Who mentions $CLS most on Surfaced?

@investmentguru_ (InvestmentGuru) mentions $CLS most — 21 mentions in the past 12 months.

When was $CLS first mentioned on Surfaced?

The earliest tracked mention of $CLS in the Surfaced archive was on Nov 5, 2025.

How many voices track $CLS?

13 FinTwit voices have mentioned $CLS 62 times in the past 12 months.

What kind of voices mention $CLS most?

$CLS is mentioned most by Long Research voices on Surfaced.

$CLS — FinTwit mentions on Surfaced

@thevalueistApr 28, 2026 · 6:04 PM

$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.

Open on X

@aiinvestorhqApr 26, 2026 · 8:12 PM

Celestica has to be the best performing Canadian company? $10k 5 years ago would now be $492,400 $CLS https://t.co/Mgo0x3kTIO

Open on X

@jukan05Apr 24, 2026 · 6:09 AM

Exclusive: Google ramps new TPU servers, Taiwan suppliers gain share Google's unveiling of its eighth-generation tensor processing unit (TPU) at Cloud Next 2026 is expected to drive the next wave of growth in the application-specific integrated circuit (ASIC) server supply chain, with Taiwanese manufacturers expanding their role, according to supply chain sources. Assembly partners, including Celestica and Foxconn, are expected to begin volume production in the second half of 2026. Key component suppliers such as Asia Vital Components and Cooler Master are preparing shipments by the end of the second quarter, with Inventec reportedly set to begin production earlier. Taiwan gains share across supply chain Taiwanese vendors are taking on a larger role in Google's TPU server ecosystem. While Broadcom has historically dominated upstream IC design, MediaTek has also entered the supply chain, according to sources. In server motherboard production, Inventec's share is reportedly rising from about 10 to 30%, gradually taking orders from Celestica. The companies declined to comment. Supply chain sources said Taiwanese firms remain competitive in terms of flexibility, quality, and pricing, supported by long-standing experience in server manufacturing. Inventec is currently the world's second-largest server motherboard supplier. ASIC servers to outpace GPU growth ASIC servers are expected to be the fastest-growing segment of the AI server market in 2026, outpacing systems based on GPUs led by Nvidia. DIGITIMES Research estimates ASIC server shipments will grow 64.2% year-over-year in 2026, compared with 43.8% growth for GPU servers. Industry sources said rising ASIC shipments are also likely to support demand for general-purpose servers. Google's continued investment in ASIC infrastructure is expected to benefit its primary general server assembler, Quanta Computer, in 2026. Dual-chip TPU highlights shift to inference Google's eighth-generation TPU introduces a dual-chip architecture for the first time, separating training and inference workloads. The lineup includes TPU 8t for large-scale training and TPU 8i for low-latency inference. Both chips are expected to launch later in 2026, with shipments projected before year-end, according to supply chain estimates. The move reflects a broader shift in AI development. Over the past three years, the industry has focused on large-scale training, but leading chipmakers, including Google and Nvidia, are placing greater emphasis on inference, indicating a widening range of AI applications. Google remains key ASIC player Google continues to be a major driver of ASIC-based AI infrastructure. DIGITIMES Research estimates the company will account for 46% of the ASIC accelerator market in 2026, with shipments reaching about 3.33 million units. The TPU v7, code-named Ironwood, is expected to remain the main product, accounting for 89% of Google's total TPU shipments. Industry sources said Celestica remains the primary assembly partner for TPU servers, but Taiwanese firms are expanding their role. This is most evident in CPU motherboard production, where Inventec's share is increasing, and in L10 to L11 assembly, where Foxconn is involved. $CLS

Open on X

@investmentguru_Apr 19, 2026 · 5:10 PM

$CLS — Celestica: The Factory That Builds the Optical Switches Hyperscalers Actually Deploy Celestica has emerged as one of the leading players in AI data center development, delivering advanced networking, storage, and computing infrastructure — including 1.6 Terabit Ethernet switches designed for AI and machine learning data centers. Q4 2025 revenue hit $3.655B, up 44% YoY. The company raised its 2026 revenue outlook to $17B and committed $1B in capex to expand AI data center capacity. Not a pure photonics play — it’s where photonics components get assembled into the actual AI infrastructure being deployed. Fabrinet’s closest competitor with more diversified AI exposure.

