Press Releases

Two commercially available 102.4 Tbps switches give AI operators a staged path from 800G fabrics to native 1.6TbE—without replacing every server endpoint at once.

At AI Infra Summit 2026, NADDOD will showcase two commercially available 102.4 Tbps Ethernet switches for different stages of AI network growth. The N9620-128OC and N9600-64PS give operators a practical way to add bandwidth today while preparing the switching layer for native 1.6TbE.

The two-switch portfolio is designed around two concurrent requirements facing AI infrastructure operators:

● Existing server endpoints equipped with 800G NICs need more fabric bandwidth.

● Next-generation networks need to begin adopting 1.6TbE switching without requiring all 800G NICs to be replaced at once.

NADDOD therefore positions 800G fabric expansion and 1.6TbE migration as successive stages of a continuous upgrade path.

Path 1: Expand the Existing 800G Fabric
As AI clusters add more GPU nodes, both server-access links and switch uplinks must carry increasingly dense east-west traffic. The N9620-128OC provides 128 × 800GbE OSFP ports and 102.4 Tbps of switching capacity based on the Broadcom Tomahawk 6 BCM78910. It supports NDD OS and ONIE (Open Network Install Environment).

The high port density enables more 800G connections and higher-capacity uplinks per switch, creating design headroom to reduce oversubscription and implement multi-rail, multi-plane network topologies. For clusters already running 800G NICs, the N9620 delivers increased link density and aggregate fabric bandwidth—integrating existing server endpoints into a more scalable network architecture.

Path 2: Move the Switching Layer to Native 1.6TbE
The N9600-64PS is the core of NADDOD’s 1.6T networking infrastructure showcase. The 4RU switch is based on NVIDIA Spectrum-6 and provides 64 × 1.6TbE OSFP224 ports, two 100GbE QSFP112 ports, and 102.4 Tbps of switching capacity. It supports Enterprise SONiC by NADDOD and ONIE for use in AI scale-out fabrics.

The N9600-64PS is more than just its 1.6TbE ports in the broader AI networking landscape. Built on NVIDIA Spectrum-6, the switch features 200G-per-lane electrical signaling—the same generation as the ConnectX-9 used in the NVIDIA Vera Rubin platform. This architecture delivers higher per-port bandwidth for next-generation AI platforms while maintaining seamless connectivity to existing 800G endpoints through link adaptation—paving a clear path toward native 1.6TbE adoption. To support this evolution, NADDOD is expanding its portfolio centered on the N9600-64PS—spanning 1.6T optical transceivers, 800G endpoint-adaptation modules, and high-speed cable assemblies. Together, the switch and interconnect portfolio delivers a complete end-to-end solution for next-generation data-center networks.

For current server environments based on 800G NICs, the showcase features a solution that links a single 1.6T port on the N9600 to two existing ConnectX-8 C8180 SuperNICs. On the switch side, an OSFP-1.6T-2xDR4 transceiver, equipped with dual MPO-12/APC interfaces, delivers two independent 800G DR4 optical signals. The two fiber links connect to separate O112-800G-DR4 transceivers installed on the ConnectX-8 adapters. At each endpoint, the O112 bridges four 200G PAM4 optical lanes to the NIC's native 8×100G PAM4 electrical interface.

This solution enables a seamless connection between a native 1.6T switching layer and existing 800G server endpoints, allowing customers to upgrade the switching infrastructure while preserving their existing NIC investments. Moreover, it underscores NADDOD's comprehensive 1.6T portfolio—spanning switch ports, optical transceivers, cabling infrastructure, and adapter-side integration—rather than being limited to the switch alone.

NADDOD Supports Deployment with Full-Stack Solutions and Platform Compatibility
NADDOD offers the optical transceivers and cabling solutions purpose-built for the featured switches showcased at AI Infra Summit. The OSFP-1.6T-2xDR4 is offered in variants for NVIDIA/Mellanox, Arista, Cisco, and standard Ethernet environments, and the O112-800G-DR4 is tailored for 800G ConnectX-8 C8180 and ConnectX-9 C9180 interfaces. Beyond hardware, NADDOD streamlines the entire deployment cycle from product selection, BOM preparation, and compatibility validation, to cable management and multi-plane patching, helping customers shorten project lead times and accelerate time-to-deployment.

The broader Ethernet portfolio also includes 1.6T optical transceivers, adapter modules for 800G and 400G endpoints, direct-attach and breakout DAC/ACC/AEC cables, and MPO/LC fiber infrastructure. NADDOD unifies BOM planning, port-mode selection, cable configurations, and link validation based on the target switch, NIC, and network topology, reducing multi-vendor coordination overhead.

With NADDOD's AI Infrastructure Deployment Services, engagements extend from network architecture and BOM planning through configuration, testing, and final delivery. For clusters at the 100,000 gpus scale, NADDOD enables efficient deployment and scaling of AI fabrics built on multi-plane network architectures. By integrating switches, interconnect products, and implementation services under a unified technical framework, NADDOD helps customers turn their product portfolio into a production-ready AI fabric.

A Parallel Path for InfiniBand XDR
NADDOD's 1.6T roadmap extends beyond Ethernet. InfiniBand remains a critical interconnect choice for large-scale AI training and high-performance computing, and XDR also advances the physical layer to 200G/lane. For InfiniBand environments, NADDOD offers a dedicated XDR product portfolio, including 1.6T and 800G optical transceivers, as well as direct-attach and breakout DAC, ACC, and AEC cables for Quantum-X800 and ConnectX-8.

NADDOD’s 1.6T strategy follows two parallel tracks: Ethernet and InfiniBand. Each track is built around the distinct protocol and platform requirements of its ecosystem, combining the appropriate switching platforms, optical transceivers, high-speed cables, and deployment support. This approach enables NADDOD to support both Ethernet-based RoCE scale-out fabrics and InfiniBand networks used for large-scale AI training and HPC.

Connect with NADDOD at AI Infra Summit 2026
At AI Infra Summit 2026, NADDOD will discuss practical approaches to scaling 800G fabrics and planning the transition to 1.6T with AI infrastructure architects, network engineers, systems integrators, and technology partners. The showcase will feature the N9620-128OC and N9600-64PS switches, 1.6T-to-dual-800G optical links for NVIDIA ConnectX-8, and Ethernet and InfiniBand interconnect solutions. NADDOD will also present its end-to-end project support, spanning solution design, deployment, and handover.

About NADDOD
NADDOD provides high-speed networking products and deployment services for AI data centers, cloud infrastructure, and high-performance computing. Its portfolio covers Ethernet and InfiniBand interconnects, AI data center switches, optical transceivers, DAC/ACC/AEC high-speed cables, fiber infrastructure, compatibility validation, and AI infrastructure deployment support. Learn more at: https://www.naddod.com/