SiGe Capacity Expansion Accelerates as AI Data Centres Shift to Optical Networking

GlobalFoundries expands SiGe capacity as AI data centres shift toward optical networking.
0
SiGe Capacity Expansion Accelerates as AI Data Centres Shift to Optical Networking
GlobalFoundries

SiGe capacity is becoming a critical bottleneck as artificial intelligence data centres move from copper interconnects toward optical networking. GlobalFoundries is expanding silicon germanium capacity to meet rising demand from customers building high-speed optical connectivity systems.

SiGe capacity at GlobalFoundries’ Vermont fabrication plant is already oversubscribed into 2027. The company said demand has reached an inflection point as AI workloads drive higher bandwidth, power efficiency and data transmission requirements.

SiGe capacity matters because silicon germanium technology supports key components in optical transceivers. These devices allow data centres to move information faster, cleaner and with less energy loss across large computing clusters.

The expansion reflects a wider shift in semiconductor materials. AI infrastructure is no longer only a story about graphics processors and advanced logic chips. It increasingly depends on photonics, optical transceivers and specialty semiconductor materials such as germanium.

Optical Networking Pushes SiGe Into a Strategic Role

AI data centres are increasing compute density and power consumption, forcing operators to rethink how data moves across networks. Copper interconnects face limits in reach, bandwidth density and energy efficiency.

Optical networking addresses those limits. It enables higher-speed data transfer across longer distances while improving system efficiency.

This shift is lifting demand for silicon photonics and SiGe technology. These materials are used in pluggable optical transceivers that convert high-speed electrical signals into optical signals and back again.

GlobalFoundries said SiGe is used in limiting amplifiers, transimpedance amplifiers and laser drivers. These components support signal amplification, conversion and cleaner data transmission inside data centre networks.

Transimpedance amplifiers and drivers are required in most data centre connections. As optical networking deployments grow, unit demand for these components is expected to increase sharply.

Satellite communications are also increasing SiGe usage. However, AI data centre optical networking is now the main growth signal attracting market attention.

Customer Prepayments and Government Support Shape Expansion

GlobalFoundries expects to double silicon photonics revenue in 2026. The company is targeting a silicon photonics revenue run rate above $1bn by the end of 2028 and $2bn in 2030.

GF already operates 300mm and 200mm silicon photonics and SiGe manufacturing facilities in New York and Singapore, with additional capacity in Germany. It also has a major US footprint in Vermont and New York.

The company increased annual wafer capacity to 2.7mn 300mm wafer equivalents in 2025 from 2.2mn in 2020. It also has an installed base of 1.6mn 200mm wafers per year.

GlobalFoundries has previously announced plans to invest more than $12bn in its New York and Vermont sites over the next decade. But future capacity growth will be tied closely to customer demand, prepayments and government financing.

That model is important. Semiconductor capacity expansion is capital-intensive, and customers increasingly need to help secure the supply chains they depend on.

Government grants and tax incentives are also becoming essential. AI, photonics, semiconductors and critical materials are now treated as strategic infrastructure, not only commercial technology.

For materials markets, the signal is clear. Germanium demand could gain support from AI-driven optical networking, especially as silicon photonics and SiGe devices become more important to data centre performance.

The Metalnomist Commentary

GlobalFoundries’ SiGe expansion shows that AI supply chains are moving deeper into specialty semiconductor materials. The next bottleneck may not be compute chips alone, but the optical and germanium-linked technologies needed to connect them efficiently.

 

No comments

Post a Comment