Vultr vs Hetzner (2026): Regional Costs, Traffic & Backups

Vultr vs Hetzner is not settled by putting a dollar headline next to a euro headline. Choose the region, architecture, usable memory, disk and operating responsibilities first. Vultr can suit a workload that benefits from its available locations and account-level transfer pool. Hetzner can suit a workload whose selected regional quote and traffic allowance fit. Neither removes the customer's responsibility for the Linux system, application, monitoring or recovery.

This comparison uses official sources checked October 16, 2026. It is documentary research, not a server benchmark, stock check, latency test or completed migration. The examples concern unmanaged Linux on shared x86 CPU with two vCPUs, 4 GB RAM and an 80 GB disk. CPU vendors and disk types differ. USD and EUR amounts remain separate, taxes are excluded where stated, and monthly examples are budgets rather than measured invoices. The Vultr review and Hetzner review provide wider context.

If the team wants managed deployments rather than responsibility for servers, start with the hosting category. A virtual machine is not a like-for-like replacement for every workflow supplied by Vercel or Render. Keeping those operational differences visible is more useful than claiming a cheaper VM automatically lowers the whole cost of running an application.

Vultr's public plan catalog lists vc2-2c-4gb at $20/month with two vCPUs, 4 GB RAM and 80 GB SSD; its location associations include Frankfurt and Singapore. The region catalog identifies those region codes. This is the Regular Cloud Compute configuration, not a High Frequency or dedicated-CPU offer substituted to improve a comparison. Catalog association does not guarantee deployable stock for the account or image you choose.

Hetzner's Regular Performance table lists CPX22 with two AMD vCPUs, 4 GB RAM and 80 GB NVMe. The selected Nuremberg offer is €19.99/month including Primary IPv4; Singapore is €26.99. Its table explicitly says IPv4 included and VAT excluded, despite a contradictory generic VAT footer. The calculator separates the VM amounts, €19.49 and €26.49, from the €0.50 Primary IPv4. Verify the final tax treatment for your billing country.

Germany is country-matched here, not city-matched: Frankfurt and Nuremberg are different locations. The selected US Hetzner CPX21 instead has three vCPUs and 4 GB RAM, while CPX11 has two vCPUs but 2 GB RAM; neither should be relabeled as an identical two-vCPU/4-GB offer. Similarly, Intel/shared/SSD and AMD/shared/NVMe do not establish equal throughput. The Node.js hosting guide can narrow the platform type, but capacity and location still need a current account-level check.

Vultr's billing documentation describes hourly charging and a one-hour minimum, with the documented 672-hour cap for ordinary Cloud Compute. However, the selected plan's public API reports $0.027/hour while the pricing page displays $0.030/hour; both show $20 monthly. Those values do not reconcile into one unambiguous partial-month invoice. The budgets below use the advertised monthly amount and explicitly do not predict hourly or short-lived-server bills.

Hetzner's billing FAQ charges the lesser of its monthly cap or rounded-up hourly usage. It also distinguishes Primary IP resources from the VM and charges retained unassigned IPv4 addresses. Its selected EUR offers are not converted into dollar prices in this comparison. A currency conversion, exchange fee and customer-specific tax could change the apparent order of the headline amounts; none is invented here.

Write down compute, IPs, automatic backups, retained snapshots, attached storage, outbound excess and optional services as separate budget lines. Add operating time and migration overlap outside the provider bill. The SaaS spending framework is useful for distinguishing a recurring platform charge from the whole workflow cost. Pricing a server without the services needed to recover it is not a complete production budget.

Vultr's published transfer policy documents free inbound traffic, global account-level pooling, an additional 2 TB monthly account allowance and $0.01/GB outbound excess. All allocations accrue hourly, including the account allowance. This is a policy introduced in 2023 and rechecked now, not a new 2026 offer. Do not assume a newly created server immediately receives every byte of the month-end pool.

The selected Vultr plan is advertised as 3 TB transfer, while its API reports 3072 GB. Use compatible provider billing units rather than silently swapping decimal and binary conversions. The useful excess calculation is billable outgoing traffic minus the eligible accrued pool, floored at zero, then multiplied by the overage rate. Other workloads in the same account can consume the shared pool; account-level capacity is not a dedicated reservation for one application.

Hetzner bills outgoing traffic but excludes incoming and defined internal traffic. Same-network-zone public traffic between its Cloud servers is treated differently from public traffic across network zones. Its selected CPX22 includes 20 TB in Nuremberg but 1 TB in Singapore; published excess rates are €1/TB in Europe/US and €7.40/TB in Singapore. Excess is rounded up in 100 MB blocks, not necessarily whole terabytes. A region label alone is insufficient: exact US plan allowances can differ too.

The reviewed Hetzner sources did not settle project pooling or the partial-month allowance formula, so neither is assumed. The examples use one continuously retained server for a full month. Usage alerts are notifications, not a hard spending limit. If transfer cost is the reason for leaving a platform, use the Vercel alternatives guide to keep managed-service scope separate from these infrastructure-only calculations.

Both scenarios assume the selected VM exists for the whole month, traffic is spread through the month, automatic backups remain enabled and one IPv4 is present. They also include twenty provider-billed compressed GB-months of separate snapshot storage. There are no attached volumes, paid operating-system licences, extra services, taxes or operator wages. These are calculated advertised-cap budgets, not product tests or promised invoices.

