Built for the traffic
nobody else wants.
There is no clever filter behind this list — it is simply what an uncounted port and a whole machine are good for. If your workload is lawful and it moves bytes, it fits.
CDN and cache origins
The box every edge node pulls from. Origin traffic is bursty and unforgiving, and it is exactly the traffic metered hosting punishes hardest.
unmetered egress · 10–100 Gbps portstreamingVideo and live streaming
Transcoding on real cores, delivery on a port that does not flinch when a stream goes wide. No per-gigabyte line on the invoice at the end of the month.
sustained egress · many coresgameGame servers
Dedicated silicon, no hypervisor stealing cycles mid-match, and root to tune the kernel the way your netcode wants it.
single tenant · low jitterstorageStorage and backup targets
Large disk arrays fed over a port with no counter — the pattern that makes metered clouds expensive the moment you need to restore.
unmetered ingress and egressMail infrastructure
Your own MTA, your own IP reputation, your own rules — legitimate mail only, and the acceptable use section means it.
clean addressing · rootDatabases and heavy compute
Workloads that hate noisy neighbours: analytics, builds, batch jobs, anything where a p99 that doubles at random is not acceptable.
predictable cores · ecc memoryAll six break the same way on metered hosting.
Every one of them has a month where the traffic is the point — a launch, a match, a restore, a campaign. That is precisely the month a transfer allowance, a 95th-percentile line or a per-gigabyte tier turns a good result into a bill. Ports of 1, 10, 100 Gbps, up to 100 Gbps on a custom build, and no counter on any of them.