Corning's AI fiber growth puts capacity delivery, customer funding in focus

Corning's optical networking surge is not a cyclical bump—it is the architectural signal that AI clusters have hit the interconnect wall. Once GPU scale crosses ten thousand units, copper's physics ceiling is breached and fiber shifts from optional to a hard design constraint. This is structural, not seasonal. The technical cascade runs both directions: upstream, preform CVD tooling demand accelerates; downstream, CPO module roadmaps compress. By 2026, optics will consume 15%+ of a new AI data center's BOM, up from under 8%, fundamentally rewriting the interconnect cost structure. Competitively, Sumitomo and Fujikura hold three-year head starts in hollow-core fiber. If Corning remains anchored to legacy single-mode capacity, the 2027 CPO production window will expose a generational gap. Meanwhile, YOFC and Hengtong's cost advantage, combined with US friend-shoring procurement policies, is redrawing the geographic radius of hyperscaler sourcing. The critical risk is not demand shortfall but timing mismatch between locked-in capacity and customer capex cycles. In 2000, Corning held 70% of global fiber capacity and endured an 18-month destocking cycle. Today, top-five customer concentration above 60% amplifies that exposure. A 2025Q4–2026H1 AI infrastructure pause would convert committed capacity from asset to liability almost overnight.