Celestica Inc. Deep-Dive Research

Its first stage can be summarized as "expansion and acquisition integration in the post-IBM contract manufacturing era." The 1998 annual report shows that revenue had already reached $3.2 billion that year, and the company completed multiple acquisitions in the IPO year, including bringing IMS into the system, which gave it important Asian manufacturing capacity in one step. At that time, Celestica, like many EMS companies in the late 1990s, believed scale, geographic coverage, and customer transfer would automatically convert into higher share and higher profit. Facts later showed that this model worked well in upcycles, but would also fail quickly when demand collapsed.

The second stage was a long period of post-bubble cleanup and structural adjustment. Public market data show that Celestica's historical low closing price occurred on 2009-03-09, at $2.59. From a large IPO star in 1998, it fell into the category of a standard low-valuation manufacturing stock. The company did not disappear during this stage. Instead, it slowly shifted its business focus from pure communications and enterprise electronics manufacturing toward ATS, where reliability requirements are higher, product life cycles are longer, and certification barriers are heavier. The company's history page lists actions such as the 2010 acquisitions of AlliedPanels and Invec, the 2011 acquisition of Brooks Automation's semiconductor equipment contract manufacturing business, and the 2012 acquisition of D+H Manufacturing. The constraint behind these decisions was practical: traditional communications and enterprise hardware outsourcing had margins that were too thin and demand that was too volatile. To live more steadily, the company had to find orders in aerospace, industrial, medical, and semiconductor equipment markets that were less easily destroyed by price wars.

The third stage began when Robert Mionis became CEO in 2015. The history page records the appointment in only one line, but in hindsight this was the real starting point of Celestica's new transformation. The key move in the Mionis era was to cut unattractive orders first, rather than chase every order. In 2020 and 2021, the company advanced a CCS portfolio review and completed a staged exit from Cisco by the end of 2020. The 2020 20-F stated directly that the exit agreement with Cisco began in October 2019, and that the overall CCS portfolio reshaping corresponded to about $1.25 billion of annualized CCS revenue decline, of which about $500 million was related to Cisco. The 2021 annual report then showed that, affected by Cisco's exit, CCS 2021 revenue fell 9% year over year, but CCS segment margin improved from 3.5% in 2020 to 3.9% in 2021. This matters. Over more than a year, Celestica told the market it was willing to sacrifice low-quality revenue in exchange for a better business mix.

The fourth stage is the AI and high-bandwidth Ethernet leap after 2022. In 2022, company revenue reached $7.25 billion, up 29% year over year, with both the CCS and ATS segments improving. In 2024, full-year revenue reached $9.65 billion, up 21% year over year. In 2025, revenue further rose to $12.39 billion, up 28% year over year. Revenue was not the only change. Margins rose in tandem: Q4 2023 non-IFRS operating margin was 6.0%, increasing to 6.8% in Q4 2024 and 7.7% in Q4 2025. Q1 2026 CCS segment margin had already reached 8.6%. This marked that the company was no longer just "soldering more boards," but was capturing higher unit value and better structural profit in AI data center networking, hardware platforms, and more complex system assembly.

This stage also had two key capital-market milestones. First, Onex completed two secondary sales in June and August 2023. After the related multiple voting shares were converted into subordinate voting shares, Celestica's control structure changed from a "historical legacy structure with a heavy PE color" into a single common-share structure. The 2025 annual report shows that in 2023 the company issued about 18.60 million subordinate voting shares for these two secondary sales to convert the former Multiple Voting Shares, and no multiple voting shares remained at year-end. Second, the company lost its "foreign private issuer" status on 2024-06-28, and from 2025-01-01 it had to report under U.S. domestic issuer rules using 10-K, 10-Q, and 8-K filings, while restating 2024 interim reports under U.S. GAAP. Together, these two changes reduced the governance discount and improved the convenience with which U.S. investors could understand and trade the stock.

By 2026, Celestica's story no longer looks like that of an "old-line contract manufacturer that survived." It looks more like a manufacturing platform company that has rediscovered its place in the AI infrastructure wave. The capability it has proven is a less visible but extremely hard-to-replicate set. Original chips or proprietary network operating systems were never its cards. Its cards were the willingness to cut low-quality revenue, reallocate the customer structure toward richer demand pools, combine global manufacturing execution with complex product introduction, and upgrade financial reporting conventions and governance structure when needed. The problem is that the market has already seen these achievements, and seen them very fully.

Longitudinal Financial Review

Celestica's financial history can be summarized in two sentences. The first is the "slimming down for quality" from 2018 to 2021: total revenue was $6.633 billion in 2018, fell to $5.888 billion in 2019, $5.748 billion in 2020, and further to $5.635 billion in 2021. The most obvious drag came from the CCS portfolio review and Cisco's exit. The second is the "re-acceleration" after 2022: revenue returned to $7.25 billion in 2022, reached $9.65 billion in 2024, reached $12.39 billion in 2025, and Q1 2026 alone had already reached $4.047 billion. The change in revenue direction shows that this company is a manufacturing platform strongly dependent on business mix and downstream capital expenditure cycles, not a steady-state cash cow.

The structure behind revenue is even more worth examining. In 2020 and 2021, the HPS business had already started ahead of total revenue: 2020 HPS revenue grew 80% year over year to $862 million, accounting for 15% of total revenue; 2021 HPS revenue continued to grow 34% to $1.15 billion, accounting for 20% of total revenue. In other words, Celestica did not suddenly catch AI data centers in 2024. As early as the Cisco exit period, it had been slowly shifting the internal center of gravity of CCS from traditional OEM outsourcing toward service providers and data center platforms. The breakout the market sees today is the realization of that thread in 800G, AI networking, and hyperscaler orders, not a new business that appeared from nowhere.

Profit quality has also changed visibly. Net income was only $70.30 million in 2021, with a 1.2% net margin. By 2025, the Q1 2026 presentation disclosed that FY2025 GAAP net income had reached $832.5 million. Q4 2024 adjusted EPS was $1.11, rising again to $1.89 in Q4 2025. Q1 2026 reached a new high of $2.16. The profit increase was not only "more volume." It came from three factors layered together: first, higher contribution from HPS and high-bandwidth networking platforms within CCS; second, mixed margin improvement after exiting low-quality projects; third, a larger revenue base spreading fixed costs. Q1 2026 CCS segment margin of 8.6%, up another 0.6 percentage points from the prior-year period, is the most direct evidence of this logic.

