25 September 2026 — WhiteFiber announced on 23 September that Continuum, an architecture connecting two data centres into what it calls a single logical GPU supercluster, was commercially available to enterprise customers. The company says the sites are 83 kilometres apart and can pool computing capacity across that distance.
Key points
- WhiteFiber says Continuum connects two HITRUST-certified QTS facilities over 12 Zayo dark fibre strands.
- The company claims 136 Tbps of bandwidth and guaranteed round-trip latency of 0.9 milliseconds.
- DriveNets supplies the network fabric; WEKA supplies storage and memory infrastructure.
- WhiteFiber says the design is intended to support pooled computing and failover while keeping sensitive workloads in their originating jurisdiction.
12 Zayo strands join the QTS sites
WhiteFiber says Continuum uses two HITRUST-certified QTS facilities connected by 12 Zayo dark fibre strands. It describes them as one logical system: computing equipment remains in separate buildings, while the connections and software are intended to let the sites contribute to the same pool of work. For a company building on Continuum, the distinction is whether capacity at the second site can be used as part of a shared cluster rather than managed as a separate installation.
The company puts the connection’s bandwidth at 136 Tbps and calls its 0.9 milliseconds of round-trip latency guaranteed. Bandwidth describes how much data a connection can carry in a given time; round-trip latency describes how long a message takes to go to the other site and return. One is the width of the road, the other the time taken to make the journey. Both matter when work and data must move between distant computers.
WhiteFiber presents that combination as a way to use computing capacity beyond the power and space limits of a single campus. Its proposed uses include adding capacity for AI training, in which a system learns from data, and inference, in which a trained system produces an answer. Those uses depend on more than a fast link: the computers at each location also need access to the data their work requires.
The company also says Continuum could bring telecom and metropolitan facilities with available power and fibre into a shared GPU cluster. In WhiteFiber’s account, those locations could contribute to AI workloads instead of operating as isolated capacity, including work delivered closer to where a service is used. That is a proposed application of the architecture, separate from the company’s stated connection figures for the two QTS sites.
DriveNets carries traffic and WEKA supplies data
WhiteFiber built Continuum with DriveNets and WEKA. According to the company, DriveNets AI Fabric provides the network connecting the facilities, while WEKA NeuralMesh supplies storage and memory infrastructure across the cluster. The arrangement gives the two partners different jobs: moving information between sites and making the information needed by the GPUs available across them.
WEKA chief executive Liran Zvibel says NeuralMesh is intended to make data accessible to GPUs separated by many kilometres as though it were local. DriveNets executive Yossi Kikozashvili says its Ethernet-based fabric is intended to move data efficiently between parts of a cluster. Their accounts describe the contribution each company expects its product to make to Continuum, rather than a measurement of a customer’s workload.
WhiteFiber says the design allows sensitive workloads to remain within their originating jurisdiction even when computing resources are pooled or work fails over between locations. It also says a training run is designed to continue if one site goes down. Those are consequential requirements for a company deciding where to place work: sharing computers across sites and deciding where sensitive work may run have to be handled together.
Using a cloud service during a site outage could be less disruptive if work could carry on at a connected location, as WhiteFiber says Continuum is designed to allow. Under the company’s proposed arrangement, sensitive work would still have to remain within its originating jurisdiction while computing capacity is shared across the sites.
Sam Tabar, WhiteFiber’s chief executive, describes enterprise use at scale, compliance requirements and continuity through a site problem as aims for the offering. For a customer considering the system, each involves a different part of the design: the link between buildings, the placement of sensitive work and the ability of one location to take on work when the other cannot.
WhiteFiber puts deployment at scale among the risks
In its forward-looking statements, WhiteFiber includes anticipated performance following full-fibre spectrum testing among the matters that may differ from its expectations. The company also identifies technical or engineering challenges affecting deployment at scale and its reliance on DriveNets, WEKA, Zayo and QTS for the integrated offering. Those risks concern the work of delivering the architecture beyond its announced configuration.
WhiteFiber says Continuum has been commercially available since 23 September. Its stated uses extend from pooled capacity for training and inference to failover and workloads that must observe jurisdictional requirements. The company lists expected applications in telecommunications, edge computing and regulated enterprise settings among its forward-looking statements, distinguishing those expectations from the availability it announced.
WhiteFiber says it has submitted patent applications covering the underlying implementation. Its forward-looking statements include the pending applications and any protection they may afford.
