Celestica: The Broadcom Scare Looks Overdone

Celestica is described as fundamentally strong despite a recent selloff tied to Broadcom's AI semiconductor miss. Ongoing Google TPU momentum is supporting AI server and switching ramps, while AMD's upcoming Helios rack-scale deployment adds another growth vector beyond Google. The note implies improving scale-up networking demand and a positive setup for Celestica shares.

Analysis

The selloff looks more like a sympathy de-rating than a fundamentals reset, which creates a useful dislocation in the hardware supply chain. The key second-order effect is that AI infrastructure spending is becoming more platform-diverse: if hyperscale compute is shifting from a single-node narrative to rack-scale and TPU-driven deployments, Celestica’s addressable content expands even if one customer’s near-term cadence is noisy. That matters because investors still tend to price ODM/EMS names as linear beneficiaries of one vendor’s cycle, when in reality the mix shift toward networking, power, and integration usually lifts margin intensity more than unit volumes.

The broader read-through is mixed for semis: weakness in one AI leader does not necessarily signal a demand air pocket, but it does raise the bar for companies selling the enabling stack. Broadcom’s move is more likely to compress expectations for near-term AI infrastructure multiples across the ecosystem than to alter the spending trajectory itself. Conversely, Google and AMD are signaling that the competitive battleground is moving to system-level deployment, which should benefit suppliers with high mechanical complexity and multi-program exposure rather than single-socket content.

The contrarian view is that the market may be underestimating how quickly rack-scale networking becomes the bottleneck, not the accelerator, of AI buildouts. If that happens, winners will be the companies with broad manufacturing, validation, and board-level integration capabilities, while pure-play accelerator suppliers may see more volatile order patterns. The main risk to the thesis is a multi-month digestion phase in AI capex after any hyperscaler pause, but the next catalyst window is still close—one to two quarters—because new platform ramps tend to re-rate suppliers before revenue fully inflects.

Key Decisions for Investors

• Add CLS on weakness over the next 1-3 weeks; use the recent selloff as an entry point for a 3-6 month long with asymmetry toward multiple recovery if AI infra orders remain intact.
• Pair trade: long CLS / short AVGO for 1-2 quarters. The idea is that supply-chain beneficiaries with multi-customer exposure can outperform a single-name AI disappointment if the market keeps penalizing perceived AI demand risk.
• Buy 3-6 month CLS call spreads rather than outright equity to express upside from TPU and rack-scale ramps while limiting exposure to a broader AI capex air pocket.
• Watch GOOGL and AMD as catalysts, not just beneficiaries: if either confirms volume ramp cadence in the next earnings cycle, add to CLS immediately; if guidance softens, reduce CLS and expect a 10-15% air pocket.
• Avoid chasing AVGO on the first bounce; wait for a confirmation of demand stabilization before buying the dip, since sentiment overhang can persist 4-8 weeks even if fundamentals are merely slowing, not breaking.