-LEVEL AI BOTTLENECKS, NOT JUST ADJACENT INFRASTRUCTURE DEMAND (READ-THROUGH 6)
Supporting call evidence: Management listed “power components” among constrained supply categories. In the Q&A on Google’s TPU architecture and data center interconnect complexity, management stated that system manufacturing design is becoming more important, especially in “liquid cooling, advanced rack scale infrastructure, thermal management.” The digital-native rack-scale program was also described as a fully integrated rack system with compute and networking that is “highly complex” and “not easy to do at scale.”
Affected companies and impact: Monolithic Power Systems, Inc. (MPWR: US) is positive, high magnitude. Vicor Corporation (VICR: US) is positive, moderate-to-high magnitude. Infineon Technologies AG (IFX: Germany) is positive, moderate magnitude. Delta Electronics, Inc. (2308: Taiwan) is positive, moderate-to-high magnitude. Vertiv Holdings Co. (VRT: US) is positive, high magnitude. Eaton Corporation plc (ETN: Ireland) is positive, moderate-to-high magnitude.
Transmission mechanism: Higher-density AI racks drive sharply higher requirements for power conversion, voltage regulation, power distribution, liquid cooling, thermal controls, and facility-level electrical infrastructure. The Celestica call confirms that power components are already constrained at the hardware production layer and that liquid-cooled rack-scale design is central to next-generation hyperscaler deployments. This supports both component-level beneficiaries such as Monolithic Power and Vicor, and data center infrastructure beneficiaries such as Vertiv and Eaton.
Near-term trading catalyst: Positive order commentary from power and cooling suppliers should be interpreted as structurally supported rather than one-off. Power-component constraints may also become a leading indicator for AI server shipment pacing.
Longer-duration fundamental shift: AI infrastructure competition is moving toward rack-level and data-center-level power density. Suppliers that solve power and thermal constraints should capture a larger share of AI capex budgets over time.
SEMICONDUCTOR CAPITAL EQUIPMENT
ATS CAPITAL EQUIPMENT STRENGTH CONFIRMS THE AI CYCLE IS BROADENING INTO WAFER FAB EQUIPMENT DEMAND (READ-THROUGH 7)
Supporting call evidence: Celestica raised its full-year ATS revenue outlook to mid-to-high single-digit growth, citing “a reacceleration of customer demand in our capital equipment business” and a strengthening market forecast for wafer fab equipment spending in 2H 2026 and 2027. Management also stated in Q&A that capital equipment is entering “a very nice cycle” with a “very strong order book” from major customers that should take Celestica through 2026 and 2027.
Affected companies and impact: Applied Materials, Inc. (AMAT: US) is positive, moderate-to-high magnitude. Lam Research Corporation (LRCX: US) is positive, moderate-to-high magnitude. KLA Corporation (KLAC: US) is positive, moderate magnitude. ASML Holding N.V. (ASML: Netherlands) is positive, moderate magnitude. Tokyo Electron Limited (8035: Japan) is positive, moderate-to-high magnitude. Celestica Inc. (CLS: Canada) is positive, moderate magnitude through ATS.
Transmission mechanism: Celestica is not a wafer fab equipment company, but its ATS capital equipment business provides supply-chain visibility into equipment customer demand. The return to growth indicates that AI-driven silicon capacity expansion is feeding through to semi-cap equipment supply chains. This is especially relevant because Celestica’s AI hardware commentary simultaneously pointed to custom silicon and memory constraints, which creates a coherent demand signal for additional foundry, logic, DRAM, HBM, packaging, and test capacity.
Near-term trading catalyst: Semi-cap order and backlog commentary for 2H 2026 should screen more positively. Capital equipment suppliers with memory, advanced logic, and advanced packaging exposure should benefit most.
Longer-duration fundamental shift: AI infrastructure demand is creating a feedback loop from server/networking deployments to semiconductor capacity expansion. This extends the AI capex cycle upstream into wafer fab equipment and advanced manufacturing tools.
ODMS, EMS, SERVER OEMS AND HARDWARE VALUE-CHAIN CONTROL
AI HARDWARE VALUE IS SHIFTING TOWARD DESIGN-CAPABLE ODMS, CREATING SELECTIVE WINNERS AND PRESSURING GENERIC EMS OR BRANDED OEM MODELS IN HYPERSCALE ACCOUNTS (READ-THROUGH 8)
Supporting call evidence: HPS generated approximately $1.7B of Q1 revenue, grew 63%, and represented 42% of total company revenue. Management said the HPS business and design work underpin a significant portion of growth, with CPO, 1.6T, 800G, and 400G programs predominantly HPS. Celestica disclosed approximately 1,350 design engineers, materially above last year, and said engineers are working on next-year and year-after programs. Management also stated that customers are looking for Celestica to “innovate with them and design with them” across liquid cooling, rack-scale infrastructure, and thermal management.
Affected companies and impact: Celestica Inc. (CLS: Canada) is positive, high magnitude. Quanta Computer Inc. (2382: Taiwan) is positive, moderate-to-high magnitude. Wiwynn Corporation (6669: Taiwan) is positive, moderate-to-high magnitude. Hon Hai Precision Industry Co., Ltd. (2317: Taiwan) is positive, moderate magnitude. Foxconn Industrial Internet Co., Ltd. (601138: China) is positive, moderate magnitude. Jabil Inc. (JBL: US), Flex Ltd. (FLEX: Singapore), and Sanmina Corporation (SANM: US) are mixed-to-negative, moderate magnitude unless they can demonstrate comparable AI design content. Dell Technologies Inc. (DELL: US), Hewlett Packard Enterprise Company (HPE: US), and Super Micro Computer, Inc. (SMCI: US) are mixed, moderate magnitude, with positive AI demand offset by direct hyperscaler ODM competition.
Transmission mechanism: The highest-value AI hardware programs are not being awarded only on low-cost assembly. They require design engineering, advanced networking expertise, thermal architecture, liquid cooling integration, supply-chain planning, and large-scale execution. Celestica’s call indicates that the value pool is moving toward ODMs and EMS providers capable of co-designing differentiated systems. This is positive for AI-focused ODMs with engineering depth and capacity, but negative for generic EMS peers if they lack similar HPS-like capability. Branded server OEMs benefit from broad AI server demand in enterprise and sovereign AI channels, but the call is less favorable for their hyperscaler share because hyperscalers appear increasingly willing to work directly with design-capable ODM partners.
Near-term trading catalyst: Investors should focus on AI program win disclosures, design engineer growth, HPS-like mix, and capex commitments across ODM/EMS peers. Generic revenue growth without evidence of design intensity is likely to be valued less favorably.
Longer-duration fundamental shift: AI hardware outsourcing is becoming more strategic and less transactional. The market should increasingly differentiate between low-margin assemblers and true co-design partners with advanced AI networking and rack-scale capabilities.