Open on X

@thevalueistJan 29, 2026 · 12:06 PM

$CLS EXECUTIVE CALL SUMMARY: (CELESTICA INC, 10/28/25) Q3 2025 results materially exceeded prior guidance on revenue and adjusted EPS, driven by an acceleration in hyperscaler networking demand within the Connectivity & Cloud Solutions (CCS) segment and continued mix-driven margin expansion. Consolidated revenue of $3.19 billion grew 28% year-over-year and was stated to be “above the high end of our guidance range,” while non-GAAP operating margin of 7.6% expanded 80 bps year-over-year and was characterized as “the highest quarterly non-GAAP operating margin in the company’s history.” Adjusted EPS of $1.58 increased 52% year-over-year and was stated to be “exceeding the high end of our guidance range.” Management raised Q4 2025 guidance and raised full-year 2025 guidance for revenue, adjusted EPS, and free cash flow, while keeping the 2025 non-GAAP operating margin outlook unchanged. The call was also positioned as an Investor and Analyst Day, and therefore included unusually detailed multi-year operating commentary, including a “preliminary high-confidence” 2026 outlook of $16.0 billion revenue (+31%) and $8.20 adjusted EPS (+39%), plus qualitative framing for 2027 growth dynamics tied to 800G-to-1.6T networking transitions, scale-up Ethernet opportunities, hyperscaler custom ASIC compute ramps, and a rack-scale custom AI system program with a digital native customer targeted for 2027 production. Key takeaways with direct earnings-call implications included: (1) demand strength in data center networking (particularly 800G switch ramps) was stronger than the company’s prior expectations, with communications end-market growth reaching 82% versus prior guidance of “low 60s” growth; (2) the enterprise end market underperformed year-over-year due to a hyperscaler AI/ML compute program transition but was guided to return to growth in Q4 and to accelerate further in 2026 with full production of a next-generation custom ASIC program; (3) margin expansion was driven by operating leverage and richer mix (higher HPS contribution), but management also acknowledged that incremental growth vectors (rack integration, compute, system-level orchestration) can be margin-dilutive at the segment level, implying a persistent mix-management challenge even in a high-growth environment; (4) capital intensity is expected to step up meaningfully in 2026 (2.0%-2.5% of revenue) to support capacity and capability expansion, with management emphasizing that growth capex is program-tied and supported by forward demand visibility; and (5) customer concentration remained elevated, with 3 customers representing 30%, 15%, and 14% of Q3 revenue (59% combined), sharpening both upside torque to hyperscaler AI spend and exposure to customer-specific procurement behavior, dual-sourcing decisions, and program timing. QUARTER PERFORMANCE AND WHAT CHANGED VERSUS EXPECTATIONS The quarter was defined by a demand and mix surprise rather than a modest beat. Management explicitly framed the results as a top-end outperformance: “Revenue of $3.19 billion was up 28% and above the high end of our guidance range,” and adjusted EPS was similarly described as “exceeding the high end of our guidance range.” Translating the year-over-year growth rate implies a prior-year Q3 revenue base of approximately $2.49 billion, underscoring that the company added roughly $0.70 billion of incremental revenue year-over-year in a single quarter, despite the enterprise end market contracting meaningfully. Profitability expansion was broad-based: adjusted gross margin was 11.7% (+100 bps year-over-year) and non-GAAP operating margin was 7.6% (+80 bps). On an implied basis, Q3 non-GAAP operating income approximated $242 million (7.6% of $3.19 billion), versus an implied prior-year operating income of approximately $169 million (6.8% of ~$2.49 billion), reflecting a substantial step-up in operating profit dollars alongside margin expansion. The core delta versus prior guidance was concentrated in CCS communications. Communications end-market revenue growth of 82% was described as “above our guidance of low 60s percentage growth,” with management attributing the upside to “very strong demand in data center networking, primarily for ramping 800G switch programs across our largest hyperscaler customers, complemented by solid demand in our optical programs.” This statement matters because it indicates the beat was driven by in-cycle volume acceleration on existing hyperscaler programs (800G ramps) rather than solely by the timing of new program awards. Such in-quarter acceleration tends to carry higher operational risk (component supply, test capacity, line readiness, yield learning curves) but can also signal that customers’ near-term build rates are increasing, which can pull forward revenue and potentially create difficult year-over-year comparisons if customers normalize build rates later. The enterprise end market behaved largely as expected relative to guidance but remains a critical swing factor for 2026-2027. Enterprise revenue declined 24% year-over-year and was described as “in line with our guidance of a mid-20s percentage decline,” attributed to “a technology transition in an AI/ML compute program with a hyperscaler customer.” The guidance narrative implies a transition trough in Q3 that is expected to reverse as next-generation programs ramp, but it also highlights an inherent volatility in hyperscaler compute programs where product cycles can cause large year-over-year swings even without end-demand weakness. The investment implication is that enterprise compute revenue can appear “lumpy” quarter-to-quarter even in a structurally growing AI market, which increases the importance of validating program-specific milestones (silicon readiness, qualification gates, production ramp curves) rather than relying on linear growth extrapolations. Segmentally, CCS carried both the growth and a meaningful portion of the margin expansion. CCS revenue was $2.41 billion (+43% year-over-year) with segment margin of 8.3% (+70 bps). This implies CCS operating profit of approximately $200 million, or the majority of consolidated non-GAAP operating income. ATS revenue was $781 million (-4%) with segment margin of 5.5% (+60 bps), implying ATS operating profit of approximately $43 million. The margin improvement in ATS despite a revenue decline supports management’s repeated theme of portfolio reshaping and improved mix, but ATS remains structurally slower-growth and more macro-exposed (industrial, capital equipment, aerospace/defense) than CCS, with the call explicitly citing “market-related softness in our Capital Equipment business” as an offset to industrial and HealthTech growth. Within CCS, the HPS business was emphasized as the principal driver of growth and mix. HPS revenue of $1.4 billion grew 79% and represented 44% of total company revenue, with management stating: “Our HPS business generated revenues of $1.4 billion… and accounted for 44% of total company revenue.” This implies HPS was also approximately 58% of CCS revenue in Q3. The degree of HPS concentration inside CCS is strategically important because the business is positioned as a technology- and engineering-differentiated platform solutions portfolio (high-bandwidth ethernet switching, liquid cooling, system-level integration), and management repeatedly indicated HPS is margin-accretive to CCS. However, the same concentration increases sensitivity to hyperscaler networking roadmaps (800G to 1.6T cadence, optical architecture choices, scale-up vs scale-out networking approaches) and to customer procurement strategies (turnkey vs consigned silicon, exclusive vs dual source). WORKING CAPITAL, CASH FLOW, AND BALANCE SHEET SIGNALS Despite the strong profitability, Q3 free cash flow was $89 million, taking year-to-date free cash flow to $302 million. Inventory increased