How to choose Minecraft server hosting: what matters

Choose hosting by three things: how fast a single CPU core is, how close the machine sits to your players, and whether the host lets you run what you need (your own server jar, extra ports, full file access). RAM matters only until your world, plugins and players fit, which for most plugin servers is somewhere between 4 and 10 GB, not the 32 GB that big plans like to advertise.

We sell no hosting and this page has no affiliate links. We do run a pinger that checks every listed server every minute, so parts of this guide come from watching how servers on different kinds of hosting behave.

What actually limits a Minecraft server

A Minecraft server runs the world in ticks, 20 per second, which gives each tick a budget of 50 milliseconds. In that time the server moves every loaded mob, runs redstone and hoppers, grows crops, handles everything players did, and runs plugin code. If a tick takes longer than 50 ms, the next one starts late, and everyone on the server feels it.

The catch is that most of this work happens in order, on one main thread. Vanilla and Paper already spread chunk generation, lighting and networking across other threads, but entities, block updates, redstone and most plugin logic run on one core per server. For a single Minecraft server, a CPU with 8 fast cores beats one with 32 slow ones. Folia is the exception, because it splits the world into regions that tick in parallel, and it comes with big trade-offs that our guide to server software covers.

Bandwidth is rarely the limit for a small server, latency and packet loss are. And when a plan says "4 vCPU", ask what that means: on many virtual machines a vCPU is a single hardware thread, often on a core that other customers use too.

How much RAM you need

There is no official number, and guides that disagree by a factor of three are not all wrong, they just picture different servers. Memory fills up with the loaded world (view distance multiplied by how spread out your players are), entities, plugin caches (web map renderers and block-logging plugins are hungry), and above all mods.

These are starting points we would give a friend, not measurements:

Three points from the PaperMC docs are worth knowing before you pick a plan:

• Set -Xms and -Xmx to the same value. Memory between the two is memory you pay for and do not use.
• Leave about 1,000 to 1,500 MB on top of -Xmx for the Java process and the operating system, and in the docs' words, "do not allocate all of your available memory on a shared host".
• More is not better forever. The same page says buying 32 GB for a server "will only waste your money with minimal returns", and it recommends "at least 6-10GB, no matter how few players", because its garbage collection settings work better with room to spare. A friends server runs on less. For a public server, aim there.

Do not judge memory by the RAM bar in your host's panel. With -Xms equal to -Xmx, Java claims the whole amount at startup, so the bar looks full on day one. Run /spark health and /spark gc instead (spark ships with Paper). What matters is how much of the heap stays in use right after a collection, and how long the pauses last.

Free hosts: what our pinger sees

Free hosts keep their costs down by putting servers to sleep when nobody is online. Someone has to wake the server again, from a web panel, a lobby, or by joining and waiting. For friends that is a fair deal. For a public server it means that most of the time, anyone who checks your address finds a placeholder instead of your world.

We can measure this because our pinger records what answers, not just whether something does. In the 30 days to 16 October 2026, five listings on our list used the address of a free host that serves a sleep page (subdomains of Aternos and Falix). Our pinger got some answer on 66.5% of its attempts, but a joinable server sat behind that answer on 5.6% at most. Four of the five were joinable for four hours or less in the whole month. In the same period, the other listings that were playable at least once had a median playable uptime of 98.9%. Live, for the last 30 days: 5 listings on such addresses, answering 64.5% of pings.

6.2%Playable uptime of listings on free sleep hosts, last 30 days

99.0%Median 30-day playable uptime of all other public listings that were playable at least once in the period, shared lobbies and impostors excluded

Measured 19 Oct 2026, 22:55 UTCHow we measure

The answers that make up the gap are the hosts' own placeholders, with first lines such as "This server is offline.", "Server is OFFLINE. Join to start it automatically!" or "Server not found." A list that only checks whether something answered would call those servers online. We count them as offline, which is also how our uptime report treats them.

Free sleep hosts5 listings

Everything else207 listings

• Never answered
• Placeholder only
• Under 50%
• 50 to 90%
• 90 to 99%
• 99 to 99.9%
• 99.9% or more

Free hosts that answer with a sleep page count as answered but not playable, which is why they sit at the bottom.

Measured 19 Oct 2026, 22:55 UTCHow we measure

Minehut works differently. When a Minehut server is asleep, its address leads into Minehut's own lobby, so a ping can come back with thousands of players. On 18 October 2026 one Minehut address on our list answered with 4,759 players and the line "Server is offline. Start via the lobby with /join". The players were real, they were just in the lobby. That is why shared-lobby addresses are left out of our network totals, as the methodology explains.

On serverlist.net the default order puts servers that were playable in the last 24 hours ahead of the rest, and placeholder answers do not count as playable. How to get your first 100 players covers what new players see in their first five minutes.

Free hosting is still the right call for testing plugins, learning to run a server, or a small group that starts the server together. If your group wants no admin work at all, Realms, Mojang's own subscription service for private worlds, is the other option, without plugins and without a public listing.

Shared game hosting, VPS or dedicated

Paid hosting comes in four shapes. The right one depends on how much of the machine you need and how much of it you want to run yourself.

A common path is shared game hosting for the first months, then a dedicated machine once a spark profile shows the CPU is the limit (our lag guide shows how to read one).

