www.steelldy-indices.com
This study analyzes the evolution of the fiber optics industry in the context of the explosion of data centers for artificial intelligence, covering the period 2026-2031. The central thesis we defend is as follows: the transition from copper to optics in data center infrastructure is an irreversible structural trend, based on fundamental physical constraints (thermal dissipation, large-scale signal loss), rather than a cyclical phenomenon. However, the current valuations of Corning (GLW), Coherent (COHR), and Lumentum (LITE) incorporate excessive optimism, and the risk of commoditization remains significant.
Dell’Oro Group data (2025) projects that the global market for optical data center equipment will reach $18.4 billion by 2029, representing a CAGR of 15.2%. LightCounting Market Research estimates that shipments of 800G and 1.6T transceivers will grow at a rate of 22% per year between 2026 and 2030. These projections are validated by the capex announcements of hyperscalers: Meta, Microsoft, Google, and Amazon have collectively announced over $300 billion in AI infrastructure investments for 2025-2027, with a growing share dedicated to internal optical interconnects within GPU clusters.
1. Transition Dynamics Copper-to-Fiber
1.1 Physical Constraints of High-Density Copper
Copper’s physical constraints in high-density environments make its replacement by optical fiber a physical necessity, not an aesthetic choice. Resistive loss follows Joule’s law (P = R × I²), where resistance increases with frequency due to the skin effect—at 100 GHz (needed for 800G interconnects), skin depth in copper is only 0.066 micrometers, concentrating current in a tiny fraction of the conductor’s cross-section and multiplying effective resistance. Shannon-Hartley theorem defines channel capacity (C = B × log2(1 + S/N)), where B is bandwidth and S/N signal-to-noise ratio. Single-mode optical fiber uses about 50 THz of bandwidth in the C-band (1530-1565 nm), theoretically enabling ~100 Tbps per fiber. In contrast, copper Direct Attach Cable (DAC) is practically limited to ~400 Gbps per channel over distances under 3 meters, making any GPU cluster architecture beyond a few dozen nodes physically impossible with copper alone.
1.2 GPU Cluster and CPO Architecture
The NVIDIA GB200 NVL72 architecture, which interconnects 72 Blackwell GPUs in a single rack, requires ultra-high-density links that exceed the capabilities of copper. Each GPU generates 1.8 Tbps of throughput, necessitating a total rack interconnection of around 130 Tbps. CPO (Co-Packaged Optics) technology, which integrates optical transceivers directly onto the processor package, represents the next frontier: it eliminates external electro-optical conversion, reducing energy consumption per bit by 50 to 70% compared to pluggable transceiver solutions. Intel, Broadcom, and NVIDIA are heavily investing in CPO, with mass commercialization expected between 2027 and 2029.
2. Theoretical Framework | Quantitative Models
The theoretical framework outlines quantitative models for analyzing optical technology adoption in data centers. The Bass diffusion model (1969) predicts adoption with parameters p=0.003 (weak innovation), q=0.42 (strong imitation due to hyperscaler networks), and M=12,000 units, targeting integral optical interconnects by 2031. Market volatility is modeled via G.(1,1) with calibrated parameters ω=0.0004, α=0.12, β=0.85, showing high volatility persistence (α+β=0.97). Three market regimes are identified using a HMM: bull (μ=0.18, σ=0.22), neutral (μ=0.04, σ=0.32), and bear (μ=-0.12, σ=0.45), with a 78% transition persistence. A K. filter estimates hidden demand for optical equipment from observable indicators like semiconductor orders and hyperscaler capex, with parameters F, H, Q, R calibrated via maximum likelihood. Bayesian inference updates regime probabilities per new data. A TVP-VAR model captures dynamic relationships between optical stock prices, copper futures, NASDAQ, and interest rates, with parameters evolving as a random walk. Portfolio optimization under regime constraints uses a quantum-classical hybrid approach, simulated on D-Wave Advantage, combining quantum exploration with classical refinement. Deep learning neural networks detect nonlinear patterns in price and sentiment data.
3. Analysis of the Three Key Actors
3.1 Corning Incorporated (GLW)
Corning, founded in 1851 in New York, is the global leader in optical fiber with approximately 50% market share in fiber preforms. The Optical Communications segment accounts for about 35% of the total revenue of $12.6 billion (FY2025). The overall gross margin is around 40%, with free cash flow generation exceeding $1.5 billion per year. Corning’s fundamental competitive advantage lies in its vertical integration: the company controls the entire value chain, from the manufacturing of ultra-pure silica glass (proprietary OVD process) to the production of cables and connectors. This integration enables economies of scale that are impossible for niche competitors to replicate.
