en.Wedoany.com Reported - DriveNets recently announced the industry's first commercial deployment of a long-distance cross-domain AI network. As part of WhiteFiber's Redwood project, the DriveNets AI Fabric connects two WhiteFiber H200 GPU clusters located 52 miles apart into a single logical GPU supercluster, with a verified bandwidth of 111.2 Tbps and a guaranteed latency of 0.9ms. Although cross-domain architectures have been widely discussed in the industry, DriveNets is the first to advance the concept from theory to actual commercial deployment, validated in a production-scale environment rather than a laboratory setting.
Increasingly, the constraints on AI infrastructure construction stem from power and space limitations at a single location, rather than computing power. Cross-domain architecture removes this limitation; AI builders are no longer restricted by the power capacity of one facility but can extend clusters to remote sites and operate distributed GPUs as a unified system. This enables larger, more resilient clusters that can be built wherever power is available, without compromising performance.
Extending clusters over distance presents greater networking challenges than simply laying cables between two sites. Links connecting remote sites typically carry lower bandwidth than the switching fabric within a facility, leaving little margin to absorb burst traffic without causing congestion. AI training further exacerbates the challenge: it generates a small number of extremely large traffic flows that arrive in synchronized bursts, which traditional data center load balancing and buffering methods cannot handle. Without a switching fabric specifically designed to absorb these bursts and manage congestion in real time, latency spikes and packet loss occur, leaving GPUs on both sides of the cluster idle. Solving this problem over long distances without sacrificing performance is why cross-domain architecture—and its switching, buffering, and congestion management technologies—is critical to the next phase of AI infrastructure development.
WhiteFiber's Redwood project connects two geographically separated GPU clusters into a single logical GPU supercluster, with the DriveNets AI Fabric solution providing the high-performance network linking the two sites. Ido Susan, co-founder and CEO of DriveNets, stated that power availability may be a major constraint on AI infrastructure development, but through this validated deployment, it is no longer an obstacle. Sam Tabar, CEO of WhiteFiber, noted that DriveNets' AI Fabric was crucial in proving that the Redwood project could deliver the performance and reliability of a single-site cluster across two locations. The validation process compared performance between GPU racks within a single site and between GPU racks across the two sites, with one GPU rack located at the primary site and the other at the remote site. Further details on the validation methodology and results are available in the DriveNets white paper.
Traditional data center interconnect links are not designed for AI workloads, which generate burst traffic that cannot tolerate jitter or packet loss. DriveNets' 9300F, 5300R, and 5301R switches, powered by its Fabric Scheduled Ethernet (FSE) technology, extend the AI switching fabric beyond a single data center through cell-based load balancing, end-to-end virtual output queuing (VOQ), and deep buffer interconnects, absorbing AI traffic bursts before they cause congestion. The result is predictable, lossless connectivity between sites, keeping GPU utilization high.