Cash-flow quality is not as attractive as the income statement, which is also the main reason I am most cautious on the current valuation. In 2025, operating cash flow was $659.5 million. In Q1 2026, operating cash flow was $356.3 million, but capex in the same period was as high as $218.4 million. Capex/Revenue reached 5.7%, materially above prior quarters, leaving quarterly free cash flow at only $137.9 million. The Q1 2026 presentation also provided a full-year framework: under a revenue outlook of about $19.0 billion, expected free cash flow is about $500 million. Based on the 2026-06-11 closing market value, this corresponds to an FCF yield of only about 1.2%. In other words, Celestica's valuation today rests on the market's belief that a large part of this high capex cycle is expansionary investment that will become higher profit and higher cash flow in the future, rather than on already realized free cash flow. That belief may not be wrong, but it must be verified by data over the next two years.

The balance sheet is the most reassuring part of this story. At the end of Q1 2026, the company had cash and equivalents of $378 million, term loans of $719 million, net debt of about $341 million, and both GAAP and adjusted trailing-12-month leverage of only 0.6x. In mid-2024, the company also completed new credit arrangements. The annual report disclosed that the amended facility included a $250 million Term A, a $500 million Term B, and a $750 million revolving credit facility, with maturities for Term A and the revolver extended into the 2029/2031 range. The Q1 2026 presentation further noted that the revolver was increased by another $1.0 billion, with total liquidity approaching $2.0 billion. For a manufacturing company in the middle of capital-intensive expansion for AI orders, this leverage level is moderate.

On capital allocation, management's behavior over the past three years is also worth noting. The company spent $150.7 million on share repurchases for cancellation in 2025 and $152.0 million in 2024. In Q1 2026, it repurchased and canceled another 73,000 shares, spending $20 million. The new NCIB launched in October 2025 allows the company to continue repurchasing about 5.70 million shares before 2026-11-02. This means management has not completely abandoned shareholder returns because of the large share-price rally, but has kept a balance among capacity expansion, deleveraging, and repurchases. At the current valuation level, however, the lift to per-share value from buybacks is already much less obvious than it was in 2023-2024.

My conclusion from the longitudinal financial review is clear: Celestica has transformed from an EMS company that "survived by scaling" into an AI hardware platform manufacturer that "raises margins by improving business mix." This transformation is real. But it has not yet reached the stage where cash-generation ability steadily exceeds market expectations. The financial statements have proven margin improvement; cash flow is still in the process of proving itself.

Share Price and Valuation History

Celestica's share-price history, like its business character, looks like manufacturing in the first half and an AI growth stock in the second half. Long-term price databases show that the company's historical low closing price was on 2009-03-09, at $2.59, while the historical high closing price was on 2026-06-02, at $472.40. The split-adjusted IPO issue price in 1998 was only $8.75. In other words, this stock spent more than a decade convincing the market that "it is only a low-valuation manufacturing stock," then forced the market over the past three years to admit that "it is no longer just a low-valuation manufacturing stock."

Breaking down the past decade, there are at least four phases. The first phase, around the period before 2018, was a long low-valuation consolidation period. The second phase, from 2019 to 2021, saw the market price in the CCS portfolio exit, Cisco decoupling, and pandemic-period material constraints, leaving little room for imagination in the share price. The third phase, from 2022 to 2024, saw both revenue and profit accelerate, and the valuation rose from the cheap range of traditional EMS. The fourth phase, from 2025 to early 2026, saw the market price Celestica almost as a scarce hardware platform asset for AI network infrastructure. The stock surged during 2025, then quickly retreated from historical highs in June 2026. This pullback does not necessarily mean fundamentals have deteriorated. It looks more like a situation where expectations had risen too high, and even "good news that is not even better" can compress multiples first.

From a capital-market structure perspective, Celestica has also undergone a change that was often ignored before but must be taken seriously today. First, trading liquidity has clearly shifted toward the U.S. primary listing venue. Google Finance shows that as of mid-June 2026, average volume for NYSE:CLS was about 2.36 million shares, while average volume for TSX:CLS was about 456,000 shares. Based on visible daily average volume, the U.S. market accounts for about 84% of two-venue trading. Second, it was added to the S&P/TSX 60 on 2025-12-22, after already being included in the S&P/TSX Composite. This brings more passive-fund attention and more stable Canadian domestic index holdings. For a stock already heavily traded by U.S. AI-theme capital, dual listing and index inclusion together magnify liquidity and volatility.

The change in valuation label is the most important part of this share-price history. Looking at Celestica today, three headline numbers stand out: based on the 2026-06-11 closing price of $372.35 and Google Finance's trailing-12-month EPS of $8.27, the static P/E is about 45x; based on company Q1 2026 full-year adjusted EPS guidance of $10.15, the forward P/E is about 36.7x; combining the 2026 full-year revenue outlook of about $19.0 billion and Q1-end net debt, EV/Sales is about 2.27x and EV/EBIT is roughly around 28x. These multiples are high in the traditional EMS/ODM world, but not the wildest in an AI networking equipment story. Celestica's problem and opportunity both sit in this "in-between valuation."

Business Model and Moat

Celestica's business engine is essentially "walking on two legs, but real valuation elasticity comes from only one leg." The first leg is CCS, or Connectivity & Cloud Solutions. The most important asset here is the company's ability to co-define, introduce, and rapidly mass-produce complex hardware platforms with hyperscalers, service providers, and large cloud customers, rather than its factories. The 2025 annual report describes CCS as serving communications and enterprise end markets, while Q1 2026 showed CCS quarterly revenue had reached $3.241 billion, up 76% year over year, with segment margin of 8.6%. The second leg is ATS, Advanced Technology Solutions, covering aerospace and defense, industrial, healthcare technology, and capital equipment. ATS provides cyclicality diversification and certification barriers, and its role is to make the company less dependent on a single cloud capital spending cycle, even though it is not the main source of the current valuation rerating.

