VPS Network Speed (2026): The Port Speeds and Transfer Allowances Providers Actually Publish

1. Four different numbers, often confused

"1 Gbps VPS" can mean four things, and providers use the phrase for at least three of them.

• The port speed is the speed of the network interface the provider gives the plan. Contabo's "200 Mbit/s Port" is this, and so is OVHcloud's bandwidth level.
• The host's uplink is the speed of the physical machine your VM shares with other customers. Hetzner's 10 Gbit and InterServer's 10 Gbps are this, and both say so.
• The network cap is a limit the provider's own rate limiter enforces on your instance, which can be lower than the port. Linode publishes one per plan; Azure publishes one per VM size.
• The transfer allowance is a monthly quantity of data, unrelated to speed: 1 TB, 2 TB, unlimited.

None of the four is the throughput you will measure. That depends on the route between your server and whoever it is talking to, on the other tenants, and on the time of day. A network comparison is therefore the table below plus your own measurement, and the fifth number, "typical sustained throughput", does not exist in public form for any provider here except Hetzner.

2. What each provider publishes

Read from each provider's own page or public API on 2026-09-22, except the Vultr, DigitalOcean, Hetzner, InterServer, BuyVM and Azure rows, which are from this site's verified pages for those providers (dates in each). A blank in the port column means the provider states nothing there, not that the link is slow.

3. OVHcloud: a port speed per model

OVHcloud's public order catalog carries a technical blob per product with a bandwidth level in Mbit/s and an unlimited flag. For the current US VPS line, labelled 2027 in the catalog, in September 2026:

This is the most specific network commitment at these prices among the providers on this page: a figure attached to the plan rather than to the host, in a machine-readable catalog anyone can query. Two generations sit in the catalog at once and the names invite a mistake. The current line is labelled "VPS-1 2027" through "VPS-4 2027" and is the one priced above; the older line is labelled "VPS-1 2026" through "VPS-6 2026" and costs more for less network (its VPS-1 is 4 vCPU and 8 GB at $7.60 with a 400 Mbit/s port). The higher year is the newer and cheaper generation, so read the year on the order page rather than assuming the larger number is the older product.

4. Contabo: a port speed per plan, unlimited transfer

Contabo's Cloud VPS page lists a port speed on every plan card and unlimited traffic on all of them: "200 Mbit/s Port" on Cloud VPS 4, then 300, 600 and 800 Mbit/s, and "1 Gbit/s Port" on Cloud VPS 16 and 18. Its summary sentence: "Every Cloud VPS plan comes with unlimited inbound and outbound data transfer. Port speeds run from 200 Mbit/s up to 1 Gbit/s. Fair usage policies apply."

The trade is visible in the numbers: Contabo sells the most memory per dollar on this site (4 vCPU and 8 GB for $6.60 on its introductory 24-month price) and the slowest entry port. For a web application serving pages, 200 Mbit/s is more than enough; for moving large files or serving video, it is the constraint, and the plan ladder is also a port ladder.

5. Linode: a network cap per plan

Linode's types API returns a network_out value in Mbps and a transfer value in GB for every plan. In September 2026: Nanode 1 GB, 1,000 Mbps and 1 TB, $5; Linode 2 GB, 2,000 Mbps and 2 TB, $12; 4 GB, 4,000 Mbps and 4 TB, $24; 8 GB, 5,000 Mbps and 5 TB, $48. The figure is an outbound cap the provider enforces, which is a different promise from a port speed and a much more specific one than most of this table. Inbound is not capped in the same field.

6. Hetzner and InterServer: a shared host port

Both publish the host's uplink rather than a per-plan figure, and both say it is shared. Hetzner's plan FAQ, quoted on this site's Hetzner specifications page, is unusually direct: the host systems have a redundant 10 Gbit connection shared by all servers on the host, there is no bandwidth guarantee, and "you can expect about 300-500 Mbits" per server. That last clause is the only published expectation on this page, and it is worth more than any advertised ceiling.

InterServer states a 10 Gbps shared port on every slice and 2 TB of transfer per slice, without an expectation figure. The InterServer guide covers what that means in practice: a ceiling for the host, not a guarantee for your server.

7. Transfer allowances and what happens when you exceed them

The allowance is the part of a network comparison that actually appears on bills. Two shapes exist: a monthly quantity (Vultr, Linode, DigitalOcean, InterServer, Hetzner, AWS Lightsail) or unlimited traffic with a port limit (OVHcloud, Contabo) or unmetered (BuyVM).

What happens after you exceed a quantity differs by provider, and is the detail to check on the provider's own billing page before you rely on a number here: Hetzner publishes a per-TB price for additional traffic in its documentation ($1.20 per TB in Europe and the US at the last reading on this site's Hetzner page). The others describe overage on their billing or pricing pages rather than in the plan API, so this page does not quote figures it has not read this month. If a large transfer bill is a real risk for your workload, the safe choices are the unlimited-traffic providers or a provider whose overage rate you have checked in writing.

A rough sense of scale: 1 TB a month is about 33 GB a day, or roughly 165,000 page views of a 6 MB page. Sites that serve video or large downloads pass it; most others never come close.

8. Measuring the server you rent

Published numbers tell you the ceiling. The throughput between your server and your users is a property of the route, and measuring takes an hour on an hourly-billed plan.

1. Install iperf3 (apt install iperf3 or dnf install iperf3).
2. Run it against a machine that matters: iperf3 -c HOST -t 30 -P 4 for upload from the server, -R for the reverse direction. A second server of your own in your users' city gives a truer answer than a public iperf3 server whose own load is part of the result.
3. Use the providers' test endpoints for a first pass. Linode runs per-region speedtest hosts, InterServer publishes a speed test page with download files from Secaucus, Los Angeles and Dallas, and Kamatera's and Hostwinds's datacenter pages have test buttons per city. The low-latency page lists them.
4. Repeat at different hours and keep the median and the spread. On a shared host the spread is the interesting number: it is the other tenants.
5. Measure latency separately with ping or mtr. Throughput and latency are different problems: a page load is usually latency-bound, a backup transfer is throughput-bound.

9. When network speed is not your bottleneck

For most sites it is not. A page that takes 2 MB to render moves that 2 MB in well under a second even on a 200 Mbit/s port, and the time the visitor experiences is mostly round trips: DNS, TLS, the first byte from the application, then the sequential requests the page makes. That is why the datacenter you choose usually matters more than the port speed on the plan, and why a cache in front of the server beats a faster port for a content site.

Network speed does decide the outcome for: large file distribution, video, backup and restore windows, database replication between regions, and anything where the transfer allowance rather than the rate is the limit. If your workload is one of those, the OVHcloud and Contabo port figures and the BuyVM unmetered plans are where this page's data is most useful; for everything else, pick on price, memory and location and measure once you are there.