DriveNets Expands AI Networking Portfolio with Broadcom Tomahawk 6 Systems

DriveNets introduced two new AI Ethernet fabric platforms based on Broadcom’s Tomahawk 6 switching silicon, targeting AI clusters that scale from hundreds to hundreds of thousands of XPUs. The new DriveNets 2600SL and DriveNets 2601S each deliver 102.4 Tbps of switching capacity across 64 ports operating at 1.6 Tbps. The platforms will begin shipping in Q3 2026 and are available in both air-cooled (2601S) and fully liquid-cooled (2600SL) versions. DriveNets says the liquid-cooled model features a fully liquid-cooled architecture rather than hybrid cooling, enabling consistent thermal performance across the entire chassis.

The new systems expand the company’s AI Fabric portfolio across scale-up, scale-out, scale-across, front-end, and AI storage networking. Each platform supports multiple port configurations, including 64×1.6 Tbps, 128×800 Gbps, 256×400 Gbps, and 512×200 Gbps, allowing operators to support current and future generations of AI accelerators using two-level and three-level fat-tree topologies. DriveNets said its software stack includes NIC driver, kernel, and Collective Communications Library (CCL) optimizations designed to improve cluster efficiency and token generation performance. The platforms also integrate with the company’s AI Cluster Orchestrator for provisioning, benchmarking, performance validation, and Day-2 operations.

Broadcom’s Tomahawk 6 Ethernet ASIC provides the switching foundation for the new platforms. DriveNets positions the solution as an open Ethernet alternative to vertically integrated AI networking architectures, enabling multi-vendor AI clusters while seeking to lower infrastructure costs and improve utilization. The company said its AI Fabric targets hyperscalers, foundation model developers, NeoCloud providers, and enterprise AI deployments supporting both training and inference workloads.

• Two new platforms: DriveNets 2600SL (fully liquid cooled) and 2601S (air cooled)
• Powered by Broadcom Tomahawk 6 Ethernet switching ASIC
• 102.4 Tbps switching capacity
• 64 × 1.6 Tbps ports, with flexible configurations down to 200 Gbps
• Supports clusters ranging from hundreds to hundreds of thousands of XPUs
• Supports 2-level fat tree, 3-level fat tree, and multi-plane AI network architectures
• Available in Q3 2026
• Includes AI Cluster Orchestrator for provisioning, benchmarking, lifecycle management, and performance tuning
• Software optimizations extend from NIC drivers through kernel and Collective Communications Library (CCL)

“The most expensive idle asset in the world today is an XPU waiting on the network,” said Ido Susan, co-founder and CEO of DriveNets. “The new additions to our AI portfolio deliver industry-leading performance at massive cluster scale and let our customers maximize infrastructure utilization and power efficiency on any AI accelerator.”

🌐 Analysis

The announcement reflects how rapidly Ethernet switching has become the preferred interconnect for large AI clusters. Tomahawk 6 represents the latest generation of merchant Ethernet silicon supporting native 1.6 Tbps interfaces, allowing vendors such as DriveNets, Arista, Cisco, HPE, Nokia, and others to build increasingly dense AI fabrics as accelerator bandwidth continues to double. The industry is simultaneously transitioning toward rack-scale architectures, liquid cooling, and higher radix switching to reduce network diameter and improve cluster efficiency.

For DriveNets, the launch extends a strategy that further extends beyond carrier networking into AI infrastructure. The company already offers software spanning scale-up, scale-out, scale-across, and storage networking, positioning itself as an independent software and systems supplier for heterogeneous AI clusters rather than vertically integrated accelerator platforms. The emphasis on open Ethernet also aligns with broader industry initiatives including Ultra Ethernet Consortium technologies and growing adoption of multi-vendor AI infrastructure.

🌐 We’re tracking the latest developments in AI infrastructure. Follow our ongoing coverage at: https://convergedigest.com/category/ai-infrastructure/