The moat in CCS is different from a software company's moat. It does not come from locking customers into code, nor from controlling a core chip ecosystem. It comes from a long series of operational thresholds: early participation in customer platform definitions, design-for-manufacturing capability, supply-chain coordination, yield ramp, quality control, global delivery, and the ability to iterate across multiple hardware generations. These thresholds are not as visible as software moats, but once a hyperscaler entrusts a large-scale platform to a supplier, switching costs are real. The issue is that these switching costs are more "execution switching costs" than "ecosystem switching costs," so they can be eroded by pricing pressure, technological generation changes, or an alternative supplier proving stronger delivery.

Customer concentration is the shadow on this moat. The 2025 annual report shows the two largest customers accounted for 19% and 13% of total revenue, respectively, and both came from CCS. The top ten customers accounted for 72%. This is normal for a company deeply serving hyperscalers, but normal does not mean low-risk. The good side is that once a platform is recognized by a large customer, revenue can ramp very quickly. The bad side is that any single customer's capex rhythm, platform path, or supplier allocation decision can directly change Celestica's growth curve. This is why the market treats it as an AI infrastructure platform story, but also why it cannot be priced like a fully diversified high-quality compounder.

The ATS segment is more like a ballast. Aerospace and defense, industrial, healthcare technology, and semiconductor capital equipment usually have more complex certification, longer product cycles, and higher requirements for reliability and compliance. This part of the business does not make Celestica as exciting as Arista or Accton, but it allows the company to carry a manufacturing base and customer breadth across cycles. In valuation terms, ATS is not the main source of high multiples, but it can reduce the probability that the entire company is dragged back to a single-cycle story. Its role is defense, not offense.

The biggest variable in the business model is whether Celestica can continue moving up the value chain from EMS toward "hardware platform partner." If the customer sees it only as capacity plus execution, then the long-term valuation ceiling should still be close to high-quality manufacturing. If the customer sees it as a partner that can participate in next-generation network and rack-level platform introduction, then margins and multiples can remain structurally above traditional EMS. Q1 2026 CCS margin of 8.6% is encouraging evidence, but not yet permanent proof. In manufacturing, the harsh reality is that every generation change brings a new round of bidding, qualification, and price negotiation.

In terms of cost structure, Celestica is still a manufacturing company. The vast majority of costs come from materials, components, outsourcing, and direct manufacturing costs, so gross margin will not suddenly become software-like at Arista levels. Yet it is not the most ordinary EMS company either, because once platform definition, validation, prototyping, customer introduction, and complex supply-chain management are locked in, customer-switching friction can be very high. The rise in Q1 2026 Capex/Revenue to 5.7% is a reminder: for this type of business to remain competitive, it must keep adding capacity, equipment, and engineering resources. When revenue rises, operating leverage is clear; when revenue falls, profit is also more sensitive to mix than in ordinary contract manufacturing.

I think Celestica has four real moats. The first is complex product introduction and global manufacturing execution. Many companies can design boards, and many can assemble servers, but few can ramp complex hyperscale customer platforms simultaneously across global sites while controlling delivery, quality, and certification timing during material shortages. Celestica's ability over the past few years to raise CCS segment margins after Cisco's exit shows this execution system is not a slogan. The second is platform embedding through co-development with hyperscalers. Once it enters the validation cycle for a customer's next-generation switch or rack platform, competition is no longer only about quoted price; it is about whether the supplier can keep up with the customer's timetable and iteration rhythm. The third is multi-industry certification and high-reliability manufacturing capability from ATS. It will not grant a high multiple, but it provides some cushion when the cycle weakens. The fourth is management's discipline in portfolio adjustment. From 2019 to 2021, it was willing to lose $1.25 billion of annualized CCS revenue to exit low-quality projects. Not every manufacturing company can do that.

The moats that do not exist, or at least have not been fully proven, also need to be made clear. Celestica does not have a network operating system and brand premium like Arista, nor silicon and ecosystem control like Nvidia or Broadcom. In front of customers, it is still mostly a "co-developed hardware platform partner," not an "irreplaceable owner of system architecture." This means its moat is more like a relationship-based barrier deeply embedded in supply chains and engineering processes, rather than an explicit pricing-power barrier. In tailwinds, this kind of barrier works well; in headwinds, customers will still renegotiate prices and reallocate orders.

At the governance level, Celestica has two positives over the past two years. First, Onex completed two secondary sales in 2023, the multiple voting shares were fully converted, and the historical control structure essentially exited. Second, after losing foreign private issuer status in 2024, the company shifted to U.S. domestic issuer rules and U.S. GAAP disclosure, which is friendlier for U.S. capital to understand financial statements, quarterly conventions, and the governance framework. In current public materials, I have not seen evidence of major financial fraud or auditor-change turmoil. The point to note is that the restatement of the company's 2024 interim reporting convention was an accounting and disclosure-status transition, not a major issue with business reality.

Industry and Cycle Analysis

When placing Celestica back into its industry, the easiest mistake is to put it only in the "EMS sector." Today it straddles two profit pools. The first profit pool is electronics manufacturing services, typically characterized by large revenue, thin gross margins, cash flow dependent on capex and receivables turnover, and persistently low valuations. The second profit pool is data center AI infrastructure, especially high-speed Ethernet switching, open-networking platforms, and more complex system-level hardware. The former determines its heavy-asset base; the latter explains why the market has rerated it. Celestica is valuable because it has secured a richer slice of AI networking infrastructure inside EMS, not because the EMS industry suddenly became attractive.

The first demand-side driver is hyperscaler capital expenditure. In Q1 2026, Meta raised its full-year capex outlook to $125.0 billion to $145.0 billion. In Alphabet's Q1 2026 trailing-12-month free cash flow convention, capex corresponded to about $109.9 billion. Amazon's trailing-12-month "purchases of property and equipment, net of proceeds from sales and incentives" through Q1 2026 reached about $147.3 billion, and the company explicitly said the increase mainly reflected AI investment. In other words, Celestica's most important industry variable is how aggressively a small number of U.S. cloud giants remain willing to spend on AI data centers, not the traditional EMS cycle. As long as that waterline is high, Celestica's CCS has soil for continued acceleration. If large customers begin talking about "efficiency" rather than "expansion," the impact on Celestica will arrive earlier and harder than for general industrial manufacturing.

