OcNOS-DC on Tomahawk 6 cuts latency and total cost of ownership while doubling bandwidth to 102.4 Tbps for next-generation AI and cloud networks

SANTA CLARA, Calif.: #Broadcom--IP Infusion, a global leader in open networking solutions, today announced plans to extend its OcNOS® network operating system to support open platforms based on the Broadcom Tomahawk 6®. Adding to the 40+ active open hardware platforms already supported, OcNOS will extend support to the Broadcom Tomahawk 6-based family, giving hyperscalers, neoclouds, and data center operators a unified network operating environment for Scale Up, Scale Out, and Scale Across networking deployments in the data center, simplifying operations and reducing operational costs.

Broadcom Tomahawk 6, the first 102.4 Tbps Ethernet switch silicon in production, doubles fabric bandwidth at the same 64-port native radix as TH5 by pushing per-port speed to 1.6T, while OcNOS-DC provides a consistent operating environment across supported platforms.

As AI networking clusters expand exponentially, the infrastructure and data center operators must engineer their networking fabrics to support that growth without expanding power and space footprints exponentially. Current state-of-the-art spine switches, based on 51.2 Tbps-class fabrics and 800 Gbps connectivity, are ill-equipped to do this. Adding additional similar-capacity switches increases latency and space and power consumption, creating the potential for congested links that can stall entire training jobs.

With the introduction of OcNOS support for Broadcom Tomahawk 6-based switches, hyperscalers, neoclouds, and data center operators are able to double the capacity of their network fabric, collapse the fabric into fewer tiers, and keep the fabric lossless as it scales-all while using a common open NOS and operational model for Scale Up, Scale Out, and Scale Across deployments.

Latency is the defining constraint in large-scale AI networking, and OcNOS-DC on Broadcom Tomahawk 6 platforms is designed to attack it at the architectural level. With 512 ports of 200G, a single Tomahawk 6 switch can connect up to 512 accelerators in a single hop for scale-up domains, while scale-out clusters of up to 128,000 XPUs can be built as a single leaf-spine pod, eliminating the super-spine layer that previous-generation fabrics require. A flatter fabric means fewer hops, fewer optics, and more predictable job completion times.

At general availability, OcNOS-DC is planned to support native 1.6 Tbps Ethernet interfaces, with up to 64 ports of 1.6TbE per switch depending on platform configuration, allowing operators to aggregate traffic onto fewer, faster links and prepare their networks for the next wave of 1.6T optics and AI accelerators.

In particular:

• Latency: Fewer hops at every scale. Scale Up: with twice the capacity in every switch, fewer devices sit in the traffic path. Scale Out: clusters of up to 128,000 XPUs fit in a two-layer leaf-spine fabric, eliminating the super-spine hop. Scale Across: a flatter fabric inside each data center leaves more of the latency budget for the links between sites, where OcNOS provides the interconnect.
• Performance: Lossless at full line rate. Scale Up: native 1.6TbE ports move AI traffic at full line rate without oversubscribing uplinks. Scale Out: OcNOS-DC support for Priority Flow Control (PFC) and Explicit Congestion Notification (ECN) keeps RoCE traffic flowing without drops as the fabric grows, protecting job completion times. Scale Across: OcNOS streaming telemetry runs from the leaf to the data center interconnect, so operators can find and fix bottlenecks across sites with one toolset.
• Capacity: Double the bandwidth, same footprint. Scale Up: Tomahawk 6 delivers 102.4 Tbps on a single 3nm chip, doubling Tomahawk 5 capacity without a proportional increase in power. Scale Out: up to 512 x 200G ports per switch let operators connect more XPUs per leaf and build larger clusters with fewer switches. Scale Across: 1.6T and 800G uplinks feed interconnect platforms running OcNOS, so capacity between data centers keeps pace with the growing fabric.

“The most powerful switch silicon deserves the most open NOS. OcNOS-DC on Tomahawk 6 gives operators lower latency, lower TCO, and total freedom of choice, so they can grow AI capacity on their own terms while keeping costs predictable,” said Miguel Alonso, Chief Product and Marketing Officer at IP Infusion. “Neoclouds and data center operators deploying a single networking operational model across Scale Up, Scale Out, and Scale Across are able to do this while delivering simplicity and operational efficiency."

IP Infusion will preview OcNOS on Tomahawk 6 at NANOG 98, Oct. 19-21 in Miami, and is previewing OcNOS on Broadcom Qumran3 carrier routing silicon at Network X 2026, Oct. 13-15, at VIECON in Vienna, Booth J72. To book a meeting with an IP Infusion engineer, visit IP Infusion at NANOG 98 or Network X 2026.

See OcNOS on the Broadcom Tomahawk 6

About IP Infusion

IP Infusion develops open network software for carriers, service providers, and data centers. Its flagship OcNOS® platform powers disaggregated networks worldwide, allowing operators to reduce TCO and avoid vendor lock-in. Headquartered in Santa Clara, Calif., IP Infusion is a wholly owned subsidiary of ACCESS CO., LTD. Visit www.IPInfusion.com.

Fonte: Business Wire