to $2.05 billion, up $129 million sequentially and $226 million year-over-year, while cash cycle days improved to 65 (down 1 day sequentially and year-over-year). The combination of rising inventory dollars with improving cash cycle days indicates inventory growth was at least partially offset by other working capital improvements (receivables and/or payables), but it still flags the operational reality of supporting rapid ramps: higher absolute inventory is commonly needed to support component lead times, buffering, and manufacturing throughput, particularly for high-complexity networking platforms and liquid-cooled systems. The investment risk is that elevated inventory levels can amplify downside if customer demand normalizes abruptly or if program transitions cause inventory obsolescence, especially in fast-iterating networking silicon and optics ecosystems. The mitigating factor cited implicitly was continued discipline in working capital and consistent free cash flow generation “on a quarterly basis going back many years,” but the quarter still demonstrated that free cash flow conversion can be meaningfully affected by growth-related working capital needs even when operating margins expand. Leverage and liquidity were positioned as conservative and supportive of stepped-up investment. Cash was $306 million, gross debt $728 million, net debt $422 million, and available liquidity was approximately $1.1 billion with no revolver draw. The gross debt to trailing 12-month adjusted EBITDA leverage ratio of 0.8x was described as improving sequentially and year-over-year. This low leverage posture is significant given the capacity and engineering investments discussed for 2026-2028; it implies a larger internal funding cushion and reduced financing risk if capex rises or if working capital absorbs incremental cash during growth ramps. Capital returns remained present but opportunistic. No shares were repurchased in Q3, with year-to-date repurchases of $115 million. Management indicated the normal course issuer bid was being renewed and expected to restart in November. The pause in Q3 repurchases, followed by an intent to resume, suggests repurchase pacing is being balanced against growth investments and working capital demands, and may also reflect share price levels relative to management’s internal return thresholds. GUIDANCE: Q4 2025 AND FULL-YEAR 2025 REVISIONS VERSUS PRIOR GUIDANCE Q4 2025 guidance was raised and framed with a notable macro-policy caveat. Management stated: “Fourth quarter revenue is projected to be between $3.325 billion and $3.575 billion,” and clarified that guidance assumes “no material changes in tariff or trade restrictions compared to what is in effect as of October 27.” The midpoint implies Q4 revenue of approximately $3.45 billion (+36% year-over-year) and adjusted EPS midpoint of approximately $1.73 (+56% year-over-year). Non-GAAP operating margin at the midpoint was guided to 7.6% (+80 bps year-over-year) with an adjusted tax rate of approximately 20%. The explicit tariff/trade assumption should be treated as a key conditionality: changes in trade restrictions can impact customer sourcing, cross-border manufacturing footprint utilization, component availability, and cost structure. The fact that it was highlighted again suggests management views policy volatility as a material uncertainty rather than a boilerplate risk disclosure. End-market guidance for Q4 implied a meaningful enterprise rebound and continued communications strength. In ATS, revenue was expected to decline in the low single-digit % range, with industrial and HealthTech growth offset by A&D portfolio reshaping and capital equipment softness. In CCS, communications was expected to grow in the high-60% range on “continued strong demand” for data center networking switches and “ongoing ramps in multiple 800G programs.” Enterprise was expected to “resume growth” with a low-20% increase driven by a next-generation hyperscaler application and AI/ML compute ramp. The enterprise reacceleration is a critical element for assessing whether Q3 was a transition trough (as positioned) or whether the enterprise portfolio will remain structurally volatile. Full-year 2025 guidance was increased meaningfully, and the comparison to prior guidance was explicitly disclosed. Management stated: “Our latest 2025 financial outlook now calls for revenue of $12.2 billion, up from $11.55 billion previously,” and “our adjusted EPS outlook is increased… to $5.90 per share,” from $5.50. Free cash flow outlook was increased to $425 million from $400 million, while the non-GAAP operating margin outlook remained 7.4%. The pattern of raising revenue and EPS while holding margin constant implies that incremental upside was expected to be broadly consistent with the existing margin framework, potentially reflecting a mix of pass-through revenue (including higher silicon content) and higher operating expense investment (R&D, engineering headcount, ramp costs) offsetting some operating leverage. This also suggests that the company may be deliberately conservative on margin guidance when visibility is high on volume but still evolving on mix (compute vs switching vs rack integration) and investment cadence. 2026 OUTLOOK AND THE DEGREE OF MANAGEMENT CONVICTION The call provided a relatively explicit 2026 financial model, described as “preliminary high-confidence.” Revenue was guided to $16.0 billion (+31% versus 2025 outlook), non-GAAP operating margin to 7.8% (+40 bps), adjusted EPS to $8.20 (+39%), and free cash flow to $500 million. Management stated: “We anticipate revenue of $16.0 billion,” and “our non-GAAP adjusted EPS is projected to be $8.20.” The guided operating margin expansion appears modest relative to the magnitude of revenue growth, and this was challenged directly in Q&A. Management’s response effectively indicated conservatism early in the year, stating the approach “isn’t that different than… previous years,” with “more to come as we go through 2026.” At the same time, management acknowledged mix trade-offs: networking/HPS is described as accretive, but “complex rack integration work… will be margin diluted,” implying that the margin trajectory is not purely a function of scale but also of portfolio composition and the commercial structure of system-level engagements. The capex framework is critical to interpreting both growth durability and free cash flow conversion. Management stated: “Next year, capital expenditures are expected to rise to between 2.0% and 2.5% of revenue,” and in Q&A quantified 2026 capex as “between $300 million and $400 million.” Maintenance capex was described as only approximately 0.4% of revenue, implying the majority of capex is growth-oriented and discretionary. Management emphasized program-tied investment and forward visibility, stating capex is “tied to customer programs” and supported by “good visibility on the demand profile going out multiple years.” Steve Dorwart added that visibility includes “about 12 to 15 months of real solid forecast inputs,” and in some cases extends further due to program-specific commitments such as ASIC quantities. The 2026 segment drivers were framed as CCS-led with sustained tailwinds. CCS revenue was expected to grow approximately 40% on continued networking strength (800G demand growth plus “earliest 1.6T programs in the 2H”) and very strong AI/ML compute growth as the next-generation custom ASIC program reaches “full-blown production.” ATS revenue was guided to be flat to mid-single-digit % growth, with healthy base demand and new wins partially offset by reshaping activities. The implied mix shift toward CCS is consistent with both the Q3 results and the investor day narrative, and it increases the structural dependence on hyperscaler and digital native AI infrastructure build-outs. TECHNOLOGY ROADMAP AND COMPETITIVE POSITIONING DISCLOSED ON THE CALL The