Kubernetes already abstracts the hardware, so a virtualization layer underneath only does it
twice. Bare-metal nodes give your clusters every core, all the memory bandwidth and NVMe at
native speed. You run the cluster, we run the servers and the network under it.
No hypervisor overhead and no slice of a shared host. Every core and all the memory on the server go to your pods.
Traffic between nodes, priced as capacity
Microservices and replicated storage move a lot of data between nodes. You pay a flat monthly rate for an unshared uplink at the speed you choose, with unmetered traffic, so none of it shows up as a per-GB line on the invoice.
Servers on demand, via API
Order servers through our GraphQL API or the client panel and join them to your cluster the way you already do. Most in-stock servers deploy the same day.
Recommended setup
What we’d put under your Kubernetes.
The starting points we recommend most often. Your account manager sizes the exact config
with you.
Multi-region usually means a cluster per region, since etcd does not tolerate
intercontinental latency well. GPU servers can join the same cluster for inference and
training.
Recommended configuration
A typical starting cluster
Control plane
3 or 5 × AMD EPYC 4585PX, 16 Cores64 GB RAM each, spread across locations. Small VMs work too.
330+ Tbps of capacity, a 50 Tbps private
backbone connecting 50 of our locations, and
300+ private interconnects with other networks. Check the routes yourself on
the looking glass.
