Some links in this article are affiliate links. We earn a commission at no extra cost to you when you purchase through them. Full disclosure.

Hosting offers are comparable only after matching workload, resources and operational responsibility. A self-managed VPS, metered application service, managed PHP host and GPU instance are not equivalent products.

This research-based update checks first-party billing and product scope on October 2, 2026. It withdraws the old non-matched “actual monthly cost” table, mixed-currency totals and universal cheapest recommendations. No provider throughput benchmark was performed.

Disclosure: We may earn a commission if you sign up through links on this page.

Start with the workload class

Provider/product	Class to evaluate	What must be matched
Railway	Metered managed application services	Runtime, average resource usage, workers, database, volumes and egress
Cloudways	Product-specific managed hosting	Flexible PHP/WordPress versus Velocity Node.js scope, resources and managed services
Contabo	Self-managed CPU VPS	CPU allocation, RAM, disk, region, setup/renewal costs and operator responsibility
Hetzner	Self-managed cloud/VPS	CPU family, RAM, disk, region, transfer and operator responsibility
Vultr	Specific compute/VPS or GPU product	Shared/dedicated CPU, hardware, region, transfer and storage
RunPod	Pods or Serverless GPU	GPU variant/count, VRAM, host resources, worker mode and storage
Render	Managed application services	Service type, instance class, database and bandwidth
DigitalOcean	Droplets or separately priced managed products	CPU/RAM class, storage, transfer, backups and database scope
IONOS	Selected VPS/cloud/server offer	Region, term, renewal, resource allocation and management scope
AWS	Selected EC2/managed/GPU services	Exact instance/count, region, storage, transfer and contract
ScalaHosting	Selected managed cloud/VPS offer	Management scope, configuration, renewal and backups

These are purchasing categories, not price or quality rankings.

Railway: a worked billing example, not a traffic prediction

Railway’s plan/rate documentation gives Hobby a $5 monthly minimum and Pro a $20 minimum, each including that amount of usage. Rates include $10/GB-month RAM, $20/vCPU-month CPU, $0.15/GB-month volumes and $0.05/GB egress.

Hypothetical resource budget: one GB-month memory + 0.1 vCPU-month CPU + 5GB-month volume + 10GB egress equals $10 + $2 + $0.75 + $0.50 = $13.25. For these eligible items, Hobby’s base minimum/usage charge is $13.25 and Pro’s $20, before taxes or separate services.

This is not a measured app bill or a guarantee that one configured GB produces that exact usage. Add database, worker and other service usage independently.

Evaluate Railway for a metered API and worker deployment if the managed-platform operating model fits.

Cloudways: select the correct product first

Flexible application hosting covers PHP-based applications; Velocity is a separate Node.js offering. Do not use a Flexible server price as proof of general Python, GPU or arbitrary-container support.

Compare Cloudways products for your PHP or Node.js hosting requirements and verify the exact product, database scope, underlying configuration and backup/transfer charges.

Self-managed VPS comparisons

For Hetzner, Vultr, DigitalOcean and IONOS, quote the same required RAM, CPU type, region, disk and transfer allowance. Record whether quoted monthly prices are caps, commitment equivalents or introductory offers. Keep currency and taxes separate. Two plans with different CPU scheduling, disks or management are not benchmark equivalents.

Use direct evidence: Hetzner cloud, Vultr compute documentation, DigitalOcean Droplet billing and IONOS VPS.

If self-management is acceptable, evaluate Vultr for a configuration-matched VPS workload or evaluate DigitalOcean for a configuration-matched Droplet workload. For a separate IONOS quote, compare IONOS VPS terms for your required configuration without assuming an advertised promotion equals renewal cost.

Also compare Contabo for self-managed CPU workloads

Contabo’s VPS specifications and pricing are direct evidence for its current CPU plans. Include location, storage option, setup fees, term and tax when matching quotes. You remain responsible for operating the server; a low advertised rental price does not establish measured performance or lower total cost.

Evaluate Contabo for a self-managed VPS workload when your application or worker fits CPU compute. This is not a GPU recommendation. Size and test any local inference separately; enough RAM alone does not prove acceptable latency.

GPU compute is a separate budget

RunPod public configurations, Pod billing and Serverless billing are direct sources. Compare exact GPU model/variant/count, worker mode, storage and availability; do not compare one GPU with an eight-GPU instance or assume 24/7 operation.

Compare RunPod configurations for the GPU portion of your workload only when accelerator compute is required. See the repaired RunPod pricing owner.

Managed VPS is another operating model

ScalaHosting’s managed-cloud product is the direct scope source. Evaluate ScalaHosting for a managed VPS configuration after checking included administration, backups and renewal terms. A managed quote should not be compared to raw VM rent without accounting for the service difference.

Build a total-cost budget

Add application compute, workers, database, storage, transfer, backups, monitoring, email, model usage and operational labor. For hosted model usage, budget measured input/output tokens at the chosen provider’s current rates; request count alone cannot establish a bill. Local inference still has hardware, memory, energy and administration costs.

Use a representative pilot and spending controls. This guide gives no monthly cost guarantee from “500 users/day,” universal free production stack or cheapest winner.

Related: Vultr vs DigitalOcean, DigitalOcean alternatives, Cloudways vs Railway vs Hetzner, managed hosting, side projects, GPU providers and model-cost optimization.