Investor and Analyst Day content provided specific technical and roadmap disclosures that are relevant because they tie directly to the durability of the Q3 demand signal and to the plausibility of 2026-2027 growth. Management emphasized the company’s positioning as a “technology platform solutions company” with vertically integrated capabilities across design, manufacturing, supply chain, software, and aftermarket services. The recurring competitive claims centered on speed-to-market, complexity management, and execution quality. A representative quote was: “Managing complexity… and helping customers achieve speed to market… are what we really excel at,” alongside execution claims such as “zero critical excursions to customers.” Networking roadmap detail was unusually explicit for an EMS/platform solutions company. The call stressed 800G leadership and an expected transition into 1.6T and 3.2T platforms. A key quantitative disclosure was the 800G/400G mix: “we’ll end 2025 with roughly a 50% split between 800G and 400G,” and management stated that “the growth is primarily going to come from 800G.” On 1.6T, management indicated both program wins and timing: a ramp was visible with “1 customer… towards the backend of next year,” further ramps were anticipated into 2027, and an unidentified company speaker stated: “we have 10 programs currently underway in 1.6T.” Management also highlighted liquid cooling as increasingly central to high-bandwidth switching, stating liquid cooling “plays a key role” in 800G and 1.6T solutions. A case study was used to illustrate perceived differentiation, describing a hyperscaler engagement to design a rack-scale liquid-cooled “1.16” networking solution leveraging Broadcom Tomahawk 6 and requiring multi-region manufacturing. Management claimed an operational prototype was produced rapidly after silicon samples, stating: “we quickly stood up an operational prototype… and believe we were the first team anywhere to have done so.” In Q&A, Steve Dorwart reinforced speed claims: “From the initial receipt of silicon to complete functional power on the systems, we’ve done it in days,” versus competitors “in terms of weeks.” These claims, if validated operationally, are material because the cadence of switching generations is accelerating and early-cycle share capture can persist across subsequent generations due to incumbency and qualification inertia. The call also framed scale-up networking as a major incremental opportunity, linked to open standards initiatives and Tomahawk 6-based program wins. Jason Phillips characterized scale-up as “an emerging multi billion dollar new market opportunity,” and positioned the company as “well positioned” due to incumbency and capability. The investment relevance is that scale-up Ethernet could expand the addressable content per AI cluster beyond traditional scale-out switching, potentially increasing the networking content intensity of AI infrastructure builds. However, scale-up is also likely to attract intense competition (silicon vendors, OEMs, ODMs, specialized systems integrators) and could see evolving customer architecture choices, increasing uncertainty around market structure and margin pools. Software and system-level integration were highlighted as key enablers of the platform solutions thesis. Management stated the software engineering team is “nearly 400” and emphasized open-source network operating systems, stating SONiC is “positioned for continued market adoption.” The company described “Celestica solutions for SONiC” that “customizes and hardens” features, while also supporting hyperscaler proprietary NOS environments via abstraction layers and test/qualification. The importance of this disclosure is that software capability can increase switching platform stickiness, expand services revenue, and raise switching from a hardware-centric engagement to an integrated platform relationship. The risk is that software expansion increases opex, may dilute margins in the near term, and requires differentiated IP or services packaging to avoid becoming a cost-of-doing-business. Market share and TAM assertions were extensive and should be treated as management-provided. Management stated: “our cumulative market share… is 41%,” measured by total ports shipped across 200G/400G/800G, and that share leadership is “more pronounced” in hyperscaler/digital native custom solutions where the company accounts for “55%” of spend. The company also referenced forecasts including high-bandwidth Ethernet TAM of $50 billion by 2029 and accelerated compute TAM “nearly $500 billion by 2029,” plus a data center capex forecast “to surpass $1 trillion by 2028.” These numbers provide context for management’s growth confidence but require independent triangulation to assess realism, definitions (what is included/excluded), and how much of the spend is truly addressable by a platform solutions provider versus being captured by silicon, optics, OEM system vendors, or in-house hyperscaler manufacturing. DIGITAL NATIVE RACK-SCALE AI SYSTEM PROGRAM: UPSIDE AND UNCERTAINTY The digital native customer program targeted for 2027 was repeatedly emphasized as a major growth vector, and it introduced both upside and uncertainty. Management stated mass production is expected to begin in 2027 and explicitly stated the 2026 outlook does not include “any meaningful level of revenue coming from this digital native opportunity.” Timing was tied primarily to silicon availability, with samples expected in “the middle of ’26” and mass production in 2027. In sizing, management stated the program could be “multiple billions of dollars,” while also noting an unresolved commercial/accounting variable: “the treatment of the silicon… is it included or is it not included?” This is a critical modeling sensitivity. If silicon is included in reported revenue, headline revenue can inflate materially with lower margin % and higher working capital intensity; if silicon is excluded/consigned, revenue could be lower but margin % and cash conversion could be more favorable. The fact that silicon treatment was still under discussion indicates that headline revenue magnitude, working capital requirements, and cash flow profile for this program remain uncertain even as the strategic relationship is positioned as robust. ATS PORTFOLIO: MARGIN QUALITY IMPROVEMENT WITH MACRO AND POLICY HEADWINDS ATS was positioned as a portfolio optimization story rather than a primary growth engine. Management described 2025 ATS revenue as approximately flat year-over-year, with margin expansion from 4.6% at the end of 2024 to 5.5% in Q3 2025 and an expectation of 70 bps full-year margin expansion. 2026 ATS revenue was guided to be flat to mid-single-digit % growth. The growth drivers cited were industrial and HealthTech, including “more than $100 million of growth” expected in HealthTech in 2026 from diabetes care wins ramping. Offsets included continued A&D reshaping and softness in semiconductor capital equipment demand, with management indicating foundries are “holding off on adding more capacity until there’s greater clarity on tariffs and trade restrictions.” This reinforces that trade policy uncertainty is not only a guidance caveat for CCS but also a demand influence for ATS capital equipment customers. Portfolio pruning in ATS was explicitly tied to both margin discipline and demand deterioration in certain sub-verticals. Management described exiting “margin dilutive customers” and disengaging from “smaller… non-strategic customers,” with specific reference to changes in clean energy incentives affecting smart energy demand. This matters because ATS margin expansion appears to be driven partly by active portfolio contraction rather than purely by operating leverage, which can improve near-term profitability but can also reduce exposure to future cyclical rebounds if pruning is not tightly targeted.

