Celestica becomes the AI factory floor

Dear WIMM Readers,

my analysis of my 3rd portfolio continues with Celestica - a technology leader dedicated to driving customer success and market advancements.

Onto the update:

TL;DR

Celestica is often described as an electronics manufacturing services company. That label is technically correct and strategically lazy. The old version of the story was simple. Customers outsourced complex manufacturing, Celestica built the hardware, and margins stayed thin because manufacturing is usually a brutally competitive business.

The new story is different. Celestica now sits much closer to the AI infrastructure buildout. Its largest segment is Connectivity & Cloud Solutions, which includes Communications and Enterprise. In plain language, this means networking switches, servers, storage, AI/ML compute programs, and hardware platform solutions for hyperscalers. The smaller Advanced Technology Solutions segment serves Aerospace & Defense, Industrial, HealthTech, and Capital Equipment.

The moat starts with scale. Customers building AI infrastructure care about speed, reliability, component availability, capacity, engineering execution, and the ability to ramp without drama. A low-cost manufacturer can quote a price. A trusted infrastructure partner can absorb complexity. That difference is the moat.

The second lever is switching cost. Once Celestica is designed into a hyperscaler program, especially in 800G, 1.6T, AI compute, or rack-scale systems, the customer relationship becomes harder to displace. The customer is now buying execution certainty in an environment where delays cost real money. Idle GPUs are expensive furniture.

The third lever is design capability. Hardware Platform Solutions matters because it pulls Celestica upstream. The company is working on design and manufacturing for advanced systems, including scale-up networking with AMD’s Helios architecture and a 1.6T co-packaged optics Ethernet switch with a hyperscaler. That is a different business from “please build this thing cheaply”.

The headline number was revenue: $4.05B, up 53% year over year. The better number was adjusted operating margin: 8.0%, a new company high. The best number was adjusted ROIC: roughly 50%. In a hardware and manufacturing business, that combination matters. Revenue growth without returns is volume theater. Celestica is showing volume with economics.

Adjusted EPS grew 80% to $2.16, above the high end of guidance. GAAP EPS was $1.83, up sharply from $0.74 last year. This was not just an AI revenue story. It was an operating leverage story. Higher volumes, better mix, and portfolio reshaping are flowing through the P&L.

The segment mix explains the inflection. CCS revenue was $3.24B, up 76%, with segment margin of 8.6%. ATS revenue was $806M, flat, but segment margin improved to 6.0%. That is a useful contrast. CCS is the growth engine. ATS is the margin repair and diversification engine.

Management raised the 2026 annual outlook from $17B to $19B of revenue. Adjusted EPS guidance moved from $8.75 to $10.15. Adjusted operating margin guidance moved from 7.8% to 8.1%. The free cash flow outlook stayed at $500M, even with $1B of CapEx. That is the trade-off. The company is reinvesting aggressively because the demand signal is unusually strong.

The CapEx number deserves attention. $1B of planned CapEx is not casual spending for a company like Celestica. It is the physical expression of customer commitments. Management said the investment is supported by awarded programs and multiyear capacity alignment with key customers. Translation: customers are pulling Celestica into bigger commitments because they need capacity and execution.

Management acknowledged in the last call the supply constraints in advanced components. One AI/ML compute ramp was partially gated by component availability, though management said this was a supply issue rather than a demand issue. That distinction matters. The bear case is demand evaporating. The current issue is supply pacing growth.

The most important disclosure was customer concentration. Three customers each represented at least 10% of revenue, at approximately 35%, 15%, and 15%. This is both the moat and the risk. Hyperscalers are very large customers. They create enormous scale, deep visibility, and program stickiness. They also have bargaining power.

Celestica is moving from being a supplier to being a capacity in the hyperscaler roadmap. That is a better business, but also a more demanding one.

Celestica’s near-term strategy is simple: execute the AI infrastructure ramp without breaking the supply chain, the balance sheet, or the customer trust layer.

The concrete initiatives are clear.

Near-term catalysts:

The most important near-term operating metric is whether margins hold while the company scales. If adjusted operating margin stays around 8% while revenue accelerates, the market will increasingly treat Celestica as an AI infrastructure execution platform, not a low-margin assembler.

In 1 to 3 years, Celestica wants the business model to look less like outsourced manufacturing and more like embedded infrastructure partnership. The company’s best future is built around design-led manufacturing for hyperscaler AI systems, networking switches, rack-scale programs, and advanced optical interconnects.

The strategic bets are obvious. 800G is today’s ramp. 1.6T is the next wave. Co-packaged optics is the proof that Celestica can participate in the next architecture shift. Rack-scale AI systems are the bridge from component-level manufacturing to system-level infrastructure.

This widens the moat in several ways. Scale improves because more hyperscaler volume flows through Celestica. Switching costs rise because design and manufacturing become linked. IP/Reg improves modestly through engineering know-how, process capability, and qualification complexity. Brand improves because hyperscaler trust compounds quietly. Network effects remain weaker than in software, but there is a form of ecosystem gravity: suppliers, capacity, engineers, and customers increasingly coordinate around proven execution partners.

Everyone has AI exposure now. The real moat is being trusted with the messy physical layer of AI: switches, racks, components, supply constraints, thermal problems, qualification cycles, and production ramps. The glamour is in the model. The money may be on the factory floor.

Verdict: Buy.

WIMM for companies: If you would like to offer the newsletter as a benefit to other employees in your company, please contact me directly. ([email protected])

Work with me: If you are interested in working with me on your personal portfolio, again, contact me. I’m a certified financial consultant.

Finally, “Where is my moat?” is designed for individual readers, though the occasional forward is absolutely fine.

Thanks for your support & have a wonderful day!