Back to articles
Cloud

Self-hosting vs cloud — TCO analysis of Cloudflare R2 and an OVH VPS on real numbers

S
Synlogia Team
••7 min read•
Self-hosting vs cloud — TCO analysis of Cloudflare R2 and an OVH VPS on real numbers

Self-hosting vs cloud — TCO analysis of Cloudflare R2 and an OVH VPS on real numbers

Deciding where to put the storage layer of an application — in object storage in the cloud such as Cloudflare R2, or on a dedicated VPS — tends to be the first serious architectural conversation in any B2B project. From our perspective at Synlogia, that conversation rarely boils down to a simple "which is cheaper". Instead, we compare total cost of ownership (TCO) together with traffic, operations, engineering time and contractual obligations. This article shows how to lay out such an analysis on real, publicly verifiable numbers.

Two billing models — the foundation of the difference

Cloudflare R2 and a classic VPS are two opposite cost models:

  • R2 (usage-based) — you pay for what you store and for operations (Class A — writes/lists, Class B — reads). Cloudflare does not charge for egress, the outgoing traffic to the internet (Cloudflare, n.d.).
  • VPS (flat-rate) — you pay a fixed monthly amount for a machine of a given specification (vCPU, RAM, disk, transfer), independent of how much you actually consume.

The first model rewards projects with unpredictable or low volume; the second — projects under stable load for which a predictable bill matters more than elasticity.

Real numbers — Cloudflare R2 (public pricing)

According to Cloudflare's public pricing page (Cloudflare, n.d.), R2 is billed as follows:

  • Storage: USD 0.015 per GB-month (after the 10 GB free tier in the Standard plan).
  • Class A operations (PUT, COPY, POST, LIST): USD 4.50 per million requests.
  • Class B operations (GET, HEAD): USD 0.36 per million requests.
  • Egress: USD 0.00 — Cloudflare promotes the absence of egress charges as a key differentiator versus competitors.

For an average application with 200 GB of data, 5 million reads and 500 thousand writes per month, the bill looks as follows: (200 × 0.015) + (5 × 0.36) + (0.5 × 4.50) ≈ 3.00 + 1.80 + 2.25 = 7.05 USD/month. Additionally, full egress of 1 TB to the internet generates no extra cost — a fundamental difference relative to classic AWS S3, where, according to Amazon Web Services documentation, the first 1 TB of internet egress is USD 0.09 per GB in most regions (Amazon Web Services, n.d.).

Real numbers — a production-class VPS

On the self-hosting side, let us focus on a typical machine for the storage layer. VPS plans on the European market (OVHcloud, Hetzner, Contabo) cover machines with 4–8 vCPU, 8–24 GB RAM, 100–400 GB SSD and several terabytes of transfer. Prices for that class today fall in the ~10–30 EUR/month range (per public pricing pages of providers as of the date of writing).

The server alone, however, is not the full bill. TCO must add:

  1. Backups — either the provider's built-in snapshots (from 1 EUR/month) or an independent mechanism such as rsync/restic to a second region.
  2. Monitoring and alerting — Prometheus + Grafana (self-hosted, "free" only in money), or an external service (Better Stack, UptimeRobot — from a few EUR/month).
  3. Engineering time — system maintenance (OS updates, certificates, hardening). In our practice we assume 2–4 engineer hours per month for a stable machine after deployment.
  4. No SLA on a single VPS — a single host without replicas is a risk that must either be accepted or mitigated by a second machine and a load balancer (and the bill doubles).

A realistic full TCO of a single, well-maintained VPS at small-company scale is 20–60 EUR/month in hard cost plus a few engineering hours, the hourly cost of which depends on the team's internal rates.

When R2 wins

  • The application serves files publicly (product images, PDF documents, CSV exports) — no egress eliminates the biggest cloud trap.
  • Volume is small or highly variable — a startup or MVP does not want to commit to a server flat fee while using only 5 % of it.
  • The team has no DevOps competence — R2 requires bucket and key configuration, but does not require maintaining the OS, kernels and certificates.
  • High availability is required without standing up your own redundancy.

When VPS wins

  • Data operations are intensive and predictable — for millions of reads daily a VPS flat fee is cheaper than usage-based billing.
  • The application needs to run additional logic next to storage (on-the-fly compression, thumbnail generation, transformations) — a dedicated server lets you run ImageMagick or ffmpeg without a cloud-specific stack.
  • Regulatory or trust requirements demand keeping data "under your own key" in a specific jurisdiction.
  • Infrastructure sovereignty — public-sector or local-government projects increasingly require EU location and identifiable infrastructure ownership in their specifications.

Hybrid — in practice the most common choice

In our deployments we most often recommend a mixed configuration: a VPS for the application and database (where steady performance and low inter-service latency matter), with R2 or an equivalent object storage for static assets and backups. This layout combines two advantages: predictable cost for the application core and elasticity for storage that scales with the number of users.

For illustration: in our own projects, moving static storage from local bind mounts to R2-class object storage reduced pressure on VPS disk monitoring and made it possible to deploy more often without risking partition exhaustion.

Decision checklist

  1. Estimate the real volume of data and operations (not a default safety margin).
  2. Compute the cost of both variants for 12 months with a 30 % growth buffer.
  3. Add engineering cost (hours × rate) to the VPS variant, and egress cost at the realistic traffic pattern to the cloud variant.
  4. Check the client's contractual requirements regarding data location and SLA.
  5. Pick the model that matches the phase of the project — early stage scales through the cloud, mature stage through VPS.

A good TCO analysis is not a bookkeeping exercise but part of architecture. The decision made today will cost or save money over the next 24–36 months of system operation.

Bibliography (APA 7)

  • Amazon Web Services. (n.d.). Amazon S3 pricing. Retrieved February 14, 2026, from https://aws.amazon.com/s3/pricing/
  • Cloudflare. (n.d.). R2 pricing. Retrieved February 14, 2026, from https://developers.cloudflare.com/r2/pricing/
  • Hetzner Online. (n.d.). Cloud servers. Retrieved February 14, 2026, from https://www.hetzner.com/cloud
  • OVHcloud. (n.d.). VPS — plan comparison. Retrieved February 14, 2026, from https://www.ovhcloud.com/en/vps/
#cloud#tco#self-hosting#r2#vps