The second supply-side driver is rising Ethernet penetration in AI data center back-end networks. In its late-2025 forecast, Dell'Oro defined 2025 as an inflection point: Ethernet adoption in AI back-end networking surpassed InfiniBand. IDC also disclosed that the global Ethernet switch market grew 35.2% year over year in the third quarter of 2025. This is critical for Celestica. If AI cluster networking equipment choices move more toward Ethernet and open-networking architectures, the content opportunity available to the company on the white-box/ODM side expands. If the industry returns to more closed software and system ecosystems with stronger binding, more value will be captured by upstream branded vendors.

Celestica's cyclicality is therefore a stack of three cycles. The first is the capital expenditure cycle, which matters most. The second is the technology iteration cycle, from 400G to 800G and then 1.6T, with every generational leap reallocating orders and share. The third is the traditional manufacturing inventory and macro cycle. In upcycles, the most beneficial variables are hyperscaler network capex, the number of platform introductions, and CCS mix improvement. In downcycles, the most fragile variables are customer concentration, price negotiation, and cash-flow mismatch caused by capex moving first while orders retreat later.

Policy and geopolitical risk are not background noise here. Celestica's operations are global, with nearly 30,000 employees at the end of 2025 and more than 20,000 employees in Asia. U.S. export controls, tariff adjustments, restrictions on advanced data center-related activity involving China, and policy friction around overseas manufacturing sites can all affect its material sourcing, site layout, and customer order allocation. The more practical point is that AI data center hardware itself has become an industry affected by national security, compute policy, and local-production demands, rather than purely commercial equipment. Celestica is not directly locked down by a single policy, but its position means it will remain exposed to these constraints for a long time, rather than suffer one round of pressure and be done.

Horizontal Peer Analysis

Celestica's comparable group is not singular. If we only ask "who also manufactures," we get EMS peers such as Jabil, Flex, and Sanmina. If we ask "who makes money in AI data center networking," we see branded and white-box switch players such as Arista and Accton. If we ask "who takes hyperscaler system-level orders," we return to server and rack ODMs such as Quanta. The most valuable comparison in researching Celestica is to see which of these three categories it resembles more, rather than build an exhaustive peer table. My answer is: in asset and cash-flow substance, it still looks like a manufacturer; in growth and market sentiment, it is closer to a data center networking hardware supplier; in valuation controversy, it sits exactly between those two worlds.

Arista is the most important "ceiling reference." Arista's Q1 2026 revenue was $2.709 billion, up 35.1% year over year, and net margin remained 37.8%. Google Finance shows its market cap at about $189.88 billion and static P/E at about 52x. Customers choose Arista because of EOS, programmability, network operating consistency, and stronger brand credibility, rarely because its hardware is cheaper. Arista sells network control. Celestica cannot reach that layer, so it should not receive Arista's valuation logic on equal terms. If the market truly puts the two too close together, it usually means Celestica has been overheated in the narrative.

Accton is the most important "parallel reference." It is also a networking and communications solutions ODM/OEM, and it directly targets open networking, white-box switches, and AI/ML hardware. Google Finance shows Accton's Q1 2026 revenue at NT$70.12 billion, net income at NT$8.34 billion, net margin at 11.9%, P/E at about 43.5x, and market cap at about NT$1.28 trillion. Customers often choose Accton because it moves faster in the white-box and open-networking camp, has a purer product lineup, and has deeper hardware collaboration with hyperscalers. Compared with Accton, Celestica's advantage is its global manufacturing network and the stabilizer of its diversified ATS business. Its disadvantage is that the "purity" of its networking assets is not as high, and investors do not understand it as directly as Accton.

Quanta is the "system-level scale reference." Quanta's Q1 2026 sales were about NT$809.2 billion, up 66.6% year over year. Google Finance shows its P/E at about 19.3x and market cap at about NT$1.43 trillion. Customers usually choose Quanta because of its scale, delivery, and cost-control capability in AI servers, full systems, full racks, and system integration. For Celestica, Quanta reminds investors of two things. First, even if AI server and rack businesses are strong, system-level ODMs do not necessarily get high multiples automatically. Second, Celestica's current valuation premium must come from the idea that it is closer to the networking platform profit pool than server ODMs; otherwise this premium cannot stand.

Jabil is the "manufacturing benchmark reference." In fiscal Q2 2026, Jabil revenue was $8.28 billion, up about 24% year over year, and it raised FY2026 revenue and core EPS guidance to about $34.0 billion and $12.25, respectively. Customers choose Jabil mostly because of customer breadth, manufacturing coverage, and execution stability, not because it controls a specific AI networking platform. The difference between Celestica and Jabil explains why Celestica could be rerated by the market. But Jabil's existence also reminds us that the average valuation center of manufacturing will not permanently relocate because of one AI cycle. If Celestica is ultimately placed by the market back into the drawer of "excellent manufacturer" rather than "platform-style AI infrastructure stock," the drawdown will be fast.

In ecosystem position, I define Celestica as a "challenger-type manufacturer for AI data center hardware platforms," not a branded leader and not a pure niche player. The gap it fills is that hyperscalers, when unwilling to give all value to branded networking vendors, need a platform partner that can co-define, co-produce, and co-iterate. The profit pool it most directly attacks comes partly from low-quality orders in traditional EMS and partly from the space between branded switches and purer white-box players. The potential attackers on its profit pool also come from two directions: above are vendors with software and brands such as Arista/Cisco; below are Accton, Quanta-related players, and a broader group of Asian ODMs. Celestica has a good position, but it is not empty territory.

Current Fundamentals and Bull-Bear Debate

The latest four quarters make Celestica's upcycle very clear. In Q4 2024, company revenue was $2.55 billion, up 19% year over year, with adjusted EPS of $1.11. In Q4 2025, revenue rose to $3.65 billion, up 44% year over year, and the company raised its 2026 outlook. By Q1 2026, revenue further reached $4.047 billion, up 52.8% year over year, adjusted EPS reached $2.16, the full-year revenue outlook was raised to about $19.0 billion, and Q2 2026 revenue guidance was $4.15 billion to $4.45 billion with adjusted EPS guidance of $2.14 to $2.34. Few traditional manufacturing companies can raise growth and margins like this over a year and a half.

