How much does a VPS cost in 2026?

How much does a VPS cost per month

How much a VPS costs depends less on the specs than on who is selling it and how they fill their hardware. As of July 2026, a virtual private server (VPS) with 2 vCPU and 4 GB of RAM sells for roughly $4 a month at the bottom of the market and roughly $40 a month at the top. The two order pages can list the same numbers. What arrives after checkout is not the same machine.

The gap is not mysterious. It comes from how many customers share each physical core, what class of storage sits under your disk, how much network traffic you are allowed to use, and how much human support is priced into the plan. This guide walks through each one, then gives you commands to check what you actually received.

Price bands as of July 2026

[
  {
    "label": "Entry, shared cores",
    "usd_low": 3,
    "usd_high": 8
  },
  {
    "label": "Mainstream, 4 to 8 GB",
    "usd_low": 10,
    "usd_high": 35
  },
  {
    "label": "Memory heavy, 16 to 32 GB",
    "usd_low": 30,
    "usd_high": 90
  },
  {
    "label": "Dedicated cores",
    "usd_low": 17,
    "usd_high": 200
  }
]

These are list prices for Linux VPS plans on monthly billing, gathered in July 2026, with the dedicated starting price updated below. Treat them as bands, not quotes. The entry band starts near 3 dollars a month, and that is where oversubscription is heaviest. Dedicated core plans rechecked on October 29, 2026, include offers near 17 dollars a month. Netcup's RS 500 G12.5 lists two dedicated cores at €10.99 per month for a 24-month term; the one-month option adds €3.88, making it €14.87 including 19% VAT, about $17 at the ECB's October 28 reference rate. The chart runs to 200 dollars a month, but larger dedicated plans cost more. With dedicated vCPUs you are paying for CPU resources reserved for you alone, which at Hetzner means one thread of a physical core rather than the whole core.

Prices in this market move, and any specific figure ages badly, which is why there is no provider by provider table here. The structure below is what stays true.

Oversubscription sets most of the price

A vCPU is a scheduling slot on a physical core, not a core. Hosts sell more slots than they have cores, because most servers idle most of the time. That practice is oversubscription, and the ratio is the single biggest reason two plans with identical spec sheets can cost ten times apart.

At a modest ratio your vCPU behaves close to a real core. At an aggressive ratio your process waits while other tenants on the same host run, and the hypervisor reports that wait to you as steal time.

vmstat 1 5

The st column is steal time as a percentage. On an idle server it should read 0 or very close to it, because hypervisor housekeeping can show a sliver of steal even on a host that is not oversold. Double digit steal while your own load is light means the host is busier than the plan suggests. Usually that is other tenants competing for the same core, sometimes an overloaded hypervisor, and on a burstable plan it can be a spent CPU credit balance. Whichever it is, the plan is cheap because the core is not really yours. Run this at your busy hour, not at 3am, because the neighbours you compete with keep their own schedule.

Memory is oversubscribed too, through ballooning and host swap. Check what you were actually given:

free -h
lscpu | grep -E 'Model name|Hypervisor vendor|^CPU\(s\)'

Hypervisor vendor: KVM tells you it is full virtualisation, so you get your own kernel. A container based VPS on OpenVZ or LXC shares the host kernel, costs the host less to run, and leaves kernel modules and access to virtualisation devices under the host's control. That is a real capability difference hiding inside a price difference.

Storage class moves the price more than capacity does

Two plans can both say "80 GB SSD" and behave nothing alike. Local NVMe attached to the same chassis answers a random 4k read in well under a millisecond. Network attached block storage crosses a network first, but its latency depends on the backend: Amazon EBS io2 Block Express delivers sub-millisecond latency. Databases, mail queues and anything that calls fsync feel that difference immediately. The mechanics are covered in the difference between NVMe and SATA SSD storage on a VPS.

Look at what you have, then measure it:

lsblk -d -o NAME,ROTA,SIZE,MODEL
sudo apt update && sudo apt install -y fio
fio --name=randread --rw=randread --bs=4k --size=512M --iodepth=32 \
  --ioengine=libaio --direct=1 --runtime=30 --time_based --group_reporting

Read the IOPS= value and the clat percentiles block in the output. A p99 completion latency in the single digit milliseconds on a quiet box is fine. A p99 in the tens of milliseconds usually means you are sharing a busy storage backend, whatever the product page called it. Rule out the ordinary causes first: a disk restored from a snapshot with lazy initialization incurs extra latency when blocks are first read, and a plan with an IOPS or throughput cap queues requests once you reach it. Amazon EBS fast snapshot restore creates fully initialized volumes immediately and removes that first-read penalty.

Bandwidth is a bill you only see later

Most VPS plans include a monthly transfer allowance. As of July 2026, plans in the mainstream band commonly include somewhere between 1 TB and 8 TB a month, and the overage rate per additional terabyte varies enough between hosts that it is worth reading before you need it. Two terms are doing a lot of work in these listings, so separate them. Port speed is how fast you can send, in Gbps. Transfer allowance is how much traffic the host lets you move in a month, and some hosts count what you receive against it as well as what you send. A 10 Gbps port with a 1 TB allowance runs out in about thirteen minutes at full rate.

