Vultr Review 2026: Pricing, Pros, Cons and Alternatives

Buy Vultr when the workload has to sit in a particular city and you will read the plan family, not the lowest sticker. The comparison page lists 33 cloud regions, and some of those cities can carry a different hourly or monthly rate, so check that city in the console before you treat the sheet as the bill.

Regular Performance is previous-generation Intel and a regular SSD, fine when the drive is not the bottleneck. High Performance and High Frequency are the NVMe rows, for when the disk matters. Optimized Cloud Compute is a dedicated vCPU when sharing a core is the problem. VX1 is a separate hourly product the pricing page does not cap at 672 hours, so a full month there can pass the other families' ceiling.

Toolradar data: on October 16, 2026 the VPS and cloud infrastructure statistics list 6 published tools inside a directory of 11,245. Pricing labels mark 5 of those 6 as strictly paid, and 1 of the 6 has a free tier. Vultr has a tool page, and so do DigitalOcean and Linode.

How we compared: six hosts a Vultr buyer actually cross-shops, checked on October 16, 2026. Nothing was provisioned. Where a total adds two published allowances or multiplies a printed rate, that sum is ours. Sibling pages cover the same hosts from the other side of the table: the developer VPS roundup.

Cloud Compute on Vultr is an SSH login on a virtual server, so you patch the machine and you pay while it exists. Regular, High Performance, and High Frequency plans share the physical core, which is the trade for the lower sticker. Optimized plans pin a dedicated AMD EPYC vCPU when that sharing is the complaint. Bare metal, Cloud GPU, block and object storage, a load balancer, and managed databases are separate lines on the same account.

A process that should cost nothing while idle is a different product. Look at the Heroku review or the Fly.io review for a git push that sleeps.

Vultr Kubernetes Engine does not bill the control plane, so a small cluster pays for workers, block storage, and any load balancer you attach. Managed MySQL and managed PostgreSQL both open at $90/mo for 4 GB of memory, 1 vCPU, 30 GB of storage, and no replica. That managed price is a different budget from the VM.

Cloud GPU is on the same pricing page, billed per GPU hour, and the billing note treats GPU months as 730 hours. An L40S configuration on that page starts at $1.671 per GPU hour, a rate for a GPU job rather than a small API. Windows licenses are not in the base VM price, and the add-on varies by plan, so price a Windows image before you treat the Linux sticker as the total.

You pick a location, a plan, and an image, and the hourly rate starts when the server is deployed, whether it is powered on or off. The minimum unit is one hour, so a short test still costs that full hour. Regular, High Performance, High Frequency, and Optimized Cloud Compute cap the month at 672 hours, so an always-on box on those families stops at the sticker. Destroy the instance to release the CPU, RAM, disk, and address, because stop leaves them reserved.

Outbound is the bandwidth meter and inbound is not charged, so uploads cost less here than serving files. The account receives 2 TB of free bandwidth each month, with no carryover, so a quiet month does not credit the next one. Each running instance also accrues its plan's bandwidth by the hour, using 672 hours as the month. The cap at any moment is free bandwidth, plus what has accrued, plus any bandwidth you bought, and an overage is measured against that running total.

An overage applies the moment outbound passes that current cap, so a spike on day two can bill before the full monthly allowance exists. Vultr's rate past the quota is the published per-GB rate, and the bandwidth FAQ states that rate worldwide, so the city can change the VM sticker without changing this overage.

Automatic backups add 20 percent to the instance's base hourly or monthly fee. They cover the instance file system, not attached block storage, so a mounted volume needs its own copy. A snapshot is a separate image at $0.05 per GB per month of compressed size, and it stays until you delete it. If you need the disk after a destroy, take that snapshot first, because Vultr keeps only the two most recent automatic backups once the instance is gone.

The worked example is one always-on regular VM, backups on, plus 1,000 GB of outbound after the combined allowance, because the plan sticker is not the invoice. Prices checked on October 16, 2026 on Vultr's pricing page, the bandwidth FAQ, and the bandwidth cap note.

The regular 2 vCPU, 2 GB plan caps at $15.00/mo ($0.022 per hour if you destroy it early), with 65 GB of disk and 3 TB of instance bandwidth. Twenty percent of that cap is $3.00, so the instance plus backups is $18.00.

