Scaling AI Data Centers Through Optical Networking

• Copper wiring becomes impractical beyond 30 meters for AI data centers, while optical networking uses light pulses through fiber to transmit data with one-third the power consumption and minimal signal loss
• The networking hardware cost per AI chip cluster is projected to increase 29x from $315,000 today to $9.4 million by 2028, with major tech companies like Nvidia investing $4 billion in laser suppliers to secure optical components
• Supply chain bottlenecks center on indium phosphide wafers produced in Taiwan, Japan, and Germany, with Goldman Sachs predicting tight supply conditions lasting through 2027

Summary: Scaling AI Data Centers Through Optical Networking

Key Investment Opportunity

The optical networking market is experiencing explosive growth as AI data centers shift from copper to light-based connectivity solutions. The market is projected to expand from $14 billion to $73 billion by 2030, representing a 39% annual growth rate.

Core Thesis

Copper wiring cannot support AI's massive data demands beyond 30 meters due to signal degradation and excessive energy consumption. Optical networking—which transmits data via light pulses through fiber—offers superior performance: longer distances with minimal signal loss, one-third the power consumption, and the ability to send multiple data streams simultaneously using different light wavelengths.

Market Economics

The networking hardware required per AI chip cluster is expected to surge 29-fold from $315,000 today to $9.4 million by 2028. Goldman Sachs estimates the AI optical networking market alone will reach $75 billion by 2028. Major tech companies are already investing heavily—Nvidia committed $4 billion to secure laser supply, while Meta and Google are deploying next-generation optical technologies.

Supply Chain Dynamics

Supply constraints are expected to persist through 2027, with the entire production chain operating at capacity. Critical components include indium phosphide wafers (sourced from Taiwan, Japan, and Germany), laser chips, modules (assembled in China and Thailand), and specialized fiber from companies like Corning and Fujikura.

Investment Risks

Key concerns include potential capex reductions by cloud providers, geopolitical supply chain exposure, and historical precedent from the 1999 fiber bubble when infrastructure buildout exceeded demand. However, the fundamental requirement that every AI chip needs optical connections at scale supports the long-term growth narrative.