Earnings Call Highlights: Celestica (CLS) | Q2 FY2026 — Revenue Surges 62%, OpenAI Rack Orders Emerge, and an EMS Company Becomes an AI Rack System Partner

2 days ago

7 min read

This quarter Celestica threw its "contract manufacturer" hat even further away. Revenue up 62% YoY, adjusted EPS up 83%, and adjusted operating margin at a company-record 8.2%—all three numbers blew through the top of guidance, and management raised its full-year outlook for the second time. But the real signal isn't this quarter; it's 2027. OpenAI's custom rack orders have officially surfaced, the AMD Helios scale-up switch is a pipeline "worth billions of dollars and still growing," and 1.6T starts ramping in the second half. This is no longer an earnings report to read for gross margin; it is a positioning chart of "who can reliably mass-produce AI racks in a world short of components."

*This article is for technology and industry trend analysis only and does not constitute investment advice.*

1. Three Conclusions to Remember First

First, this quarter was a broad beat plus another raise, not a one-off surprise. Revenue, EPS, and operating margin all broke through the top of guidance, and both CCS end markets as well as ATS beat expectations. Management then raised full-year revenue guidance to $20.5B (up 65% YoY) and adjusted EPS to $11.30. When a company raises guidance for the second time by mid-year, the visibility usually comes from materials already ordered and contracts already signed, not from optimism.

Second, the growth engine is shifting from selling boards to selling systems. HPS (Hardware Platform Solutions, covering Celestica's own-design networking and compute systems) reached $1.9B in the quarter, up 58% YoY and 41% of total revenue. This business carries richer margins than pure contract manufacturing and is the key reason CCS margins moved up this quarter.

Third, half of the 2027 script is already written. Celestica confirmed for the first time that its "digital native customer" is OpenAI, and that it is supporting OpenAI's multi-generation custom accelerators together with Broadcom (the first, codenamed Jalapeno, enters volume production in 2027); the AMD Helios scale-up switch ramps in the first half of 2027 as well. Both are billion-dollar-class opportunities, and both are still growing.

2. Revenue and Financials: Every Line Says "Record High"

First, the hard numbers for the quarter (all on the company's comparable / adjusted non-GAAP basis, consistent with the earnings card):

• Revenue of $4.70B, up 62% YoY and about 16% QoQ, above the top of guidance.

• Adjusted operating margin of 8.2%, up 80 bps YoY and a company record—the single most important number this quarter.

• Adjusted gross margin of 11.5%, down about 0.2 percentage points YoY, mainly due to CCS product mix.

• Adjusted net income of $295.4M, adjusted EPS of $2.54, up 83% YoY (up $1.15).

• Adjusted ROIC of about 55%, up roughly 20 percentage points YoY—the result of operating leverage plus disciplined working capital.

By segment: CCS $3.81B (up 84% YoY, 81% of revenue), with Communications up 62% (ahead of the ~50% outlook, driven by the 800G switch ramp plus unexpectedly resilient 400G) and Enterprise up 167% (well above the ~130% outlook, driven by a hyperscaler pulling in an AI/ML compute program and by storage demand). ATS $888M (up 8% YoY, 19% of revenue), with growth across all businesses. Segment margins both rose: CCS 8.7% (+40 bps) and ATS 6.3% (+100 bps).

One easily overlooked but critical number: three customers each accounted for more than 10% of revenue this quarter, at 32%, 17%, and 14%—63% combined. That is the flip side of a hyper-growth phase, and Section 5 expands on it.

To understand why the same customers and the same capex can feed companies in such different segments as Celestica, Marvell, and Nokia at the same time, see our earlier piece The AI Capex Ceiling Gets Pushed Higher Again, but Optical Communications Is Starting to Diverge; this quarter Celestica sits squarely on the side that hasn't diverged and keeps benefiting.

3. Business Highlights: OpenAI Emerges, AMD Helios Expands, 1.6T Starts Ramping

The biggest news on this call wasn't in the financials but in Rob Mionis's business update.

OpenAI officially revealed. The customer management had vaguely referred to as a "digital native customer" for several quarters was named outright as OpenAI. Celestica's role is "strategic systems partner," supporting OpenAI's multi-generation custom accelerator roadmap together with OpenAI and Broadcom; the first accelerator, codenamed Jalapeno, enters volume production in 2027, and initial custom rack samples begin shipping by year-end. Management framed this as a long-term relationship "supporting OpenAI's multi-gigawatt deployments"—not a one-off order, but a roadmap.

AMD Helios is the other scale-up pillar. Celestica is the R&D design and manufacturing partner for the Helios scale-up networking switch, the interconnect backbone of the entire rack. Samples arrive at the end of 2026 and volume ramps in the first half of 2027; management called it an opportunity "worth billions of dollars, with a pipeline that is still growing," adding that "we are involved in all of these programs." For a full breakdown of why Helios writes open Ethernet into the rack backbone and what it really means for the optical supply chain, see AMD Writes "Open Ethernet" into the Rack Backbone.