Add the account's 2 TB and a full month of that plan is 5 TB before overage, a pool you have at month's end rather than on day one. Another 1,000 GB at the published per-GB rate is $10, and the VM, backups, and that extra 1,000 GB come to $28.00. That $10 assumes the transfer happens after the allowance exists, and an early spike can bill sooner.

High Performance at 2 vCPUs and 4 GB is $24.00/mo with 100 GB of NVMe and 5 TB, the step when the regular SSD is the problem. Optimized general purpose opens at $30.00/mo for 1 dedicated vCPU, 4 GB, 30 GB, and 4 TB, the buy when you want the core to yourself. VX1 general purpose at 2 vCPUs and 8 GB is $0.060 per hour on block storage. A 730-hour month at that rate is $43.80, and the pricing page does not stop VX1 at 672 hours, so this family can outrun a capped plan.

On DigitalOcean the matching small production shape is the 2 GiB Basic Droplet with two vCPUs: 60 GiB of SSD, 3,000 GiB of outbound, and the public address inside the cap shown in the table. Another 1,024 GiB at $0.01 per GiB is $10.24, so that Droplet plus the extra 1,024 GiB is $28.24. RAM units differ (GiB there, GB on the Vultr row above, which also has 65 GB of disk).

Bundled Droplets bill each second after a 60-second minimum and stop at 672 hours. The 512 MiB Basic row is $4/mo if that smaller size is enough.

Akamai's shared sheet puts 1 GB and 1 TB on Nanode, and 2 GB with 2 TB on the next shared size at $12/mo. Past the pool the printed rate is $0.005 per GB, the line to read when the month is mostly files. Their compute terabyte is 1,024 GB, so one of those past the pool is $5.12 and the 2 GB shared VM plus that block is $17.12.

Regional prices can differ, and powering a Linode off does not remove the charge. The Akamai Linode review is the page for the dedicated families that dropped their cap.

For a US Hetzner order, the June 15, 2026 price list, excluding VAT, shows CPX11 at $20.49/mo before the address. Add the primary IPv4 at $0.60/mo and the two lines are $21.09 before tax, already above a small Vultr regular plan. Specs are 2 shared vCPUs, 2 GB, and 40 GB, with at least 1 TB in US sites, and a further terabyte in the EU or the US is $1.20.

The Germany and Finland CPX22 row on that list is $22.99/mo before IPv4, and EU cloud servers include at least 20 TB, which is why Hetzner wins when the city can be there. Shared cost-optimized plans are flagged limited availability on the cloud page, so they are not treated as in stock here.

Lightsail's first Linux bundle that includes a public IPv4 is half a gigabyte, two vCPUs, 20 GB of disk, and 1 TB, at the price in the table. Traffic in both directions fills the allowance, which is stricter than Vultr, where only outbound counts. Only the outbound excess is billed, and in Northern Virginia that starts at $0.09 per GB, so 100 GB over is $9.

The printed transfer is halved in Sao Paulo, Hong Kong, Malaysia, Jakarta, Sydney, and Mumbai, so those cities get a smaller allowance than the US card. OVHcloud's US VPS-1 opens at $4.54/mo for 2 vCores, 4 GB, 40 GB of NVMe, traffic without a terabyte quota, a 500 Mbps public port, and a daily backup, which is the pick when a port ceiling beats a transfer pool.

Vultr fits a developer who will pick the city first and then the plan family. A staging API on the 1 GB regular row, a worker that must live near users in a city the other hosts do not list, and a Kubernetes node pool whose control plane should not be a line item are the jobs this sheet is built for. Destroy the box when the work is done.

It is a weak default when the month is mostly egress and the server can sit in Germany or Finland, where Hetzner's included traffic is the product. It is also a weak default when you wanted the public address folded into a Droplet sticker, or when a stopped server should cost nothing, because stop is not a pause here.

Skip it for a git-push app that should idle at no charge, because that job is the Heroku or Fly.io page, not a VM. A database you do not want to patch belongs on Postgres hosting, because the managed 4 GB plan is a different budget from the VM. A page-only front end is the Vercel review, and a box you administer yourself is the self-hosting tools guide.

Devshot data: this list reaches 28,000+ developers, and the open rate in the audience file is 32%, updated October 18, 2026. Later price moves belong in Devshot's daily brief.