Optical fiber TAM growth is driven by the multiplicative effect of more AI data centers × more fiber per data center × faster link speeds requiring more parallel strands. The base case reflects hyperscaler fiber procurement continuing at current pace through 2028 before moderating as Corning and Prysmian bring new capacity online. Bear case assumes CPO deployment accelerates faster than expected, reducing intra-data-center fiber demand from 2028. Bull case assumes Stargate and other mega-cluster builds require fiber at a scale that strains even expanded Corning capacity through 2030, sustaining pricing power and TAM above the base.
“AI-optimized data centers require 16–36× more fiber than conventional cloud facilities. Meta signed a $6B supply agreement with Corning in January 2026; Amazon struck a separate multibillion-dollar Corning deal in June 2026. Corning disclosed two additional unnamed hyperscaler deals in Q1 2026 earnings — total fiber commitments from hyperscalers in H1 2026 exceeded $12B.”
“800G transceiver production running 40–60% below demand through 2027 (McKinsey). TrendForce and LightCounting flag fiber and transceiver demand 30%+ above supply. Fiber cable lead times extended to ~12 months. NVIDIA locked up Lumentum and Coherent EML laser supply ($4B commitment), further constraining the transceiver supply chain for non-NVIDIA buyers.”
“Corning Q1 2026 revenue beat driven by Data Center segment growing 40%+ Y/Y. Management raised full-year guidance citing record optical cable backlog. Corning controls ~45% of global single-mode fiber market and has pricing power in a supply-constrained environment — ASPs increasing for the first time in a decade.”
“Each AI data center speed transition (400G → 800G → 1.6T) requires not just faster transceivers but more fiber strands per link — parallel optics architectures use 8–16 fibers per 800G link vs. 2 fibers for 100G. AI clusters are driving a structural step-up in fiber density per data center, compounding the raw demand growth from more facilities.”
“Silicon photonics CPO integration could reduce fiber consumption inside data centers if implemented at scale — on-chip optical signaling eliminates some fiber runs between racks. However, CPO is still 2–3 years from volume deployment; near-term fiber demand is unaffected. DCI (data center interconnect) fiber between campuses is CPO-immune and growing independently.”
Meta ($6B), Amazon (undisclosed billions), and two unnamed hyperscalers all signed long-term Corning supply agreements in H1 2026. This mirrors what NVIDIA did with EML lasers — strategic actors securing physical supply before the shortage becomes acute. Corning's fiber capacity is essentially pre-sold through 2028. Any buyer without a signed contract faces spot-market pricing at 30–50% premiums and 12-month lead times.
Meta deal: $6B over ~3 years = $2B/yr fiber commitment
Amazon deal: estimated $3–5B over 3 years
+ 2 unnamed hyperscaler deals (similar scale): ~$4–8B
Total hyperscaler fiber commitments H1 2026: ~$13–20B over 3 years
Less existing fiber baseline spend: ~$8–15B net incremental TAM
→ Corning capacity pre-sold; spot market at 30–50% premium
Every AI cluster speed upgrade requires not only faster transceivers but more fiber strands per link. 800G parallel optics uses 8–16 fibers per link (vs. 2 for 100G duplex). As AI clusters upgrade from 400G to 800G to 1.6T, fiber strand count per data center increases 4–8× even with constant rack count. This multiplier compounds raw demand growth from new data center construction.
100G duplex: 2 fibers/link
800G parallel optics: 8–16 fibers/link (4–8× increase per link)
AI cluster with 50K GPU nodes: ~200K inter-rack 800G links
200K links × 16 fibers × 100m avg run = 320M fiber-meters per cluster
vs. 100G equiv: 200K × 2 × 100m = 40M fiber-meters (8× more)
→ Each AI cluster upgrade to 800G = 8× more fiber demand
Beyond intra-data-center fiber, AI mega-clusters require massive data center interconnect (DCI) fiber between campuses for model synchronization, data replication, and inference load balancing. Stargate's Texas cluster needs fiber to West Coast, East Coast, and international AI hubs. DCI fiber is immune to CPO disruption — it always requires physical fiber regardless of what happens inside the data center. New AI corridors (Texas to Virginia, Memphis to Phoenix) are driving fiber conduit construction ahead of cable installation.
New AI corridor fiber (TX–VA, MEM–PHX etc): ~5,000 route-km each
5,000km × 288 fibers/cable × $8/fiber-meter = $11.5B per route
3–4 major new AI corridors by 2028: ~$10–15B in dark fiber
Less existing long-haul infrastructure reuse: ~$4–8B net new fiber
→ Prysmian, Sumitomo benefit on long-haul; Corning on campus/campus DCI