WhiteFiber Launches Continuum, Linking Two Data Centers 83 km Apart Into One GPU Supercluster
WhiteFiber has launched Continuum, a networking architecture that makes two separate data centers work as a single GPU supercluster. The first deployment connects two QTS facilities 83 kilometers apart. It’s commercially available now.
How it’s built
The two sites are linked by 12 strands of Zayo dark fiber carrying 136 Tbps, with a guaranteed round-trip latency of 0.9 milliseconds. DriveNets supplies the AI networking fabric across the link. WEKA’s NeuralMesh provides a shared storage and memory layer so both sites see the same data.
“It turns two data centers into one engine,” says WEKA CEO Liran Zvibel.
Why spread a cluster across sites
The limit on AI clusters today is often power. Getting enough of it at one site can take years. If a cluster can span two facilities, operators can add GPUs wherever power exists, without waiting on a new greenfield campus.
There are other benefits too. Training can keep running if one site goes offline. Customers with data sovereignty rules can keep sensitive workloads in their home jurisdiction while still using the combined capacity. And overflow demand can spill from one site to the other.
“Today we are making it commercially available to enterprises that need AI compute to perform at scale” and “hold up under compliance requirements,” says CEO Sam Tabar. Both QTS sites are HITRUST-certified, which matters for healthcare and other regulated customers.
The catch
Spreading one training job over 83 km only works if the network is fast enough that GPUs aren’t left waiting. That’s why the latency guarantee is the headline number. Sub-millisecond round trips are what make it believable. The idea of distributed training has been around a while. WhiteFiber is now selling it as a product.