The Cloud Tax: The Same Workload on 6 Clouds (2026)

Cloud price comparisons are usually apples-to-oranges, so here is exactly what we held constant. Every VM is a dedicated-vCPU, general-purpose instance at a 4 GiB-per-vCPU ratio: AWS m5, Azure Dv5, GCP n2-standard, and the dedicated-CPU tiers from DigitalOcean, Vultr, and Hetzner (not their cheaper shared-vCPU lines, which would not be a fair fight). All prices are on-demand, Linux, list — no commitment discounts — in each provider’s canonical US region (AWS us-east-1, Azure eastus, GCP us-central1). DigitalOcean and Vultr price flat across US regions; Hetzner is EU-native, shown at its US-dollar list price. Monthly cost is the hourly rate × 730 hours.

These are the same unit prices the SpendArk calculator uses, verified against each provider’s public pricing on 2026-07-25. Where a provider genuinely doesn’t sell a product (Hetzner and Vultr have no managed CDN, for example), we leave the cell blank rather than inventing a number — the full methodology is documented here.

Start with a large general-purpose box: 16 vCPU and 64 GB of RAM, the kind of instance that runs a busy application server or a mid-size database. Same specs everywhere. Here is the monthly bill.

Hetzner’s CCX43 lands at $325.49/mo. The same specs are $560.64/mo on both AWS (m5.4xlarge) and Azure (D16s_v5). Google Cloud (n2-standard-16) sits in between at $453.74, and Vultr ($480.00) and DigitalOcean ($504.00) fill the middle. That’s a 1.7× spread — $2,821 a year — for a machine that boots the same OS and runs the same containers.

The pattern holds at the small end too. A 2 vCPU / 8 GB instance is $70.08/mo on AWS and Azure versus $50.49 on Hetzner, $56.72 on GCP, $60.00 on Vultr, and $63.00 on DigitalOcean — a 39% premium at the bottom of the range that widens to 72% at the top, because the hyperscalers’ per-vCPU rate scales linearly while the independents give you volume headroom.

One honest caveat: this is on-demand list price. Commit to a 1- or 3-year AWS Savings Plan or Azure Reserved Instance and you can cut the hyperscaler compute number by 30-50%, which closes most of this gap — if you can predict your usage that far out and are willing to lock capital into it. The independents are already at that price with no commitment. See reserved vs spot vs on-demand for how far commitments actually get you.

Compute is where you notice the tax. Egress — data leaving the cloud to your users — is where it becomes indefensible. It’s also the line item that ambushes teams, because you don’t provision it: it just accrues with every download, video frame, and API response. Here is the bill to serve 1 TB (1,024 GB) of outbound CDN traffic in a month.

AWS CloudFront charges $0.085/GB, Azure CDN $0.081, and Google Cloud CDN $0.080. DigitalOcean’s Spaces CDN is $0.01/GB — one eighth of the price. Serve a terabyte a month and that’s $87 versus $10; serve 20 TB and it’s $1,740 versus $205. Nothing about the bytes is different. This is the single biggest structural reason egress-heavy products — video, downloads, image-heavy apps — are quietly moving off the hyperscalers. Our cloud egress costs guide breaks down why the hyperscalers price transfer the way they do.

The tax shows up on the smaller line items too, in the same direction. Standard object storage is $0.023/GB-month on AWS S3, $0.020 on GCP Cloud Storage and DigitalOcean Spaces, $0.018 on Vultr, and just $0.0068 on Hetzner Object Storage — a 3.4× spread. Azure Blob Hot sits at $0.0184. For a team parking 10 TB of assets, that’s $235/mo on AWS versus $70 on Hetzner.

A managed load balancer follows suit: $16.43/mo on AWS (ALB), $18.25 on Azure, $18.26 on GCP, $12.00 on DigitalOcean, $10.00 on Vultr, and $8.57 on Hetzner. The hyperscaler LB costs roughly twice the Hetzner one. Individually these are rounding errors; across a full stack with redundant load balancers, multi-terabyte storage, and steady egress, they compound into the difference between a $2,000 and a $4,000 monthly bill.

Managed databases are the one place the comparison gets genuinely hard to normalize — AWS RDS, Azure Flexible Server, and Cloud SQL are true managed control planes, while a fair Hetzner/Vultr equivalent is self-hosted Postgres on a VM. Among the three that offer apples-to-apples managed relational DBs, a 2 vCPU / 8 GB Postgres instance is $129.94/mo on AWS and Azure versus $101.18 on GCP. We keep databases out of the headline table below for exactly this reason — the calculator models each provider’s real DB shape so you don’t have to.

Every number below is on-demand list price, US region, verified 2026-07-25. A dash means the provider doesn’t sell a comparable managed product (documented in our methodology), not a zero.

VM specs: 2 vCPU/8 GB and 16 vCPU/64 GB, dedicated-vCPU general-purpose, Linux, on-demand list, US region, ×730 hrs. Source: SpendArk pricing catalog, verified 2026-07-25.

Want these numbers for your exact stack — your instance sizes, your storage, your traffic, in your region and currency? That’s what the calculator is for.

None of this means AWS, Azure, and GCP are overpriced for everyone. The premium buys real things — the question is whether your workload consumes them. It’s worth paying when you need the breadth of managed services (200+ on AWS versus a handful on Hetzner), global region coverage and edge presence, deep compliance tooling (HIPAA, FedRAMP, SOC 2 across the whole catalog), or a large hiring pool of engineers who already know the platform. Enterprise procurement, committed-spend discounts, and existing credits also change the math.

It is not worth paying when you’re running a handful of straightforward VMs, serving a lot of egress, storing bulk data, or you’re a small team without a FinOps function to chase commitment discounts. That’s a large share of real workloads — and it’s exactly where the 1.7× compute and 8.5× egress gaps turn into thousands of dollars a year for nothing you actually use.

The point of pricing it out isn’t “always pick the cheapest.” It’s to know the size of the premium before you commit, so paying it is a decision rather than a default. Related reading: AWS vs GCP pricing, why your AWS bill is so high, and the 2026 cloud cost benchmark.