Silicon Photonics Capacity Expands as AI Data Centre Demand Accelerates

Silicon Photonics

Silicon photonics capacity is expanding rapidly as artificial intelligence data centres require faster, higher-bandwidth and lower-latency connections between processors. Semiconductor firms are increasing investment as SiPh technology moves from telecoms and enterprise networks into hyperscale AI infrastructure.

The shift reflects a structural change in data centre architecture. Large AI clusters need optical interconnects that can move massive volumes of data with lower energy consumption than traditional copper-based systems.

Silicon photonics capacity growth is therefore becoming a materials story as well as a semiconductor story. While silicon wafers form the base platform, high-performance SiPh components also rely on indium phosphide, gallium arsenide, gallium nitride, germanium compounds and, in some cases, lithium niobate.

AI Workloads Push Optical Interconnects Beyond Copper

AI workloads are increasing data centre scale and network complexity, forcing hyperscalers to adopt technologies that reduce latency and energy use per bit. This is accelerating the replacement of copper in high-speed data communications equipment.

Industry spending on data centre switches for AI back-end networks is now forecast to exceed $100 billion by 2030. The upgraded outlook reflects rising demand from agentic AI, physical AI, humanoid robots, autonomous vehicles and military systems.

This growth increases demand for optical transceivers, modulators, photodiodes, lasers and high-speed detection systems. These components are closely tied to minor metals and compound semiconductor materials that support faster optical transmission.

Tower and STMicroelectronics Scale SiPh Platforms

Tower Semiconductor is working with Coherent on high-speed data transmission using a silicon modulator built through a production-ready SiPh process. The work targets next-generation optical transceivers for AI data centre applications.

Tower is also investing heavily in silicon germanium products that support its SiPh platform. The company raised its 2026 capital expenditure budget for SiPh and silicon germanium by $270 million, in addition to the $650 million announced in late November.

STMicroelectronics is also expanding aggressively. The company plans to quadruple SiPh production capacity by 2027, supported by long-term capacity reservation commitments from customers.

STMicro recently entered high-volume production for its SiPh-based PIC100 platform. The platform is used by hyperscalers for optical interconnection in data centres and AI clusters, and it is designed to reduce signal loss while improving modulator, photodiode and chip-to-fibre performance.

The Metalnomist Commentary

Silicon photonics capacity is becoming a hidden bottleneck in AI infrastructure. As optical interconnects scale, demand for indium, gallium, germanium and lithium niobate-linked materials will become more strategically important to the semiconductor supply chain.