But what truly determines the share price is not only whether results beat, but whether expectations can keep rising after the beat. After Q1 2026, the market is clearly trading two variables. The first is whether CCS growth can remain high, especially whether share and pricing can be preserved through the transition from 800G to 1.6T. The second is capex and free cash flow. In Q1 2026, company capex surged to $218.4 million, far above the prior-year period. This shows the company is currently in "rapid expansion," not "mature cash collection." Therefore, although the market narrative is AI, the more precise fundamental question is whether "AI-driven scale expansion can eventually convert into stable, high-quality cash flow."

Analyst sentiment is currently clearly optimistic. Google Finance shows that among 13 analysts covering CLS over the past three months, 12 rated it Buy and 1 rated it Hold. This consistency itself deserves caution. When almost the whole market stands on one side, the next earnings report is more likely to trigger "the realization that it is not as good as imagined" than "the realization that the company is good." Celestica's share price now looks like a high-revving engine. Continued upside requires Q2, Q3, and Q4 to keep proving to investors that $19.0 billion of 2026 revenue is only a waystation, not the cycle peak.

The core bull evidence has three layers. First, the company is confirming high growth through official guidance, rather than relying on the market to dream on its behalf. Second, the product generation transition is moving quickly. The Q1 2026 presentation stated that the 800G portfolio is ramping and that the 1.6T platform will begin ramping with two hyperscalers in the second half of 2026. Third, downstream demand variables remain strong, and the most important cloud vendors, Meta, Alphabet, and Amazon, have not shown signs that they have "stopped investing." Put together, bulls would say Celestica is not at the end of realization, but in the middle of it.

The core bear evidence also has three layers. First, customer concentration is already very high. The top two customers accounted for 32% of 2025 revenue, meaning any change in order rhythm at a single hyperscaler can hit revenue directly. Second, cash flow has not caught up with the profit narrative. The combination of $500 million of full-year free cash flow and a $42.8 billion market value leaves little room for error. Third, competition is not one-directional: above sit branded and software-layer moats represented by Arista; below sit the price and delivery competition of Accton, Quanta, and broader ODM players. Celestica certainly has a position, but "a good position" is not the same as "a secure position."

My summary of the current state is: fundamentals are very strong and the narrative has real support, but the share price has already turned "strong" into the baseline rather than a surprise. What this stock should fear most next is good news becoming old news, not bad news.

Valuation Analysis

Start with the conclusion: Celestica's current valuation cannot be judged only by P/E, nor can it be dismissed entirely using the low multiples of traditional EMS. It is more like a "platform growth stock with a manufacturing base," so a reasonable method should blend three sets of numbers: forward earnings, owner earnings, and the multiple coordinates of peers with different business models. Looking only at the first set makes it look too cheap; looking only at the second makes it look too expensive.

Historically, Celestica's valuation center has clearly moved up. Based on the 2026-06-11 closing price of $372.35, static P/E is about 45x. Based on 2026 adjusted EPS guidance of $10.15, forward P/E is about 36.7x. Compared with Arista at 52x, Accton at 43.5x, and Quanta at 19.3x, Celestica is now placed by the market in a position "above system-level ODMs, below the strongest branded networking vendor, and close to pure white-box networking players." This relative coordinate is not absurd, but it has clearly moved away from traditional EMS valuation habits.

Once cash flow is looked through, valuation immediately becomes less comfortable. Q1 2026 company full-year free cash flow outlook is about $500 million. Using current market value, the FCF yield is only about 1.2%. Even acknowledging that a meaningful part of 2026 capex is expansionary rather than pure maintenance, I still find it hard to describe today's share price as "cheap based on realized cash flow." The more reasonable explanation is that the market is willing to temporarily ignore the current low FCF yield and buy more mature profit and cash flow in 2027-2028. The problem is that this is equivalent to paying for the next two years of execution in advance. The margin of safety is therefore naturally thin.

The company has not provided a precise split between maintenance capex and expansion capex. I can only make a conservative inference: 2026 capex is materially higher than in previous years and lines up exactly with 800G volume, 1.6T introduction, and larger customer platform ramps, so much of it clearly has expansionary characteristics. Even so, I am unwilling to fully ignore this cash outflow in valuation, because for a company like Celestica that must continually follow next-generation platforms and capacity needs, so-called "expansion capex" is not necessarily a one-off project. It may well be the new normal required to maintain its position over the next few years. In other words, owner earnings can be higher than 2026 reported FCF, but not nearly as optimistic as GAAP or adjusted net income.

Valuation Scenarios

The following table is a price scenario analysis under the research framework and does not constitute investment advice. I use 2027 as the multiple anchor, rather than focusing only on 2026, because Celestica's current valuation is already essentially trading 2027 sustained growth.

For the price signals corresponding to the table, I apply stricter rules: the conservative scenario first produces intrinsic value, then I apply an approximately 20% margin of safety to derive the ideal buy zone. For the bull scenario, I add about 10% upward as the "clearly overvalued" warning zone. Based on this, my price bands are as follows: ideal buy zone $190-$220; acceptable hold zone $330-$390; above $500-$570 should be treated as clearly overvalued. The current price of $372.35 is inside the "acceptable hold" zone and toward its upper end. The support for this judgment is that under the base scenario it is not cheap, but also not expensive enough to require immediate exit. Conversely, new capital does not receive enough margin of safety at this point.

Where could expectation gaps come from? I see only three places. First, the mass-production pace of 1.6T. The Q1 2026 presentation says "beginning in the second half of 2026, introduction with two hyperscaler customers." If the market mentally turns that into "fully contributing profit in the second half" before earnings, expectations can become too full. Second, CCS segment margin. As long as it can stay around 8%, the market will continue to view the company as platform hardware. Once it falls back to 7% or below, the valuation anchor will quickly move back toward manufacturing. Third, free cash flow. The company has already explained the growth story. What will most change the bull-bear balance next is showing cash flow, not telling the growth story again.

On margin-of-safety review, my judgment is "not obvious." Based on the conservative scenario intrinsic value estimate above, the current price is not at a discount; it is at a significant premium. Among the three scenarios, the most fragile assumption is that the market remains willing to grant a mid-to-high multiple above 26x. If that assumption is cut by 30%, meaning the market compresses the base scenario multiple from around 28x back to roughly 20x, base scenario value would fall to around $250, far below the current share price. If earnings do not grow over the next three years, Celestica may still sustain a valuation above traditional EMS because of its technology position and platform attributes, but I find it hard to call that return curve a margin of safety. Good company, bad price is at least true at today's level.