Open on X

@thevalueistJan 28, 2026 · 9:44 PM

$CLS (Bloomberg) -- Celestica reported adjusted earnings per share for the fourth quarter that beat the average analyst estimate. FOURTH QUARTER RESULTS Adjusted EPS $1.89, estimate $1.75 (Bloomberg Consensus) Revenue $3.65 billion, +44% y/y, estimate $3.48 billion FIRST QUARTER FORECAST Sees revenue $3.85 billion to $4.15 billion, estimate $3.62 billion Sees adjusted EPS $1.95 to $2.15, estimate $1.83 YEAR FORECAST Sees revenue $17.0 billion, estimate $16.34 billion Sees adjusted EPS $8.75, estimate $8.45 COMMENTARY AND CONTEXT Raises 2026 Annual Outlook "As demand for AI-related data center technologies continues to strengthen, we now expect revenue of $17.0 billion and adjusted EPS (non-GAAP) of $8.75 for 2026." "We now anticipate our 2026 capital expenditures will be approximately $1 billion, or approximately 6% of our latest annual revenue outlook." "Driven by very strong results in 2025, and improved momentum into 2026, we are pleased to be raising our annual outlook." "We believe the revenue growth trajectory that we anticipate in 2026 will be sustained into 2027, and as a result, we are strategically increasing our planned capital investments to $1 billion this year."

