Celestica Q2 Preview: The 1.6T Supercycle Meets The Overbuild Dilemma

The setup is less about near-term revenue and more about how quickly a highly valued, utilization-sensitive manufacturer can go from scarcity premium to cyclical multiple. At ~31x 2026 EPS and ~71x free cash flow, CLS is being priced as if AI networking demand can compound smoothly through the next spending cycle; that leaves the stock vulnerable to even a modest pause in hyperscaler ordering because cash conversion typically lags the point where build rates slow.

The first-order winners from an AI hardware digestion phase are not necessarily the big cloud names, but better-diversified EMS peers with lower concentration and less stretched multiples, such as JBL and SANM, which can absorb a mix shift away from AI-specific programs. Second-order losers would be the broader component stack — PCB, connector, optical, and logistics suppliers — because order pullbacks usually show up there before they are visible in hyperscaler capex disclosures. If overcapacity emerges, pricing pressure tends to hit assembly/service providers first, then works upstream.

The market may be underestimating how quickly sentiment can turn once one or two hyperscaler capex comments normalize. The key falsifier is not general AI demand, but whether CLS can keep raising guides and sustaining margin/FCF conversion through the next 1-2 earnings prints; if that continues, the valuation can persist longer than skeptics expect. Absent that, the risk/reward favors a de-rating trade rather than a fundamental collapse, with the sharpest move likely over the next 1-3 months if capex digestion becomes visible.