In Germany with 500 GB external outgoing traffic, Vultr's example is $20 compute plus $4 automatic backups plus $1 snapshots, or $25.00. Hetzner's is €19.49 compute plus €0.50 IPv4, €3.898 backups and €0.286 snapshots, totaling €24.174 before final rounding, or €24.17. The two totals cannot establish a cheaper provider without an explicit currency basis. Matching RAM and disk size also does not make the performance identical.

For Singapore with 2 TB outgoing traffic in the provider's billing units, the selected Vultr instance remains within its advertised allowance, so the example is again $25.00. Hetzner's 1 TB allowance leaves 1 TB at €7.40, producing €39.974 before rounding, or €39.97. This example deliberately stays below Vultr's instance allowance without relying on the extra account pool. For a burst early in a partial month, obtain the actual accrued allowance instead of copying this result. The startup hosting guide can help decide whether operating a VM is justified in the first place.

Both documented automatic-backup charges are twenty percent of the parent VM price, not twenty percent of an invoice containing IPs, traffic and separate storage. Vultr's backup FAQ describes scheduling choices and excludes attached storage from the instance backup. Hetzner's backup and snapshot FAQ explicitly excludes Volumes from both image types. A database placed on an attached volume therefore needs its own protection plan.

Hetzner provides seven rotating daily backup slots tied to the server; deleting the server deletes those backups. A separately retained Snapshot survives server deletion and continues to incur storage charges. That distinction matters when following advice to snapshot and delete a powered-off machine. Confirm that the surviving recovery object actually exists before removing the only running source, and retain independent application data where the recovery requirement calls for it.

A VM image also does not prove application consistency. An active database, external object storage and a service holding secrets may need separate, coordinated backups. In a future authorized restore exercise, verify the database, uploaded files, ownership, permissions and application startup. This article did not run a restore. For workloads connected to DigitalOcean or another provider, evaluate each service's coverage separately instead of assuming one VM snapshot includes the entire stack.

Vultr's snapshot FAQ prices compressed stored size at $0.05/GB-month. Hetzner's current pricing lists €0.0143/GB-month, and its billing FAQ also bases charges on compressed storage and the time it exists. The nominal 80 GB disk, filesystem usage, compressed billed size and restore-target disk size are different measurements. The twenty-GB snapshot assumption in the budget is a billing input, not a compression guarantee.

The Vultr data-portability guide explicitly says snapshots and backups cannot be downloaded directly. Its documented route restores a temporary instance and retrieves data through operating-system tools. That can add temporary compute, transfer and administration costs. An image identifier or API resource URL is not evidence of a transferable raw disk download, and a saved provider-native snapshot is not automatically a vendor-independent backup.

Hetzner documents restoring or creating servers from its native images, with architecture constraints; that is not proof of a seamless cross-provider image migration. Both providers describe regional reuse inside their own systems, which does not establish available destination stock or a complete change of data location. Hetzner's Singapore image storage remains in Singapore under its documented rules. Keep file, database, configuration and secret recovery separate in the software evaluation checklist.

Both providers bill a server while it exists, including when powered off. Stopping the guest operating system or clicking power off is not cancellation. Deleting a machine stops that resource's future compute accrual but can leave billable snapshots, attached storage or IPs behind. Hetzner explicitly bills retained unassigned Primary IPv4 addresses. Inspect the account's resources rather than treating a successful VM deletion message as a complete zero-bill confirmation.

Do not delete first and discover missing data later. Record the recovery copy, its access owner, required credentials, retention decision and expected remaining charges. The person approving shutdown should know which backups disappear with the server and which snapshots remain. The cloud-storage guide distinguishes storage from a backup process; merely placing a file somewhere is not evidence of tested recovery. Refund or unused-credit eligibility is not established by stopping future resource charges.

A shared vCPU count is not a benchmark. CPU generation, neighbors, disk behavior, network path and the application can dominate the result. Hetzner's server FAQ discusses shared resources, rescaling and architecture restrictions; Vultr's product families likewise differ. Do not compare a shared plan against dedicated cores or substitute a High Frequency result into the Regular plan budget.

If latency or throughput decides the purchase, run an authorized small evaluation in the actual target region and record the plan identifier, image, software versions, dataset, concurrency and test period. Measure the application behavior and restore time you care about, not just a provider's marketing score. Germany-to-user latency cannot be inferred from Singapore pricing. Stock and account restrictions can change the feasible plan before any performance comparison begins.

The customer still needs an owner for OS patches, SSH access, firewall rules, monitoring, database maintenance and backups. A provider compliance label does not certify a particular deployed application or settle its contractual data-location requirements. If that operational burden is unsuitable, compare Railway or Render as different service models instead of choosing the lowest compute amount and leaving administration unassigned.

Map the existing system before moving: compute architecture, databases, uploads, background jobs, DNS, certificates, mail delivery, monitoring and secrets. Obtain a consistent data copy, restore to an authorized destination, validate record counts and file access, then rehearse cutover and rollback. Allow for an overlap period where both providers bill. An application repository is only one part of the system; it does not automatically contain production state or recovery credentials.

Choose Vultr provisionally if the actual available regional configuration and accrued transfer pool fit the workload and the team can operate it. Choose Hetzner provisionally if its available regional plan, currency, IP and transfer terms fit better. Confirm the hourly discrepancy before a short-lived Vultr deployment and the unresolved allowance rules before a partial-month Hetzner estimate. Neither choice guarantees faster pages, uptime or search rankings. A current quote and a measured, permission-safe workload test are the decision evidence.