Location and ping

Light in fiber covers about 200,000 km per second, so every 1,000 km between a player and your server adds roughly 10 ms to the round trip, before real cable routes add their detours.

Our pinger sits in western Europe and records each server's latency floor, the fastest round trip in a period. For playable servers that declare a single country, the medians between 21:00 and midnight UTC on 17 October 2026 were:

Live values: Europe 10 ms, North America 78 ms, Asia 147 ms.

Put the machine where most of your players are during your busiest hours. Your listing's analytics show which countries your visitors come from, and when Minecraft servers are busiest shows how peak hours differ by region. PvP players notice differences that builders on an SMP never will, so the tighter your combat, the closer the server should be. If you court both Europe and North America, the US East Coast is the usual middle ground. Nearly half of the North American servers in that sample (22 of 46) answered our pinger in 70 to 90 ms, roughly what one trip across the Atlantic costs. Before you pay, ping the host's test address from where your players live, or ask three of them to do it.

One trap: 12 of those 46 North American servers answered our European pinger in under 15 ms. They sit behind protection networks that answer at the edge nearest to whoever is asking, so a ping test measures the edge, not your machine. The trip from that edge to your server still happens on every packet.

DDoS protection and proxies

Minecraft servers attract attacks out of all proportion to their size. A banned player with a grudge is enough. Protection comes in three layers, and you want to know which ones you have before the first attack.

1. Your host's network filtering. Ask whether it is included and always on, whether it understands game traffic or only stops large floods, and what happens during a big attack: is it filtered, or is your IP switched off for a few hours (a null route)?
2. A Minecraft-aware protection proxy. Players connect to the provider's edge, which passes clean traffic to your hidden server. Your server then needs each player's real IP back. Some providers ship their own plugin for that, many use the PROXY protocol: haproxy-protocol = true in velocity.toml, or proxies.proxy-protocol: true in Paper's config/paper-global.yml when there is no Velocity. Turn it on only when the protection really sits in front, because a server that expects PROXY headers turns away normal connections.
3. Your own discipline. A hidden IP stays hidden only if nothing else points at it: an old DNS record, the Bedrock address, the Votifier port, a website on the same machine, or a staff member who pasted the numeric IP into Discord.

Check Bedrock separately. Geyser needs its own port, 19132 by default, and the current GeyserMC port forwarding guide asks for it to be open on both TCP and UDP. Some protection services cover Java's TCP traffic only.

Questions to ask a host before you pay

1. Which CPU model will my server run on, and are its cores dedicated to me or shared? Look the model up in a public single-thread benchmark.
2. What kind of storage? NVMe SSDs keep chunk loading and saves fast.
3. Which locations can I choose, and is there a test IP I can ping?
4. Can I get extra ports? NuVotifier needs a TCP port (our NuVotifier guide), Geyser needs 19132 (our Geyser guide), and a web map needs one too.
5. Do I get full file access over SFTP, and can I upload my own server jar?
6. Which Java versions can I pick? Minecraft 26.1 and later need Java 25, according to PaperMC's requirements table.
7. How often are backups made, where are they stored, how long are they kept, and can I restore one myself?
8. What does DDoS protection filter, does it include UDP, and what happens during a large attack?
9. Can I move to a bigger plan without migrating files by hand?
10. What is the refund window, and what happens if the node I am on gets crowded?
11. When is support available in my time zone, and how fast does it answer?

A host that answers the first question with "high performance CPUs" and nothing else has told you something too.

Moving hosts without losing players (DNS and SRV records)

Players save your address in their multiplayer list, so the address must belong to you, not to the host. Use your own domain from the first day and a move becomes a DNS change instead of a lost community.

1. A day before the move, lower the TTL of your server's DNS records to 300 seconds, so the change spreads in minutes when you make it.
2. Set up the new server, copy plugins and configs, and test it by its numeric IP.
3. Announce a short downtime. Stop the old server first, then copy the world, so nobody keeps playing on a copy you are about to throw away. Since Paper 26.1 the Nether and the End live inside the world folder under world/dimensions/. Older Paper versions keep them next to it as world_nether and world_the_end.
4. Start the new server and point your DNS record at it.
5. Keep the old machine for a day or two if you can, running a tiny server whose MOTD says you moved, so players with an old cached address see a message instead of a timeout.
6. If the Votifier host or port, or the Bedrock address, changed, update them on every list you use. On serverlist.net that is in your dashboard, and the Votifier tester confirms votes arrive.
7. Check from outside with the server status tool, then watch your listing's public stats page: a step in the response time chart is the move, a long gap in uptime means something went wrong.

If your host gives you a port other than 25565, an SRV record lets Java players keep typing a clean address. The record name is _minecraft._tcp. followed by the address players type:

_minecraft._tcp.play.example.com.  300  IN  SRV  0 5 25577 node7.example-host.net.

The fields after SRV are priority, weight, port and target. The target must be a hostname with its own A record, not a bare IP. Move hosts and you change only that line, and players never see a new port.

Two limits. Bedrock ignores SRV records, so Bedrock players type the host and the port themselves (19132 unless your host assigned another). And if your DNS is on Cloudflare, set the Minecraft records to DNS only, the gray cloud: Cloudflare's normal proxy carries web traffic, not Minecraft connections.

If the new machine lags too, find the cause before you fix it. A bigger plan does not cure a hopper farm.