Wall Street may be sleeping on this networking stock tied to Google and OpenAI

Bernstein says a data-center "bandwidth wall" is creating a multibillion-dollar investment runway for high-speed networking and optical hardware, extending AI spending beneficiaries beyond chipmakers. Analyst Daniel Zhu identified Celestica as a networking stock positioned to benefit from demand tied to major AI customers including Google and OpenAI. The thesis is constructive for AI infrastructure suppliers, though the article provides no company-specific financial estimates or target-price changes.

Analysis

CLS is levered to the less-appreciated shift from compute-centric AI capex toward cluster-level throughput: once accelerator utilization is constrained by interconnect congestion, customers prioritize networking spend even if GPU deliveries remain supply-limited. The key earnings sensitivity is not merely higher revenue, but mix: systems integration and rack-level deployment can carry structurally better dollar profit per deployed AI cluster than lower-complexity manufacturing programs. A sustained step-up in hyperscaler 800G/1.6T buildouts could therefore support estimate revisions and a multiple re-rating over the next 2-4 quarters.

The more attractive relative expression is likely CLS versus hardware vendors with heavy exposure to enterprise IT refresh cycles, where AI spend may be offsetting rather than incremental. GOOG's benefit is indirect: more efficient internal networking lowers the effective cost of serving inference and improves the return on AI infrastructure, but the financial impact is too diluted to drive the stock absent evidence of improving Cloud margins or AI monetization. Second-order beneficiaries include ANET, CIEN, COHR and LITE, while optical-component supply tightness could shift value toward qualified manufacturers and away from commoditized server assemblers.

Consensus may be underestimating the duration of the networking upgrade cycle, but it may also be over-crediting every AI-adjacent supplier with hyperscaler economics. The thesis fails if accelerator utilization does not improve after network upgrades, if customers adopt lower-cost scale-up architectures, or if a concentrated customer delays cluster deployments; CLS's customer/program concentration makes a single-quarter guide-down materially more damaging than for diversified networking OEMs. Near-term confirmation should come from order-book conversion, AI-program mix, and management commentary on 800G/1.6T deployment rather than broad AI capex headlines.

Key Decisions for Investors

• Initiate a 3-6 month long CLS position only on confirmation that AI-program revenue and backlog conversion are accelerating; target a 15-25% upside from upward EPS/multiple revisions, with a 8-10% stop or exit on a material customer-program delay.
• Express the infrastructure bottleneck through a basket: long CLS and ANET, with smaller satellite exposure to CIEN/COHR; size CLS below ANET because concentrated manufacturing/customer exposure creates larger gap risk around earnings.
• Pair trade for the next 1-3 months: long CLS / short a broad enterprise-hardware proxy such as HPE, provided relative valuation does not already imply an extreme premium. The intended payoff is AI-networking capex remaining incremental while conventional enterprise spending stays budget-constrained.
• Do not use GOOG as the primary implementation. Reassess only if Google Cloud margin expansion or disclosed AI-infrastructure efficiency metrics demonstrate that networking investment is translating into monetizable inference economics.
• Set an earnings watch item: reduce or avoid CLS if management signals slower hyperscaler deployment, declining AI mix, or weaker backlog conversion; these are more thesis-relevant than aggregate data-center capex guidance.