Risk Analysis

The first risk is customer concentration and a swing back in hyperscaler capex. This is a medium-probability, extremely high-impact risk. In 2025, two customers accounted for 19% and 13% of revenue, both from CCS, and the top ten customers accounted for 72%. This means Celestica is capturing alpha from a small number of customers, not diversified industry beta. If any large customer enters a stricter capex review period in 2027, or reallocates network orders to another ODM, system vendor, or branded vendor, Celestica's CCS growth and margins would come under pressure at the same time, and the market would quickly cut the multiple. The indicators to watch are whether single-customer concentration rises again in the next annual report and whether CCS growth begins to diverge from the capex direction of a few large customers, not media reports.

The second risk is that during the generational transition from 800G to 1.6T, Celestica may not preserve share and margin together. This risk has medium-to-high probability and high impact. Management has clearly stated that 1.6T will be introduced to two hyperscalers in the second half of 2026, which is of course an opportunity. But generational transitions are also the moments when requalification, repricing, and order reallocation are most likely. Above Celestica are players such as Arista and Cisco that control network systems and brand support. Below are Accton, Quanta, and broader Asian ODMs with cost and flexibility. If 1.6T order growth comes with ASP declines faster than expected, CCS margin will peak before revenue does. The red lines investors should watch are whether CCS segment margin falls below 8% for two consecutive quarters and whether the company stops emphasizing multi-customer 800G/1.6T ramps.

The third risk is capex and cash-flow mismatch. Its probability is high and its impact is medium-to-high. Celestica is now in a phase of expanding capacity while delivering, not a phase where "orders arrive and cash simply comes in." Q1 2026 Capex/Revenue had reached 5.7%, and full-year free cash flow guidance was about $500 million, making the FCF yield relative to current market cap very thin. If high capex over the next two years does not smoothly convert into higher cash recovery, or if customer payment rhythms slow and receivables and inventory usage rise, the valuation framework will be hit before revenue forecasts. The reason is that the market bought the stock assuming "cash will settle after expansion." Once that assumption loosens, the share price will retreat faster than profit.

The fourth risk is that the rerating came from mix improvement, and valuation can also be cut by mix deterioration. This risk has medium probability and high impact. Celestica's greatest success in recent years has been exiting low-quality revenue and raising the share of HPS and networking platforms. But the reality of manufacturing is that once downstream customers enter a calmer investment phase, they will renegotiate prices, disassemble BOMs, compare lead times, and test substitutability. Celestica does not have Arista's software lock-in, nor Nvidia/Broadcom's silicon and ecosystem control. Its valuation premium therefore depends heavily on the market believing that "it is moving toward a higher-quality hardware platform supplier." The signal to watch most closely is revenue still growing while CCS margin begins to decline persistently. That usually means mix is deteriorating, not scale improving.

The fifth risk is geopolitics and policy. Probability is medium and impact is medium, but once triggered it can last a long time. Celestica's employee base and manufacturing network are highly globalized, with a high share of employees in Asia. AI data center hardware also sits at the intersection of export controls, tariffs, energy, and local-capacity policies. For Celestica, these risks usually show up as slower introduction of certain platforms, higher costs at certain sites, more component substitution, or more conservative order allocation by certain customers. They rarely show up directly as "one ban causes revenue to go to zero." It is more like a chronic condition than a one-off accident.

Catalysts and Tracking Indicators

The most meaningful positive catalysts are: first, 1.6T truly starting to form revenue rather than staying at product release; second, full-year guidance being raised again, especially the $19.0 billion revenue target being reconfirmed ahead of schedule; third, CCS segment margin remaining above 8% during high growth; fourth, capex beginning to fall back in 2027 while free cash flow is released materially faster than market expectations; fifth, ATS not dragging the company down and instead providing additional resilience through aerospace, healthcare, and capital equipment.

Negative catalysts are easier for the market to magnify. First is a guidance cut, especially weaker order visibility in CCS. Second is the AI customer capex narrative shifting toward efficiency, or at least from "raising" to "pausing increases." Third is CCS margin falling before free cash flow rises. Fourth is competitors using more aggressive pricing in 1.6T or rack-level platforms to win orders. Fifth is that after the extreme 2025 rally, Celestica itself carries a sentiment premium, and as long as the slope of high growth flattens slightly, the share-price reaction will usually be larger than the fundamental change.

Tracking Dashboard

In this table, the four fundamental indicators to watch most closely are not share price, but the combination of CCS growth, CCS margin, Capex/Revenue, and free cash flow. As long as the first two remain stable and the latter two improve, Celestica's high valuation can still be digested by time. If the first two begin to loosen and the latter two do not improve, the market will quickly pull it back from "AI hardware platform" to "cyclical manufacturing stock." Among external indicators, Meta, Alphabet, and Amazon capex is the waterline for Celestica's real orders.

Horizontal-Longitudinal Synthesis

From the longitudinal perspective, what Celestica has truly proven is that "it can actively refresh itself in long-term manufacturing competition," not only that "it caught AI." Many old-line EMS companies choose to defend revenue when the cycle is weak, eventually pushing themselves into a loop of low gross margin, low valuation, and low growth. Celestica did the opposite from 2019 to 2021: it first endured revenue decline and the pain of a major customer exit, then exchanged that for higher-quality customers, more complex platforms, and higher-growth end markets. This capability is far more important than a single quarterly beat. Without that painful restructuring, there would be no sustained rise in CCS margins later and no current position in AI data center networking.

Its past success of course benefited from the times. Without the current hyperscaler capex cycle, which easily reaches the hundred-billion-dollar level, Celestica could not have risen into what it is today. The structural change described by Dell'Oro, with Ethernet surpassing InfiniBand in AI back-end networks, also gives it a better narrative position than traditional server ODMs. But it would be unfair to attribute all success to the cycle. What truly allowed it to break out was the groundwork management laid before the cycle arrived: cutting low-quality projects, increasing HPS investment, keeping ATS as a stabilizer, moving the company from FPI status to a disclosure framework more understandable to U.S. capital, and simplifying the control structure. The era provided the wave; Celestica at least knew how to build the boat.

