en.Wedoany.com Reported - DriveNets has launched a new AI network platform based on the Broadcom Tomahawk 6 (TH6) ASIC, with a total capacity of 102.4Tbps and 64 1.6Tbps ports. Industry analysts expect these platforms to account for the majority of newly deployed AI ports by 2027. The platform series is scheduled to ship in the third quarter of 2026, offering air-cooled (DriveNets 2601S) and liquid-cooled (DriveNets 2600SL) configurations to suit different data center environments.
Ido Susan, co-founder and CEO of DriveNets, stated that the new platform delivers industry-leading performance on large-scale clusters, helping customers maximize infrastructure utilization and energy efficiency. He noted that these capabilities will become even more critical as the industry moves toward heterogeneous AI architectures. Unlike other vendor solutions, the DriveNets 2600SL features a 100% liquid-cooled design, ensuring consistent system thermal efficiency and thereby maximizing energy efficiency.
The new platform expands DriveNets' AI Fabric portfolio, supporting vertical scaling, horizontal scaling, cross-dimensional scaling, and integrated front-end and storage networks. These platforms offer rack-level performance and high reliability, suitable for large-scale clusters. The platform supports flexible port cardinality configurations, including 64x1600Gbps, 128x800Gbps, 256x400Gbps, or 512x200Gbps, with low-latency characteristics.
The platform can support current and next-generation XPU data rates and clusters, utilizing 1.6Tbps or 2x800Gbps network interface card (NIC) connections, employing various network architectures (including 2-level fat tree, 3-level fat tree, and 2-level fat tree multi-plane), with cluster deployments supporting from hundreds to hundreds of thousands of XPUs. DriveNets' AI Fabric portfolio offers a high-performance alternative comparable to single-vendor solutions for open multi-vendor AI clusters, while reducing infrastructure costs.
DriveNets' solution improves token economics through full-stack performance optimization, covering NIC drivers, kernels, collective communication libraries (CCL), and hardware efficiency. The platform solution covers a wide range of AI use cases, including training and inference, with customers including hyperscale cloud providers, foundation model developers, NeoCloud providers, and enterprises.
Charlie Kawwas, president of Broadcom's Semiconductor Solutions Group, stated that as AI systems scale to hundreds of thousands of XPUs, the network fabric has become a key factor in AI performance, infrastructure utilization, and overall token economics. He believes that the combination of Broadcom's Tomahawk 6 chip with DriveNets' high-performance AI Fabric delivers unprecedented performance and reliability on large-scale clusters.
The software running on the new platform benefits from end-to-end fine-tuned performance optimization, covering NICs, drivers, kernels, and CCL. This software stack is complemented by the DriveNets AI Cluster Orchestrator, which manages the entire cluster lifecycle, including configuration, benchmarking, and large-scale operations. The platform enables rapid deployment through built-in performance validation and tuning, while DriveNets Infrastructure Services provides expert professional services to ensure reliable deployment achieving optimal performance.
Alan Weckel, founder and technology analyst at 650 Group, believes that Ethernet has become the dominant structure for AI infrastructure. He noted that the rack itself is rapidly changing, with liquid cooling, massive AI bandwidth, and new high-performance systems reshaping the market landscape. This market did not exist two years ago, but now the total addressable market (TAM) has exceeded $100 billion.
