The race to power generative AI and next-generation data centers is being won at the optical layer. As switch silicon leaps from 51.2 Tbps to 102.4 Tbps with the arrival of Broadcom's Tomahawk 6, the transceivers that feed those ports have to keep pace — and the new 1.6 Terabit per second OSFP transceiver is exactly the component the industry has been waiting for.

Our System Engineer at the EPS Global Switch Lab in Dublin, Welly, took a closer look at this state-of-the-art module in the latest episode of our Unboxing Open Networking series.

A True 1.6T Pluggable — And Four LC Connectors

The headline figure speaks for itself: 1.6 Tbps of bandwidth in a single OSFP pluggable. But what really sets this module apart is how that bandwidth is delivered to the fiber plant. Rather than relying on a traditional MPO trunk, this transceiver uses four LC connectors — a first-of-its-kind engineering approach for a module at this speed.

For network engineers, that means simpler, field-friendly terminations, easier troubleshooting, and a cleaner migration path from existing duplex LC infrastructure into 1.6T speeds. It's a small detail with a big operational impact in real-world data center deployments.

Reach: Up to 2 km Over Single-Mode Fiber

The transceiver is built for the distances that modern AI clusters and data center interconnects actually require. With a reach of up to 2 kilometers over single-mode fiber, it comfortably covers:

• AI/ML training clusters spanning multiple halls
• Hyperscale data center fabrics
• Campus-style data center interconnect between buildings on the same site

This longer reach gives architects far more flexibility in how they distribute compute, storage, and switching across a facility — critical when GPU clusters are growing faster than any single building can house them.

Full CMIS Compliance and DDM Visibility

The module is fully compliant with CMIS (Common Management Interface Specification) and exposes all of its operational data through DDM (Digital Diagnostic Monitoring) directly to the host switch. That means real-time visibility into:

• Optical transmit and receive power
• Temperature and voltage
• Lane-level diagnostics

For NOC and operations teams running open NOS platforms like SONiC or IP Infusion's OcNOS, that level of telemetry is essential for keeping AI fabrics healthy and predictable.

Built for Both Ethernet and InfiniBand AI Fabrics

Here is where the transceiver really earns its place in modern AI deployments. It supports both Ethernet and InfiniBand — and that matters more than ever as AI cluster architects debate which fabric is right for them.

As Welly explains, AI workloads push millions of packets per second between GPUs. While Ethernet continues to evolve rapidly to handle these flows, InfiniBand has long been the established choice for ultra-low-latency, lossless GPU-to-GPU communication. With this module, customers get a single transceiver that supports:

• 800 Gbps and 1.6 Tbps speeds for Ethernet AI fabrics
• 400G, 800G, and beyond for InfiniBand deployments

That dual-protocol flexibility future-proofs the optical investment — whichever fabric direction a customer chooses, the transceiver is ready.

Key Specifications at a Glance

• Form factor: OSFP, standard
• Data rate: 1.6 Tbps
• Connector: 4x LC (industry first at this speed)
• Reach: Up to 2 km over single-mode fiber
• Management: Full CMIS support, DDM-enabled
• Operating temperature: 0°C to 70°C (commercial grade)
• Protocols: Ethernet and InfiniBand

The 1.6T Era Is Here

This transceiver is one of the practical building blocks of the 1.6T era. Paired with Broadcom Tomahawk 6-based switches delivering 102.4 Tbps of switching capacity, modules like this complete the picture: 1.6T optics, 1.6T silicon, and an open ecosystem of switches and NOS platforms ready to deploy them.

For organizations building or upgrading AI clusters, GPU fabrics, or hyperscale data centers, the message is simple — the components needed to scale beyond 800G are no longer roadmap items. They are here, they are shipping, and they are being tested in our lab today.

Need a Custom Solution?

EPS Global specializes in open, disaggregated networking — from switches and NOS platforms to the optical transceivers that tie them all together. If you are designing your next AI cluster or planning a 1.6T upgrade, our engineers can help you choose the right optics, validate compatibility, and get hardware in your hands faster.

Get in touch with us today!