From the horizontal perspective, its advantage versus peers lies in the combination of "engineering platform + global manufacturing + customer introduction + demand absorption," not in a single extreme capability. Arista's advantage is software-defined networking and brand control. Accton's advantage is white-box networking purity and a more direct open-networking bet. Quanta's advantage is system-level scale and full-rack server execution. Jabil's advantage is customer diversification and manufacturing breadth. Celestica sits among these companies. Its best side is that it has both partial networking hardware platform attributes and manufacturing execution capability. Its worst side is that none of these capabilities is so strong that competition disappears. This weakness is not temporary; it is structural. Therefore, Celestica's reasonable valuation should always be below the strongest branded networking vendors, but can be above standard manufacturing peers. The current share price is hard to judge precisely because the market has already priced this intermediate state quite optimistically.

The market's most likely misjudgment is to treat "high growth" directly as "high-quality free cash flow growth." Celestica's financial statements look very strong today, but cash flow has not yet reached a level sufficient to support an extremely high premium. Strong Q1 2026 capex and thin full-year FCF guidance mean it remains one step away from being a "self-reinforcing cash-cow growth stock." If 1.6T and more complex platform orders over the next two years can raise profit and cash flow together, today's premium can be digested. If not, the share price will return to a more conservative framework before profit does. For a company that has already completed a successful rerating, this step is not far, but the price consequences are large.

The most important variable over the next 1 year is whether the shipping pace of 1.6T and CCS segment margins can remain stable. The most important variable over the next 3 years is whether hyperscaler AI networking and rack infrastructure investment can transition from "surging" to "stable at a high level," rather than swing back quickly. The most important variable over the next 5 years is whether Celestica can truly convince the market that what it has captured in CCS/HPS is a repeatable profit pool, not a one-time dividend from an extreme capex peak. If it can, today's high multiple only reflects part of the future in advance. If it cannot, today's high multiple is overdrawn.

For me, the path for this company to become a better investment candidate is clear. Either the price comes down and the margin of safety reappears, or data over the next three quarters proves that free cash flow is being released much faster than I currently expect. Conversely, two situations would make me overturn the original view: one is CCS margin falling below 7.5% for two consecutive quarters; the other is the company still spending heavily on capex while full-year revenue and EPS no longer get revised up. That would mean the core assumptions the market is paying for have started to fail.

Bull Case

• The company has moved from the 2019-2021 stage of removing low-quality revenue from CCS into the 2025-2026 stage of high growth driven by AI data center networking platforms. The transformation is not a concept; it has already been written into revenue, margins, and guidance.

• Q1 2026 CCS revenue grew 76% year over year and segment margin reached 8.6%, showing this growth is not simply low-price volume expansion.

• Management has clearly stated that 800G continues to ramp and 1.6T will be introduced to two hyperscalers in the second half of 2026, so product generation progress has not stopped at 800G.

• Meta, Alphabet, and Amazon AI infrastructure investment remains high, and the most important external demand variable has not turned negative.

• The balance sheet is sound. Q1 2026 net debt was only about $341 million and leverage was 0.6x, leaving the company financial room to continue expanding and withstand volatility.

Bear Case

• The top two customers accounted for 32% of 2025 revenue, making customer concentration high enough for a single customer's rhythm change to directly rewrite full-year profit.

• At the current price, the 2026 guided free cash flow yield is only about 1.2%, leaving a thin cash-flow margin of safety.

• Celestica does not have Arista's software and brand moat, and it is not as "pure networking hardware" as Accton, so a high premium requires continuous proof.

• The 1.6T generational transition is both a catalyst and a window for requalification, repricing, and order reallocation.

• The company's current valuation is materially above server and system-level ODMs such as Quanta, meaning the market is already paying in advance for growth over the next two to three years.

Pre-mortem

The first scenario I worry about most happens in 2027. After completing a round of front-loaded construction in 2026, Meta, Amazon, and Alphabet shift AI data center investment from "racing for time" back to "checking returns." At the same time, Accton and broader Asian ODMs use more aggressive pricing to win 1.6T orders, forcing Celestica to protect volume by cutting prices. The result is that CCS segment margin falls from around 8% in 2026 to 6%-6.5%, adjusted EPS retreats to the $8-$9 range, and the market multiple falls from the low 30s back to 18-20x. A share price falling to $150-$180 would not be surprising then, implying a three-year decline close to or above 50%.

The second scenario is milder but still painful for shareholders. Celestica continues to grow, revenue still rises above $20.0 billion in 2027, and EPS does not decline materially. But capex stays high, free cash flow never comes through, and the market gradually admits that it is more like a "high-growth manufacturing company" than a "scarce platformized hardware asset." In that case, the company may not suffer an earnings collapse, but valuation could slowly contract to a forward EPS multiple in the 20s. The share price could then consolidate for a long time in the $250-$300 range, consuming all the time value of investors who chased the stock in 2025-2026.

Final Research Conclusion

My final judgment on Celestica is this: it is a company that has truly improved its business structure, but not a stock that can ignore price. What deserves respect is that it has pulled itself out of the old low-quality EMS framework and moved onto the main path of AI data center networking and platform hardware. What makes me hesitate is that the current share price has already written in a considerable amount of victory over the next two years, while cash-flow verification has not fully caught up. Buying Celestica today is essentially a bet that "the company will keep getting better, and get good enough to support this multiple structure," rather than a bet on "whether the company will keep getting better." These are different things.

If you already hold it, I do not think there is enough evidence to crudely classify it as a bubble and exit immediately. The demand environment, product generation transition, and company execution are still enjoying tailwinds, and the market remains willing to pay high multiples for AI infrastructure hardware stories of this kind. But if you are deploying new capital today, I think the most important discipline is to wait for better odds rather than chase confirmation. Celestica's biggest shortcoming now is margin of safety, not growth. It is still a "good company," but no longer a "good price."

The conditions under which I would change my view are also clear. On the upside, if the company proves over the next 2 to 3 quarters that 1.6T and new platform introductions can also lift free cash flow, I would be willing to acknowledge that the current base valuation is too conservative. On the downside, as soon as CCS margin weakens consecutively, or full-year revenue/profit guidance stops being raised while high capex continues, this story needs to be repriced.

