Amsterdam and Frankfurt are under 4 ms apart, so for most of Europe the choice is a wash. At the edges it is not: the UK, Ireland, the Benelux and the Nordics lean Amsterdam. The latency table, the VPN and delivery workloads that suit it, and a sizing rule for each.
An Amsterdam dedicated server puts your hardware in the Netherlands, at the other end of Europe's busiest internet corridor from Frankfurt, and for a large class of workloads it is the better of the two. Amsterdam is closer to the UK, Ireland and the Nordics, it has two large internet exchanges in the same metro, and it has long been a favourite for delivery, VPN and streaming infrastructure. This article covers who an Amsterdam server reaches best, how it compares with Frankfurt, which workloads it suits and how to size one.
The two cities are 364 km apart — under 4 ms of round-trip floor — so for most of continental Europe the choice between them is a wash. The differences are at the edges of the map:
| Audience | Better from | Why |
|---|---|---|
| UK and Ireland | Amsterdam | 3–4 ms shorter round trip to London and Dublin |
| Nordics | Amsterdam | Shorter paths north via Hamburg and Copenhagen |
| Benelux | Amsterdam | Users are a few milliseconds away |
| Germany, Austria, Switzerland | Frankfurt | Users are a few milliseconds away |
| Italy, central and eastern Europe | Frankfurt | 2–4 ms shorter to the south and east |
| Even spread across Europe | Either | Differences under 5 ms either way |
If your analytics show a westward or northward lean, take Amsterdam; if they lean south or east, take Frankfurt. For a primary and a standby, the two cities make an excellent pair: close enough that replication is cheap, far enough apart to be separate failure domains. The Frankfurt guide covers the other half of that pair.
Same method as everywhere in this series: the straight-line distance divided by 100 gives the theoretical round-trip floor in milliseconds, and real paths land 30 to 50% above it plus a millisecond or two for routers.
| From Amsterdam to | Great-circle | Theoretical floor | Working estimate |
|---|---|---|---|
| Brussels | 173 km | 1.7 ms | 3–5 ms |
| London | 358 km | 3.6 ms | 6–7 ms |
| Frankfurt | 364 km | 3.6 ms | 6–7 ms |
| Hamburg | 365 km | 3.7 ms | 6–7 ms |
| Paris | 430 km | 4.3 ms | 7–8 ms |
| Copenhagen | 621 km | 6.2 ms | 9–11 ms |
| Dublin | 756 km | 7.6 ms | 11–13 ms |
| Oslo | 913 km | 9.1 ms | 13–16 ms |
| Stockholm | 1,125 km | 11.3 ms | 16–19 ms |
| New York | 5,863 km | 58.6 ms | 77–90 ms |
London at 6–7 ms is the number that matters most for many buyers. The UK is one of Europe's largest online markets, and an Amsterdam server reaches it nearly as quickly as a server in London itself, with EU jurisdiction kept intact.
Amsterdam is home to AMS-IX, one of the world's largest internet exchanges, and NL-ix, which interconnects data centres across the Netherlands. Large content networks, access providers and carriers maintain a presence in the city, which is why traffic from an Amsterdam server so often reaches its destination network without a long detour through transit.
The question to put to any host in Amsterdam — or anywhere — is not "are you in Amsterdam" but "which networks do you hand my traffic to, and where". Peering vs transit explains what the different answers mean for your users, and our network page lists the carriers and content networks we connect with.
VPN exit nodes. The Netherlands has long been one of the most requested exit countries for VPN services, and Amsterdam's connectivity to the UK, the Nordics and western Europe makes it a natural first or second location after Frankfurt. VPN traffic is symmetric and heavy, so a guaranteed rate matters more than CPU; our servers for VPN providers are configured around that.
Delivery to the UK and Ireland. Download mirrors, game patches, video catalogues and software updates for British and Irish users are well served from Amsterdam, with a round trip close to that of a London server.
Streaming and media. Amsterdam has a long history as a media-hosting centre, and an origin there reaches the Benelux, the UK and the Nordics in a few milliseconds.
A second European location. If Frankfurt is your primary, Amsterdam is the shortest-distance standby that is still a separate metro, separate facilities and separate power grids.
VPN and delivery workloads are bandwidth-bound long before they are CPU-bound. A rough sizing rule for each:
VPN exit: peak concurrent users × average active rate per user
Delivery: peak concurrent downloads × average download rate
A VPN pool with 1,500 concurrent users averaging 3 Mbps each needs 4.5 Gbps at peak — the case a 5 Gbps dedicated server is built for. A software mirror pushing 200 parallel downloads at 20 Mbps needs 4 Gbps. Neither needs much more than a dual Xeon; both need the guaranteed rate at 21:00 local time, when everyone is online at once.
For VPN specifically, watch the CPU cost of encryption. WireGuard and AES-GCM are fast on any CPU from the last ten years, but at multi-gigabit rates the work spreads across many cores, so check that your VPN software actually uses them.
The location is a free choice at checkout, so Amsterdam is priced exactly like our other 12 cities. The tiers that suit VPN and delivery work start here:
| Guaranteed bandwidth | Entry configuration | Per month |
|---|---|---|
| 1 Gbps unmetered | Dual Xeon E5-2630 v4, 20 cores, 128 GB | €209 |
| 5 Gbps unmetered | Dual Xeon E5-2630 v4, 20 cores, 128 GB | €519 |
| 10 Gbps unmetered | Dual Xeon E5-2630 v4, 20 cores, 128 GB | €659 |
Larger builds, up to dual EPYC, are available on every tier. Pick Amsterdam in the location selector when you configure the server; most in-stock configurations are online within an hour. For the full ladder from 1 to 200 Gbps, the unmetered dedicated server buyer's guide works out the cost per terabyte of each tier.
Users in southern or eastern Europe. Madrid, Rome, Bucharest and Athens are all further from Amsterdam than from Frankfurt, and further again from both than from a server in their own region.
Audiences outside Europe. North American users are better served from the US east coast, and traffic from Africa, the Middle East and Asia enters Europe mostly through Marseille.
Workloads where the metro does not matter. A backup target or a batch-processing box gains nothing from Amsterdam's interconnection. Put it wherever stock and price are best, in a different city from the data it protects.
For an audience spread across continental Europe the difference is a few milliseconds either way. Choose Amsterdam if your users lean towards the UK, Ireland, the Benelux or the Nordics, and Frankfurt if they lean towards Germany, Italy or eastern Europe.
The straight-line floor is about 3.6 ms round trip, and real paths typically land between 6 and 7 ms. That is close to a server in London itself, while the data stays inside the EU.
Yes. It reaches the UK, the Nordics and western Europe on short paths, and it is one of the most requested exit locations. Size the port for your peak concurrent users and check that the VPN software uses all available CPU cores.
No. Amsterdam is priced like every other city we sell: from €209 a month for 1 Gbps guaranteed, or €519 for the 5 Gbps tier that suits most VPN pools. Stock differs between cities, which can affect how soon a particular build is ready.
Frankfurt is the default European location, and usually the right one: one of the world's largest internet exchanges, and no major capital in the west or centre of the continent more than about 25 ms away. The latency table, the workloads that fit, and the cases where another city wins.
Unmetered says the bytes are not counted; it does not say the port is yours at the busy hour. The five contract clauses to check, how to size a tier from your peak, what each tier costs per delivered terabyte, and the 13 cities where the server can live at the same price.
Generating a token means reading the model's active weights once, so a CPU server's speed ceiling is its memory bandwidth divided by the model's size. The arithmetic, the catch in prompt processing, why two sockets are not twice as fast, and the workloads a big-memory EPYC box genuinely suits.