GlobalFoundries and Marvell Scale Up Silicon Photonics for AI Data Centers

The Tech TL;DR:

• The Hardware Shift: GlobalFoundries is increasing silicon germanium (SiGe) and silicon photonics manufacturing capacity at its Burlington, Vermont facility to build energy-efficient optical interconnects.
• The Architectural Driver: According to Marvell Chief Executive Matt Murphy speaking at Computex 2026, copper interconnect distance limits force rigid server partitioning, necessitating an industry-wide transition to optical solutions like the Colorz 1600 and Ara 1.6 Tb/s product families.
• Enterprise Impact: Hyperscale cloud providers can leverage these developments to pool compute and memory across thousands of kilometers, bypassing legacy physical boundaries in AI infrastructure.

Overcoming Copper Interconnect Limits in AI Data Centers

As artificial intelligence workloads swell beyond the physical limits of single server racks, data center architecture faces a hard latency and bandwidth wall. Traditional scale-up systems, such as Nvidia’s NVL72, rely heavily on copper cables where CPUs must sit in close physical proximity to memory banks. However, according to Marvell Chief Executive Matt Murphy, these rigid physical boundaries begin to fracture as cluster demands scale up. Workloads no longer fit neatly inside a single localized facility, pushing hyperscale cloud service providers to construct expansive multi-facility campuses linked by high-speed networks.

Copper wiring inherently struggles with distance attenuation at high frequencies. To eliminate these physical choke points, Marvell has detailed a vision for optically interconnected infrastructure that spans thousands of kilometers. This long-haul scale relies on advanced coherent optics, such as the upcoming Colorz 1600 1.6 Tb/s coherent optical solution built on a 2nm DSP, which targets inter-data-center connectivity and is slated to sample later this year. Concurrently, the firm is developing the Ara 1.6 Tb/s family of interconnect solutions using 3nm DSPs alongside the Teralynx T100 102.4 Tb/s Ethernet switch supporting up to 512 ports running at 200 Gb/s or 64 ports running at 1.6 Tb/s.

For systems operating at scale, integrating dedicated networking hardware requires robust verification. Enterprise deployment teams often lean on specialized infrastructure testing pipelines and low-latency systems engineering consultants to validate high-bandwidth optical fabrics prior to production rollouts.

Manufacturing Silicon Photonics at Scale in Vermont

To support this transition away from copper, foundry execution is critical. GlobalFoundries has ramped up its operational capabilities through an expanded partnership with Marvell, focusing on advanced silicon germanium (SiGe) technology. Production is scaling directly out of GlobalFoundries’ Burlington, Vermont plant, supported by its broader manufacturing footprint in Malta, New York, and Singapore.

According to technical specifications outlined by GlobalFoundries, their silicon photonics platform supports pluggable optical transceivers, linear drive pluggable optics (LPO), near packaged optics (NPO), and co-packaged optics (CPO). The process technology integrates photonics, RF components, and advanced packaging under one manufacturing umbrella. This includes support for the Optical Compute Interconnect Multi-Source Agreement (OCI MSA) via their SCALE CPO solution, alongside manufacturing features such as through-silicon vias (TSVs), sub-100 µm copper pads, and detachable fiber attachment methods.

Deploying electro-optical transceivers into production container clusters demands rigorous configuration management. Enterprises updating their network topologies frequently coordinate with enterprise network architects and dedicated DevOps automation agencies to manage automated provisioning scripts and maintain strict uptime.

Implementation: Querying Optical Switch Port Statuses

Managing high-density switches like the Teralynx T100 in an optical data center environment requires automated telemetry retrieval via modern network APIs or CLI tooling.

curl -X GET "https://api.internal-net-management.local/v1/switches/teralynx-t100/ports/telemetry" 
     -H "Authorization: Bearer ${API_TOKEN}" 
     -H "Accept: application/json" 
     -d '{"metrics": ["optical_power_dbm", "ber", "temperature_c"], "ports": [1, 64, 512]}'

Toward the Disaggregated Optical Data Center

The convergence of silicon photonics manufacturing and coherent long-haul optics signals a fundamental shift in enterprise hardware design. By replacing copper connections with optical links, hardware developers can completely disaggregate compute, memory, and accelerators (XPUs) into isolated resource pools. As Matt Murphy emphasized regarding future infrastructure, compute and memory will operate as a unified system where millions of resources work in tandem across vast distances. For engineering organizations scaling next-generation clusters, establishing partnerships with vetted silicon integration specialists remains essential to navigating this transition.