1.6T moves from slideware to the production line. Management said it has 10 active 1.6T programs, entering volume production with two hyperscalers from Q3, while 800G will still grow strongly in 2027 and 400G has been "surprisingly resilient." In other words, Celestica's networking business isn't waiting for 1.6T to take the baton; 400G, 800G, and 1.6T are all growing side by side. For how this generational step redistributes BOM and margin, see 800G vs 1.6T Optical Modules: A BOM Teardown of Who Really Captures the Margin.

One more small but important signal: the 1.6T CPO program Celestica won samples in the first half of 2027 and enters volume production in the second half of 2027, and Celestica continues to develop next-generation technology with that customer. For the backstory on CPO going from "crying wolf" to an actual production schedule, see CPO Is Finally No Longer "Crying Wolf".

4. Management Outlook: A Second Raise, with Language of "Confirmation," Not "Hope"

Q3 FY2026 guidance: revenue of $5.25B–$5.55B (up 69% YoY at the midpoint), adjusted EPS of $2.88–$3.08 (up 89% at the midpoint), and an adjusted operating margin of 8.4%. By end market: Communications about +60% (800G stays strong and the first 1.6T volume production begins), Enterprise about +190% (continued AI/ML compute pull plus a storage recovery), and ATS growing in the mid-teens (driven by the capital equipment business).

Full-year FY2026 outlook (second raise): revenue raised to $20.5B (up 65% YoY, from $19.0B), adjusted EPS raised to $11.30 (from $10.15, up 87% YoY), adjusted operating margin of 8.4% (from 8.1%), and free cash flow raised to $600M (from $500M, including about $1 billion of CapEx).

The line that really deserves underlining is the qualitative guidance for 2027: management said explicitly that 2027 revenue growth will "accelerate even faster than 2026's 65%," and that because operating margin keeps expanding, EPS growth will "exceed revenue growth." Making a call in mid-year that next year's growth will accelerate is "confirmation"-grade language, backed by long-lead-time components that have already been ordered. It also explains why CFO Mandeep Chawla said they are now discussing capacity planning with customers for 2027, 2028, and even 2029—a snapshot of how this hyperscaler capex cycle is pulling visibility forward, and the other end of the same "capex plus data center interconnect" thread we saw in our breakdown of the Alphabet FY2026 Q2 Earnings Call.

5. Supply Chain and Customer Clues: What's Short Is Materials, Not Demand

Management repeatedly stressed one point this quarter: what limits growth is materials—not capacity and not demand. Celestica said its 2026 and 2027 capacity is already in place; the real bottleneck is components, with some lead times now beyond 52 weeks, so it has had to place 2027 material orders early. Seen the other way, it is precisely this shortage that has produced exceptional visibility: customers placing NCNR (non-cancelable, non-returnable) orders for long-lead silicon and signing supply agreements for hard-to-source parts such as memory and PWBs. Here the shortage isn't bad news; it is an endorsement that the demand is real.

Global capacity footprint: Thailand is the flagship site; Japan is expanding at customer request and taking on part of the switch business; and the U.S. is growing (Richardson, Texas is expanding, and Fort Worth is online and will expand further). Management said customers now want "equal growth in Asia and the U.S."—geographic diversification has become a customer demand, not something Celestica is pursuing on its own.

Customer concentration is the risk to watch. Three customers each exceed 10% of revenue, totaling 63%, and with OpenAI and AMD deals snowballing, any single customer's schedule or capital position will directly amplify swings in Celestica's results. The CFO was blunt: "We are not a bank"—Celestica won't finance its customers' growth, but it will protect both sides through contract terms (including 12–18-month binding forecasts and customer-funded dedicated equipment). That is necessary discipline in a hyper-growth phase.

6. Conclusion

This quarter Celestica staged a public identity upgrade. Record financials were only the price of admission; the real point is that it has pushed itself from "board supplier for AI racks" to "rack system partner for OpenAI and AMD," and used the logic of "shortage = high visibility" to lock 2027 growth into orders ahead of time. For anyone tracking the AI infrastructure supply chain, this is a rare chart that maps upstream demand to downstream assembly.

Three things to watch next quarter (and next year):

1. The 1.6T ramp slope: with 10 active programs and two hyperscalers entering volume production in Q3, whether actual shipments keep pace is the leading indicator for 2027 acceleration. 2. Revenue recognition cadence of the OpenAI/AMD deals: management has hinted that OpenAI will be consigned (materials not booked, revenue excluding materials), so the focus is the timing and scale of rack samples moving to volume production rather than the headline revenue number. 3. Customer concentration and materials: with the top three customers at 63% and component lead times above 52 weeks, any change in scheduling or shortages anywhere in the chain will be magnified by this growth multiple.

In one sentence: while the market is still arguing over whether AI capex has a ceiling, Celestica has already ordered next year's materials and said OpenAI's name out loud.