Open on X

@thevalueistJan 27, 2026 · 3:34 PM

$CLS This worked. https://t.co/zFpj9BnNLn

Open on X

@yasutaketinDec 22, 2025 · 10:28 AM

そう。オニール流のような「基本」をきっちりやるということ。例えば $CLS セレスティカって何倍にもなったけど、最高値更新中だけ乗っかって、最近下がり始めたから早めに降りる。 https://t.co/fRFB4I3B5L

Open on X

@thevalueistDec 19, 2025 · 4:22 AM

$CLS Public information indicates that Celestica’s relationship to OpenAI is best described as “highly likely but not formally confirmed by the parties” rather than explicitly disclosed. Celestica has consistently described, in company communications, a “leading Digital Native Company” program award in its Hardware Platform Solutions (HPS) portfolio that entails a full rack AI-optimized system solution leveraging Celestica’s AI/ML server capabilities, 1.6T-class switching, and advanced liquid-cooling technologies, with production expected to begin in the latter part of 2026. Market commentary and certain sell-side channel checks have explicitly named OpenAI as the Digital Native customer tied to this rack-scale program, but that attribution appears to be sourced from analyst interpretation rather than a direct customer disclosure by Celestica or OpenAI. As a result, the relationship must be treated as “inferred from converging datapoints” rather than “verified by direct identification,” with the practical implication that the commercial scope and economics are material enough to influence Celestica’s forward financial profile, while remaining subject to the risks inherent in opaque customer attribution and contract confidentiality. The highest-confidence anchor point for the OpenAI linkage is the alignment of Celestica’s disclosed program timing and content with OpenAI’s stated infrastructure roadmap. In October 2025, OpenAI announced a strategic partnership with Broadcom to develop custom AI accelerators and rack-scale systems, with initial deployment targeted for 2H 2026 and scaling to approximately 10 GW of compute capacity by 2029. Broadcom’s corresponding announcement reinforces the same directional roadmap: purpose-built accelerators and rack-scale systems intended to support OpenAI’s compute expansion over several years. Celestica’s disclosed Digital Native “full rack AI system” award points to production starting in late 2026 and explicitly calls out the same rack-level integration problem space (servers, high-speed switching, and liquid cooling) that would be central to deploying custom accelerators at data center scale. The temporal overlap is notable because it reduces the probability that Celestica’s late-2026 rack-scale award is merely a generic AI infrastructure engagement; rather, it is consistent with a custom-silicon-driven rack platform entering production as soon as silicon supply and system validation permit. At the same time, the absence of direct naming remains a gating uncertainty, and attribution should be framed probabilistically rather than asserted as fact. If OpenAI is the Digital Native customer, Celestica’s supplier role would be expected to center on non-semiconductor system value rather than on silicon economics. The Celestica press release describing the award highlights a “full rack AI-optimized system solution” and specifies three major content buckets: AI/ML servers, 1.6T switching, and advanced liquid cooling. This framing strongly implies a rack-as-a-product model rather than a component-only model. In practice, that typically means Celestica is responsible for portions of mechanical design, board and system design integration, high-speed interconnect topology, power distribution and power delivery constraints, thermal architecture (air and liquid paths), system-level validation and test, manufacturing process engineering, and final rack integration. Even when the accelerator silicon is customer-furnished or procured through a separate channel, the integrator captures meaningful “non-silicon bill of materials” revenue and, more importantly, engineering and integration value add. This distinction matters because much of the debate around the Digital Native opportunity centers on whether accelerators are included in Celestica-reported revenue (turnkey model) or excluded/consigned (pass-through minimized), with major implications for reported top-line magnitude versus gross margin rate. The product set likely supplied into an OpenAI rack program can be decomposed into 3 layers that map directly to Celestica’s disclosed capabilities and to OpenAI’s architecture needs. The compute layer would include AI server platforms and rack integration around customer-specified accelerators, NICs/DPUs, memory/storage configurations, and power delivery constraints. The fabric layer would include high-bandwidth Ethernet switching that is increasingly being pushed toward liquid-cooled and higher-speed platforms as cluster scale rises. Celestica has publicly positioned itself as a leading supplier in 800G and 1.6T-class Ethernet switching, including liquid-cooled designs intended for AI workloads. The thermal layer would include liquid cooling subsystems (cold plates, manifolds, CDU interfaces, leak detection, validation processes, serviceability design) and would increasingly be system-level rather than component-level. Because the Celestica award language explicitly references advanced liquid cooling and 1.6T switching in the same rack-scale program statement, it suggests an integrated “compute plus networking plus cooling” rack-level system rather than a discrete switch-only program. A crucial nuance in interpreting “what Celestica supplies to OpenAI” is that OpenAI’s near-term capacity build is not exclusively custom-silicon-driven; it is also being scaled rapidly on NVIDIA-based infrastructure. OpenAI’s Stargate initiative, announced in January 2025, describes an intent to invest up to $500 billion over 4 years to build AI infrastructure in the US, with an immediate deployment plan starting at $100 billion. In July 2025, OpenAI announced that Oracle and OpenAI entered an agreement to develop 4.5 GW of additional Stargate data center capacity in the US. The Abilene, Texas Stargate campus—widely described as a flagship early deployment—has been built by Crusoe and is explicitly described as operating on Oracle Cloud Infrastructure (OCI) for OpenAI. Crusoe’s own September 2025 announcement states Oracle began delivering NVIDIA GB200 racks in June 2025 and that the campus had begun