The AI Trade Is Rotating From Memory To Interconnects, And Corning Is The Best Way To Play It

Corning's Q2 2026 EPS rose 30% year over year, while Springboard operating margins exceeded 20%, achieving the target one year early. The company is positioned to benefit from hyperscalers' increasing demand for high-density fiber and cluster-scale AI optical interconnects. Multi-year, customer-funded supply agreements with Meta, NVIDIA, Verizon and others provide capital-allocation support and revenue visibility through 2030.

Analysis

GLW is transitioning from a cyclical display/telecom materials valuation framework toward an AI-capex infrastructure multiple, but the investable issue is whether optical content per GPU cluster rises faster than hyperscaler procurement pricing compresses. The strategic scarcity is not commodity fiber alone; it is qualified, high-density cable/connector assembly capacity with deployment-grade reliability. That favors GLW over more component-exposed optical peers such as LITE and COHR, whose economics remain more sensitive to transceiver pricing and inventory cycles.

The second-order beneficiary is META: proprietary cluster architectures that require materially more optical connectivity can improve usable GPU utilization and reduce the effective cost of training/inference capacity. Conversely, NVDA faces a mixed effect over 6-18 months—networking attach revenue benefits, but rising cluster-level interconnect cost strengthens the incentive for customers to pursue custom silicon, photonics integration, and alternative fabric architectures. VZ is a lower-beta read-through; enterprise/edge fiber demand is additive but unlikely to move consolidated earnings absent a sustained acceleration in fiber subscriber or wholesale monetization.

Near term, GLW's risk is that investors capitalize contracted demand as if it were irrevocable volume. Customer-funded capacity reduces capex risk but does not eliminate timing, acceptance, pricing, or concentration risk; the key verification points are backlog conversion, prepayment/deposit treatment, incremental gross margin, and whether AI-related sales displace lower-return legacy capacity. A hyperscaler capex pause would likely hit the narrative multiple before it materially affects reported revenue, creating a 1-3 month drawdown risk even if the 2030 demand case remains intact.

Consensus may underappreciate that the bottleneck can migrate rapidly from fiber availability to installation labor, connectorization, switch-port density, and power/cooling constraints. If deployment bottlenecks dominate, GLW's shipment growth could lag announced AI capex despite healthy orders; ANET and CIEN are useful cross-checks because improving optical demand should ultimately be accompanied by resilient network-system order trends.

Key Decisions for Investors

• Initiate a 6-12 month long GLW position only on confirmation that AI/enterprise optical revenue is growing faster than total company sales and incremental operating margin remains above 20%; target a 15-20% upside on sustained multiple re-rating, with a 8-10% stop or exit on margin slippage and weaker backlog conversion.
• Express relative value as long GLW / short a basket of LITE and COHR over 6 months. GLW's customer-funded capacity and system-level qualification should produce less volatile gross margins; cover the short leg if optical-component pricing stabilizes materially or either peer reports a sharp inventory normalization-led margin recovery.
• Use META as the preferred hyperscaler exposure rather than NVDA for the cluster-interconnect theme over 6-18 months: better fabric efficiency can raise returns on AI capex, while NVDA's upside requires networking attach to outrun a potential shift toward customer-designed interconnect. Reassess after the next capex guide and disclosed AI infrastructure depreciation trajectory.
• Set an earnings watch item—not a trade trigger—for GLW: require disclosure of customer concentration, cancellation/deferral protections, and capex-to-prepayment coverage. Absence of these details would argue that current visibility is being overstated and warrants reducing exposure into results.