AMD is reportedly establishing two new research and development centers in Taiwan, with a strong focus on silicon photonics, heterogeneous integration, and AI-related technologies, according to a report by the Liberty Times. These areas—particularly silicon photonics and system-in-package (SiP) heterogeneous integration—are seen as essential to enabling the next generation of AI and high-performance computing (HPC) components, as traditional performance scaling can no longer meet market demands.

Taiwan's Ministry of Economic Affairs (MOEA) confirmed that the new facilities will be located in Tainan and Kaohsiung, though AMD has yet to make an official announcement. The Kaohsiung R&D hub is expected to collaborate with National Sun Yat-sen University and other academic and industrial partners to advance research and talent development in silicon photonics and heterogeneous integration for AI applications.

The total investment in the two R&D centers reportedly exceeds $280 million (NT$8.64 billion), with AMD contributing about $170 million (NT$5.33 billion) and the remaining $110 million (NT$3.31 billion) supported by government funding.

Silicon Photonics: The Next Frontier for AMD

Optical interconnects offer higher data speeds, longer transmission distances, and superior energy efficiency compared to traditional copper connections, making them ideal for AI and HPC data centers. By integrating optical and electronic functions on a single platform, silicon photonics allows for direct optical interconnects without bulky adapters, reducing latency and power consumption.

Following the footsteps of Intel and NVIDIA, AMD has been actively moving toward incorporating silicon photonics into its product roadmap. The new Taiwan-based R&D centers are expected to accelerate its development of co-packaged optics (CPO) and related technologies.

Earlier this year, AMD acquired Enosemi, a California-based photonics design firm that had been one of its key development partners. The acquisition brought in a team with deep expertise in combining light-based communication with conventional silicon, as well as a portfolio of IPs and 16×112G optical chiplets for single-chip transmit and receive (Tx/Rx) functions. This move gives AMD greater control over its optical interconnect architecture, reducing dependency on external vendors and enabling tighter integration of CPUs and GPUs within large-scale AI data centers.

Toward Vertically Integrated AI Systems

With its silicon photonics and optical interconnect capabilities enhanced by Enosemi, AMD is expected to pursue vertically integrated rack-scale AI platforms—similar to NVIDIA's NVL72 and NVL144 systems—initially leveraging copper-based connections and later transitioning to optical links.

In the longer term, AMD's strategy could involve extending optical communication between chips to significantly improve performance and power efficiency, enabling the seamless connection of millions of processors across a single data center.

The addition of Enosemi's technology complements AMD's growing data center portfolio, which already spans CPUs, GPUs, DPUs, FPGAs, network solutions, and rack-scale integration through its partnership with ZT Systems. With specialized talent, proprietary IP, and early production capabilities now in place, AMD is positioned to accelerate the commercialization of silicon photonics and co-packaged optics solutions for next-generation AI systems.