running early training and inference workloads. Independent data center industry reporting also describes the Abilene campus scaling to a multi-building, 1.2 GW site with OCI as the serving cloud layer for OpenAI. The implication is that OpenAI’s supply chain is operating in parallel across 2 major tracks: a near-term NVIDIA rack deployment track delivered through OCI/Stargate infrastructure partners, and a medium-term custom-accelerator rack track (Broadcom partnership) entering production in 2H 2026. Under that framework, Celestica’s most plausible direct exposure to OpenAI is skewed toward the medium-term custom-silicon rack program (late 2026 production, 2027 scale), while near-term NVIDIA racks at Stargate may be sourced through other ODM/integrators unless evidence emerges that Celestica is part of the GB200 rack supply chain for OCI. The “how Celestica works with OpenAI” question is best answered through the procurement and deployment topology that OpenAI has constructed. OpenAI is not merely renting generic cloud instances; it is co-developing and/or specifying dedicated infrastructure capacity with hyperscale and neo-cloud partners, then deploying that capacity into data centers that are built and operated by those partners (Oracle/OCI, Crusoe as developer, and others). In such a model, the rack supplier’s operational interface can take 2 forms. The 1st is a direct ship-to-operator form: the rack supplier contracts with the end-customer (OpenAI) but delivers equipment into the operator’s (Oracle/OCI) facilities under operator compliance, security, and acceptance test regimes. The 2nd is an operator-procured form: the operator (Oracle/OCI) contracts for hardware on OpenAI’s behalf and delivers a managed service to OpenAI, with hardware procurement treated as part of the service build-out. Public statements confirm Oracle is delivering NVIDIA racks into Abilene for OpenAI workloads, which strongly indicates at least some hardware is operator-procured in the Stargate context. For Celestica, this matters because the customer-of-record determines credit exposure, working capital mechanics, warranty/service obligations, acceptance milestones, and the degree to which revenue recognition is tied to operator acceptance versus end-customer acceptance. The Oracle and Crusoe dimension is often misinterpreted as a “hardware supplier relationship” rather than an “infrastructure hosting and deployment venue.” The highest-confidence public data describes Crusoe as the data center developer/partner building the Abilene Stargate campus, and OCI as the serving cloud layer used for OpenAI workloads at that site. Oracle is described as delivering NVIDIA GB200 racks into that campus in June 2025. This does not, on its face, establish Celestica as a supplier to Oracle for Stargate racks. Instead, it establishes that OpenAI’s compute deployment venue includes OCI-operated facilities built by Crusoe, and that at least the initial rack deliveries are NVIDIA GB200 systems supplied through Oracle’s procurement channel. For Celestica, the key analytic question becomes whether Celestica’s “full rack AI system” award is designed to be deployed into the Stargate/OCI environment (as a parallel or incremental cluster type) or whether it is intended for a separate OpenAI-controlled deployment footprint. The available public record supports the inference that OCI/Stargate is becoming a major deployment venue for OpenAI, making it a likely destination for any rack-scale program intended to power OpenAI training and inference at scale, but it does not confirm vendor-level participation for Celestica within the Abilene NVIDIA rack supply chain. The “neo-cloud” layer introduces additional complexity and reduces confidence in any single-vendor share assumption. OpenAI has explicitly broadened its compute sourcing beyond traditional hyperscale channels. Reuters reporting indicates OpenAI deepened its partnership with CoreWeave via a contract worth up to $11.9 billion (reported March 2025) and then signed an additional contract expansion worth up to $4 billion (reported May 2025). OpenAI’s own Stargate communications have also referenced ongoing projects with CoreWeave alongside the Oracle partnership, reinforcing that multiple infrastructure partners are being used in parallel. This multi-partner approach implies that the relevant “customer” for a rack supplier might vary by cluster type and deployment venue, and that OpenAI’s procurement is increasingly sophisticated and multi-sourced across both infrastructure partners and hardware integrators. It also implies that OpenAI’s rack-scale spend will not be monolithic; different venues may standardize on different rack architectures, cooling solutions, and network fabrics based on power availability, deployment speed, and operator preferences. Competitive dynamics in OpenAI rack supply are further highlighted by OpenAI’s disclosed partnership with Foxconn. Reuters reported in November 2025 that Foxconn signed a partnership with OpenAI to jointly develop AI data centers, and that Foxconn would design and manufacture data center racks, including cabling, networking, and power solutions. This directly overlaps with the “non-silicon rack content” that would otherwise be core to Celestica’s Digital Native rack opportunity. Consequently, even under a scenario in which Celestica is supplying OpenAI, the supplier landscape appears to be multi-vendor with meaningful overlap in capability sets. The most likely segmentation is by platform generation (NVIDIA GPU racks versus custom accelerator racks), by deployment venue (OCI/Stargate, CoreWeave, other hosted environments), and by subsystems (compute trays, network racks, cooling integration, assembly/test). Under that segmentation, Celestica’s relative advantage would be strongest where high-speed networking and liquid-cooled rack integration are tightly coupled and where design/engineering incumbency is critical to achieving rapid platform bring-up and stable yield at scale, while Foxconn and other ODMs may compete effectively on large-scale manufacturing capacity and broader commodity integration. The reference to “Caruso” does not map cleanly to widely documented OpenAI infrastructure partners in publicly available sources reviewed. The closest and most consistently documented entity in this context is Crusoe, which is repeatedly