【Company Profile Scorecard】

• Fundamental quality: high

• Growth: high

• Moat: medium

• Financial stability: strong

• Management credibility: high

• Valuation attractiveness: low

• Risk level: medium-high

• Suitable investor type: long-term growth / cyclical

【Investment Rating】

• Rating: Hold

• One-sentence investment thesis: The business reinvention has already been recognized, but the current share price looks more like it is paying for continued perfection over the next two years.

• Three price signals: 【Ideal/Fair Buy Price】190-220 USD Basis: this corresponds to conservative scenario intrinsic value of about $240-$275 after applying an approximately 20% margin of safety; only a return to this range would restore meaningful odds. Acceptable hold price: 330-390 USD

• Clearly overvalued price: 500-570 USD

• Current price classification: acceptable to hold

• Worth waiting for a better price: yes. If the share price returns below $220, while CCS margin remains around 8% and full-year guidance has not been cut, I would consider the odds materially improved. The opportunity cost of waiting is that the company may continue using performance to "earn into" the high valuation.

• Target holding period: 3-5 years

• Expected annualized return: conservative about -12%; base about -1% to +2%; bull about +9%. These are three-year scenario projections and do not constitute investment advice.

• Maximum loss risk: if hyperscalers cut network capex in 2027, 1.6T competition intensifies, CCS margin falls back to 6%-6.5%, and valuation compresses to 18-20x, the share price could retreat to $150-$180, about 50%-60% below the current price.

• Signals that trigger reassessment: CCS segment margin below 7.5% for two consecutive quarters

• Full-year revenue guidance cut below $18.5 billion

• Free cash flow fails to cover high capex for two consecutive quarters, and full-year FCF margin remains below 2%

• Single-customer revenue share in the annual report exceeds 25%

• Management language on the 1.6T introduction pace moves back from "production introduction" to "validation/pilot production"

The above price bands and return estimates are based on the conservative/base/bull assumptions in this section for 2027 earnings power, cash flow, and multiple ranges, and are calculated in reference to the current market price.

【Valuation Range】

• current: 372.35 (as of the 2026-06-11 close)

• bear (conservative · ideal buy zone): [190, 220]

• base (reasonable · acceptable hold zone): [330, 390]

• bull (optimistic · above clear overvaluation line): [500, 570]

Key Data Tables

The table above is a high-confidence longitudinal data summary, compiled from Celestica 2020/2021 20-F filings, the 2025 10-K, 2022/2024/2025 full-year earnings releases, and the Q1 2026 results announcement. Cash-flow details for a few years were not fully captured in the current review, so they are left blank rather than guessed.

The most important part of this table is "what the market is paying for," not "who is more expensive." Arista's high multiple buys system and software control. Quanta's lower multiple shows that even high-growth system-level manufacturing does not automatically command a high valuation. Accton shows that pure white-box/networking ODMs can receive valuations far above traditional server ODMs. Celestica's current valuation is hard because it touches all three logics, while none is strong enough to fully overwhelm the other two.

Research Uncertainties

First, HPS standalone revenue often cited by the market, the share of the top three CCS customers, and some specific design-win lists are not fully disclosed in currently parseable first-party announcements. In the body of the report, I prioritize officially disclosed CCS aggregate data and customer concentration, and do not treat segment figures lacking clear first-party text confirmation as core valuation inputs.

Second, the company has not clearly split maintenance capex and expansion capex. The 2026 capex increase is clearly related to AI-related capacity expansion, but how much of it will fall back in the future determines whether owner-earnings valuation can be revised upward.

Third, I did not obtain sufficiently clean first-party confirmation text on roles related to rack-level solutions such as AMD Helios in this review. I treat this as a potential upside option, not a main assumption in the base valuation.

Fourth, quarterly data and valuation conventions for Taiwanese peers are less available and less structured on English-language sites than for U.S.-listed companies. Therefore, in the horizontal comparison, I use "high-confidence key figures + business positioning" for Accton and Quanta, rather than making an overly detailed financial breakdown.

Fifth, this report uses the 2026-06-11 closing price, the latest complete trading day, as the primary convention. Because the research reference date is 2026-06-12 in the Tokyo time zone, U.S. and Canadian stocks had not completed a new close that day, so I did not replace the closing convention with intraday prices.

Reference Sources

• Celestica official investor relations and historical materials: 2025 10-K, 2021/2020 20-F, Q1 2026 results announcement and presentation, history and milestones, investor FAQ, and annual reports page.

• Celestica historical earnings releases: 2022, 2023, 2024, and 2025 full-year and Q4 results.

• Capital-market and liquidity data: Yahoo Finance historical prices, Google Finance quote pages, and public materials from S&P Dow Jones Indices on the S&P/TSX 60 and Composite.

• Downstream hyperscaler capex and industry demand: Meta, Alphabet, and Amazon Q1 2026 earnings materials, and public summaries of the data center switch market from Dell'Oro and IDC.

• Horizontal peers: Arista Q1 2026 results, Jabil FY2026 Q2, Flex FY2026, and public market data and investor materials for Accton and Quanta.

Other Securities Mentioned in the Report

• ANET.US — Branded switch leader, used to compare the software and brand profit pool that Celestica cannot capture.

• 2345.TW — Accton is the most important comparable in the white-box/open-networking camp, used to test the competitive intensity of 800G/1.6T.

• 2382.TW — Quanta represents the system-level scale logic of AI server and rack ODMs.

• JBL.US — Jabil is a beneficiary of the AI cycle among traditional EMS companies, used to compare the normal valuation framework for manufacturing.

• FLEX.US — Flex reflects the margin and cash-flow benchmark of a mature electronics manufacturing leader.

• META.US — Meta's capex is one of the most important external anchors on Celestica's demand side.

• GOOG.US — Alphabet/Google Cloud's AI infrastructure spending affects demand for Ethernet and platform hardware.

• AMZN.US — Amazon/AWS data center investment determines how long the industry cycle can last.

• CSCO.US — Cisco was both a key historical customer transition point for Celestica and represents the competitive direction of branded networking vendors.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.