WhiteFiber has commercially launched Continuum, a cross-data-center architecture designed to make GPU infrastructure in two physically separate facilities operate as one logical AI supercluster.
The company says the design links two HITRUST-certified QTS facilities 83 kilometers apart over 12 Zayo dark-fiber strands, with a target aggregate bandwidth of 136 Tbps and 0.9 milliseconds of round-trip latency. Independent testing coverage notes that full-fiber-spectrum validation is still underway, an important qualification to the commercial specification.
The New Constraint Is the Size of One Campus
Large AI clusters are normally designed around a single campus because distributed training is extremely sensitive to latency, packet loss and bandwidth. That architecture becomes a problem when one site runs out of power, rack space or interconnection capacity before demand for GPUs stops growing.
Continuum attacks that physical ceiling rather than simply adding more servers. It combines optical transport, DriveNets AI Fabric and WEKA NeuralMesh storage infrastructure so compute at two sites can be presented as one logical resource pool.
136 Tbps Across 83 Kilometers
The commercial architecture uses 12 dark-fiber strands between the two facilities. WhiteFiber says additional wavelengths bring the design to 136 Tbps of aggregate bandwidth, while round-trip latency remains 0.9 ms across the 83 km route.
The distinction between tested and designed performance matters. Earlier R&D testing demonstrated 111.2 Tbps using only part of the available fiber spectrum. The 136 Tbps figure reflects the expanded commercial design, and WhiteFiber says full-spectrum testing is continuing.
Why Distributed AI Clusters Matter
The AI infrastructure market is already trying to solve similar bottlenecks at several layers. BitcoinVersus.tech recently covered 616 Blackwell GPUs activated across 77 nodes, showing how GPU capacity can be packaged as a cloud resource without customers owning the underlying facility.
Networking is evolving just as quickly. 1.6T optical links, 3.2T-class networking technology and massive silicon-photonics manufacturing expansion all point toward the same problem: GPU performance is rising faster than traditional infrastructure boundaries were designed to accommodate.
Storage Has to Cross the Boundary Too
A fast optical link alone does not create a unified cluster. Training and inference jobs need access to model checkpoints, datasets and intermediate results without turning the storage layer into the next bottleneck.
That is why WEKA’s NeuralMesh is part of the architecture. BitcoinVersus.tech has previously examined AI storage built around parallel data access and the role of RDMA in high-speed computing. Continuum applies the same basic principle at metropolitan scale: moving data efficiently is as important as locating the GPUs.
Power Becomes a Network Problem
The most interesting use case may be power rather than networking. If one data center cannot obtain additional megawatts, a second energized facility can provide capacity without waiting years for a new campus or utility interconnection.
That fits a broader shift already visible across AI infrastructure. Energized data-center capacity is becoming a strategic asset, while physical networking manufacturers are expanding production to support denser deployments.
What Continuum Has—and Has Not—Proved
WhiteFiber has moved the architecture from an R&D announcement to commercial availability. That is a meaningful milestone. It does not establish that every distributed AI workload will scale across 83 km with the same efficiency as a tightly coupled single-campus cluster.
Workload behavior, collective communication patterns, storage traffic and failure handling will determine real-world performance. The company also still has full-spectrum testing underway for the expanded 136 Tbps configuration.
The larger signal is architectural: as AI facilities hit local power and space ceilings, the network connecting data centers may increasingly become part of the compute system itself.
Live Project Updates
DriveNets directly highlighted the Continuum launch in this project-specific post, linking the network architecture with WhiteFiber and WEKA.
WhiteFiber’s correct X account also posted the Continuum announcement directly.
BitcoinVersus.Tech Editor’s Note
Support independent technology reporting: BTC donations may be sent to 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb.
Follow BitcoinVersus.tech on X/Twitter for AI infrastructure, Bitcoin mining, semiconductor and data-center reporting.
Disclaimer: BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.
Leave a comment