described as a data center developer/partner for OpenAI’s Stargate campus in Abilene, operating on OCI for OpenAI workloads. If “Caruso” refers to a distinct entity, public confirmation of a material OpenAI infrastructure partnership was not identified in the sources reviewed, and therefore should not be assumed as a meaningful node in the OpenAI-Celestica supply chain absent additional documentation. Investment implications for Celestica, under an “OpenAI as Digital Native customer” base case, are material and multi-dimensional. The opportunity size implied by market commentary and management-level qualitative framing is large enough to reshape revenue concentration and growth visibility. Sell-side commentary cited in secondary sources has described a rack revenue opportunity on the order of $10 billion tied to OpenAI expansion and framed Celestica as a beneficiary of “non-semiconductor content” in that build-out. Even allowing for potential overstatement in secondary reporting, the directional conclusion remains that a single rack-scale customer ramp can plausibly create multi-billion-dollar annual revenue contributions at scale for an integrator that is already operating at a roughly low-teens billions revenue base. The primary risk is not simply “win versus lose” but “economics, scope, and accounting treatment.” If accelerators are consigned and excluded from reported revenue, reported top-line uplift could be materially smaller than implied by total rack value, but gross margin % could be structurally higher. If accelerators are included as pass-through, revenue could be larger but margin rate lower, with working capital and supply chain risk higher. This creates a scenario where headline revenue guidance can diverge meaningfully from underlying value capture, making margin dollars, free cash flow, and ROIC more decision-relevant than revenue alone. Execution and timing risk is also elevated because the most plausible OpenAI-linked Celestica program appears gated by silicon availability and platform readiness. OpenAI’s Broadcom partnership frames first deployment in 2H 2026, which implies a narrow integration window for rack-scale platforms incorporating new accelerators, new thermal solutions, and potentially new network fabrics. Celestica’s Digital Native program likewise points to late-2026 production start. In rack-scale systems, small slips in silicon qualification, optics availability, power/cooling constraints, or software stack maturity can translate into 1-2 quarter revenue shifts, which can be disproportionately impactful when the ramp is expected to be steep. The presence of multiple OpenAI infrastructure venues (OCI/Stargate and CoreWeave) adds a second-order risk: platform standardization may diverge across venues, increasing the probability of multi-sourced hardware designs and lowering the likelihood of single-vendor dominance for any given rack architecture. Strategic positioning implications are constructive but should be assessed relative to competitive encroachment. The strongest “moat” argument for Celestica in an OpenAI-linked ramp is the coupling of high-speed networking, liquid cooling, and rapid new-platform bring-up. Celestica’s public messaging around 1.6T-class switching, liquid-cooled designs, and rack integration suggests the company is investing directly into the pain points that emerge as AI clusters scale in bandwidth and power density. OpenAI’s infrastructure strategy, as evidenced by Stargate and the Broadcom partnership, is explicitly oriented toward scaling dedicated capacity and deploying custom rack-scale systems, which is an environment where design-led integrators can capture increased value relative to traditional box-build EMS. However, OpenAI’s Foxconn partnership demonstrates that equally scaled competitors are actively positioning for the same rack content categories. This raises the probability that competitive bidding pressure will emerge after initial ramps, with OpenAI and its infrastructure partners seeking second-source options once platforms are stabilized, particularly for later waves of deployment. A final implication is that the OpenAI relationship, even if real and sizable, would likely increase Celestica’s already-high customer concentration and cyclicality to the AI capex cycle. OpenAI’s multi-venue expansion plan (Stargate with Oracle, plus CoreWeave) signals sustained demand, but it also signals procurement sophistication and potential renegotiation leverage once supply constraints ease. In that regime, Celestica’s differentiation must remain anchored in speed-to-market, yield, reliability, and system-level performance rather than in commoditized assembly. The most investable framing is that an OpenAI-linked program would be a “proof point” for Celestica’s transition toward higher-value platform solutions and could meaningfully extend visibility through 2027-2029, but it would simultaneously increase sensitivity to program-specific timing, multi-sourcing decisions, and the evolution of OpenAI’s platform architecture as it blends NVIDIA-based deployments with custom-accelerator racks.

Open on X

@thevalueistDec 18, 2025 · 4:41 PM

$CLS Celestica's AI Pivot, Growth Strategy, and Management Quality The video provides a comprehensive financial and management analysis of Celestica, focusing heavily on its successful transformation into an AI infrastructure platform company. The analysis details the strategic pivot toward high-margin, design-led solutions within the Connectivity & Cloud Solutions (CCS) segment, highlighting robust financial guidance, strong growth in high-performance computing (HPS), and the risks associated with customer concentration and competition. It offers a deep assessment of the senior management team, praising the leadership's strong operational track record, clear strategic execution (particularly the shift to high-value platforms), and exceptional alignment with shareholders through high equity ownership and performance-based compensation that has resulted in massive total shareholder returns. Collectively, the documents argue that the company's high valuation is justified by both its strategic positioning in the booming AI market and the quality of its highly aligned executive team, despite inherent cyclicality and concentration risks.

Open on X

$CLS

$CLS

About this stock

Related stocks

Frequently asked