AI Data Centers Shift to Optics, Creating New Infrastructure Bottleneck

The article frames optical connectivity as the next critical bottleneck in AI infrastructure buildout. As AI innovation accelerates, the computational demands on data centers are forcing a technology shift: optical fiber and photonic systems offer higher bandwidth and lower power consumption than copper, which cannot keep pace. This creates a supply-chain opportunity across multiple layers—from the chips that convert electrical signals to light (optical DSPs like those from Marvell and Broadcom) through the light sources themselves (Lumentum's lasers) to the complete transceiver modules (Coherent, Applied Optoelectronics) and the physical routing and fiber backbone (Amphenol, Corning, Ciena).

The concentration of advantage in specialized players is notable. Lumentum's position as a near-monopoly supplier of high-power indium phosphide lasers—even for use in competitors' systems—highlights how vertical integration and manufacturing scale create defensible positions in hardware supply chains. Similarly, Coherent's end-to-end integration from raw substrate to receiver, and Applied Optoelectronics' in-house laser manufacturing, position them as indispensable partners to cloud providers. However, Applied Optoelectronics' customer concentration with Microsoft introduces risk, even as the company ramps capacity to meet demand. The article suggests that some players like Amphenol will benefit from the optics shift regardless, since they also supply components used in legacy Nvidia NVLink copper systems.