For years the 1-gigabit uplink was the default. Every server shipped with a 1GbE port, and it was enough. It is not enough anymore. With us, a 1G port has exactly one job left: out-of-band management, the IPMI or BMC port. Everything that carries actual data runs on at least 10GbE.
10G is the new 1G. And honestly, we are already well past it.
The Math That Settles It
A 1GbE link tops out at roughly 125 MB/s. That is not a guideline, it is the ceiling: 1000 megabit divided by 8. A single NVMe SSD delivers around 7 GB/s today, more than fifty times as much. The bottleneck stopped being the disk a long time ago. It is the network port in front of it.
Put differently: you buy fast NVMe, fast CPUs and plenty of RAM, and then all of it hangs off a link your storage saturates in seconds. The server is not slow. The network makes it slow.
Where 1G Fails in 2026
The workloads that matter today are exactly the ones that bring 1G to its knees:
- Backups: A nightly full backup crawling over 1G is not done by morning.
- Object storage: S3-compatible storage lives on throughput. At 125 MB/s per connection you feel every copy.
- VM migration: A live migration over 1G blocks your maintenance window instead of shortening it.
- Egress: When your users pull data, 1G is the wall your traffic runs into.
Our Data Path
With us the data plane starts at 10GbE. Most of the time it runs on dual 25GbE, sometimes on 100GbE, wired into our own 100G backbone, whose sites sit right next to DE-CIX Frankfurt and AMS-IX. All of it runs on our own network, not on rented third-party capacity. The 1G port stays where it belongs: on the management interface.
The short version
1GbE ≈ 125 MB/s. An NVMe ≈ 7 GB/s. The port is the bottleneck, not the disk. That is why 1G is the management port here, nothing more.
What It Means Per Product
Storage servers: Dual 10GbE (LACP) is standard, not a paid add-on. The NVMe or HDD array is not throttled by the uplink.
Managed Proxmox: The operations that hurt on a slow link are live migration and networked storage. A large VM migrated over 1G crawls, and disks served from a shared Ceph or storage backend over 1G are throttled on every I/O. On our 25G fabric, migrations finish quickly and the storage backend stops being the ceiling.
Managed Kubernetes: Image pulls, persistent-volume I/O and east-west pod traffic all share the node's uplink. On a 1G node they contend and stall the moment you scale out or move real data. On 10 and 25G nodes they have headroom.
The Point
Fast hardware has become the standard. The network behind it has not. Anyone still selling 1G as a data link in 2026 is selling you the bottleneck along with it. With us, the network carries the hardware, not the other way around.