"Unmetered" usually means no counted allowance but a lower port speed. "Unlimited" is a marketing word with a fair use clause behind it. Check whether inbound traffic counts, and whether the allowance is pooled across your servers or applied per server.

sudo apt install -y vnstat
vnstat -m

vnstat -m prints a monthly total per interface once it has collected data for a while, which lets you check the host's counter against your own before an invoice surprises you.

Support and the network you never see

An unmanaged VPS gives you a root login and a control panel. Everything above the hypervisor is yours: patching, firewall, backups, recovery. A managed plan puts staff behind those, and salaries are the most expensive input any host has, so managed plans sit one band higher for the same specs. Neither model is wrong. Buying unmanaged and then paying somebody hourly at 2am is what goes wrong.

Network quality is the other invisible input. Transit from tier 1 carriers with direct peering costs a host far more per megabit than cheap blended bandwidth, and the difference reaches you as latency and packet loss rather than as a line on the spec sheet. Test it from where your users are before you commit, using mtr or a plain ping sampled over a few hours.

Billing games to read for

The listed price and the price you pay in year two are often different numbers. Watch for these:

• A promotional rate for the first term that renews at a higher list price. Ask what the renewal figure is, in writing.
• A setup fee on the first invoice, which is most common on the cheapest plans.
• IPv4 addresses billed separately. Addresses are scarce and genuinely cost money now, so a plan that includes one is quietly more generous than a plan that does not.
• Backups, snapshots and extra block storage sold as add-ons, often per gigabyte per month.
• A refund window measured in days. That window is one cheap way to test a host.

On yearly against monthly billing: where a host sells an annual plan, paying twelve months up front lowers the effective monthly rate, and as of July 2026 that discount usually lands between 10 and 30 percent. Not every host discounts prepayment. Some, Linode among them, let you deposit credit in advance that pays ordinary monthly invoices. The trade is that you are committed to a host you may not have tested under your real workload. Buy one month, run your actual application, measure steal time and disk latency, then convert to annual if it holds up.

What to actually compare

Ignore the headline price and compare these instead:

• The renewal price per GB of RAM, not the promotional price.
• Steal time under your own load, measured on the box.
• Storage class and measured p99 latency, not the word "SSD".
• Transfer allowance, overage rate, and whether inbound traffic counts.
• Whether IPv4, backups and snapshots are included or extra.
• The refund window, and how long a support ticket takes to get a human reply.

A checklist for the first hour on any new server, including the security steps that belong in the same session, is in the first ten minutes on a new VPS.

When the cheap VPS stops being the cheap option

Cost per unit of work matters, not cost per month. A $5 plan that loses 30 percent of its CPU to steal and stalls on disk may need three instances to carry the load one honest $20 plan carries alone, and you now pay for three sets of patching and monitoring as well. Past a certain size the same arithmetic points the other way, off shared virtualisation entirely, which is the comparison in a VPS against a dedicated server.

Start from the workload. Decide what you are running, size it, then shop. Picking a price first and hoping the work fits inside it is how people end up buying twice. If you are still deciding what to put on the server, the self hosting shortlist for 2026 is a reasonable place to start.

FAQ

How much does a VPS cost per month?

As of July 2026, entry Linux VPS plans with shared cores run about $3 to $8 a month and mainstream plans with 4 to 8 GB of RAM run about $10 to $35. As of October 29, 2026, dedicated core plans are available near $17: netcup's RS 500 G12.5 costs €14.87 on a one-month term including 19% VAT. Those are list prices on monthly billing. The same specs cost more where cores are less oversubscribed, storage is local NVMe, and support is staffed by people.

Why is one VPS $3 and another $50 for the same specs?

Because a vCPU is a scheduling slot, not a core, and each host decides how many slots to sell per core. Aggressive oversubscription, network attached storage instead of local NVMe, cheap blended bandwidth and unstaffed support all cut a host's cost per plan. You see the result as steal time in vmstat, as high p99 disk latency in fio, and as slow ticket replies.

Is a yearly VPS plan worth it?

It is worth it once you have tested the host, and only if the host actually discounts it. Where an annual discount exists, it usually lands between 10 and 30 percent as of July 2026, but some hosts offer no discount and simply hold prepaid money as credit against monthly invoices. Run one month first with your real workload, check steal time and disk latency, confirm the renewal price in writing, then commit for a year.

What is the cheapest cloud hosting?

The July 2026 entry band above covers shared core VPS plans at $3 to $8 a month, sometimes on container virtualisation rather than KVM. It is fine for a small static site, a chat bot, or a personal VPN. It is a poor fit for a busy database or anything latency sensitive, because the saving comes directly from resources you share with strangers.

Are there hidden costs in VPS hosting?

The common ones are renewal pricing above the promotional rate, setup fees, IPv4 addresses billed separately, backups and snapshots as per gigabyte add-ons, and bandwidth overage once you pass the included transfer allowance. None of these are hidden in the legal sense. They sit below the fold on the order page, so read the renewal terms and the transfer allowance before you compare two prices.