AWS Alternatives for Small Teams: Start With the Egress Bill

Compute isn’t why small teams leave AWS. An EC2 instance is expensive but comprehensible.

Egress is why they leave, and the gap isn’t a percentage — it’s two orders of magnitude.

The line item that causes the migration

AWS charges for data leaving its network. The first 100 GB a month is free, then $0.09/GB up to 10 TB, $0.085 to 50 TB, $0.07 to 150 TB and $0.05 above that.

$0.09/GB is $92 per terabyte. Hetzner’s overage rate at its European locations is about €1.00 per terabyte.

A media-heavy app serving 5 TB a month pays $452 to AWS and about $10 to DigitalOcean — on a $24 droplet that includes 4 TB before overage starts.

That’s the migration in one table. Nothing about compute, storage or engineering effort moves the needle like this does.

Why the bill is also unforecastable

The second complaint is subtler and, for small teams, often more painful than the total.

Egress is metered per gigabyte. API Gateway, CloudWatch and S3 are metered per request. A loop that polls something more often than you intended doesn’t produce a slightly higher bill — it produces a bill with a shape you didn’t predict, discovered a month later.

Flat-rate hosting removes an entire category of anxiety here. A €19.99 server costs €19.99 whether your code behaves or not. For a solo developer, the predictability is worth more than the saving, and the saving is already large.

Pick by workload, not by vendor

“What replaces AWS” has no answer. “What replaces the part of AWS I use” has four.

“I run a few containers and a Postgres”

Railway is the closest thing to the developer experience people actually like about modern AWS, without the console. You push, it builds, it runs.

The caveat we found pricing it for n8n: its metered model bills for residency, so anything that must stay awake to catch webhooks is billed as though it’s always on — because it is. For bursty workloads it’s excellent; for permanently-resident ones a flat VPS wins, and we worked that through separately.

DigitalOcean’s App Platform occupies similar ground with more predictable pricing and a slower deploy experience. Its tracked link isn’t working on our end, so we’ve named it without linking.

“I run VMs”

Hetzner is the reason this article’s egress table looks the way it does, and it’s the default answer for anyone comfortable on a terminal. Two caveats to go in with: it repriced its shared CPX line upward by as much as 2.75× in June 2026, and its cheap CX tier has been showing as unavailable — so check both before you plan a migration around it.

DigitalOcean is the middle path: dearer than Hetzner at $24 for 4GB, far cheaper than AWS, with documentation and an API that behave predictably and generous included transfer.

IONOS is worth a look purely on cores — four vCPU on a 4GB plan at $11, which is the best price per vCPU in our VPS index. Its promotional pricing runs only three months before renewing at that rate.

Cloudzy bills month-to-month with no term, which suits a migration you’re not certain about.

“I deploy a Next.js front end”

That’s a different market with the same disease — metered egress. Vercel alternatives prices it out: Vercel Pro includes 1TB then charges $0.15/GB, where a $11.99 VPS includes 4TB.

“I need GPU”

RunPod prices GPUs by the hour with published rates rather than a sales call — an RTX 4090 at $0.34/hour on its community tier, an A100 PCIe at $1.19, an H100 PCIe at $1.99. Against on-demand EC2 GPU instances that’s a large saving, and the hourly billing means idle time costs nothing.

One line to budget that isn’t in the hourly rate: volume storage for model weights bills whether the pod runs or not, and at a higher rate when idle. We covered that in Ollama VPS cost.

“I need managed everything”

Liquid Web is the fully-managed end — you’re buying an operations team rather than a machine. It prices well above everything else here, and that’s the point: if the reason you’re on AWS is that nobody on your team wants to run infrastructure, moving to a cheaper VPS doesn’t solve your problem, it relocates it.

Do not leave AWS if

The honest counter-list, because most “AWS is too expensive” articles don’t have one.

You’re regulated and multi-region. If you need data residency guarantees across regions, audited compliance attestations and the paperwork that goes with them, AWS’s coverage is genuinely hard to replicate. The alternatives here are mostly single-region operations with far smaller compliance surfaces.

Worth checking whether you’re actually bound by a residency rule before paying for one, though. When we read the Canadian requirements, British Columbia had repealed its residency rule in 2021 and Quebec’s Law 25 turned out to require an assessment rather than localisation — so the real constraint is usually a customer contract, not a statute.

You use IAM properly. Fine-grained, cross-account, role-based access with tenancy boundaries is one of the genuinely excellent things AWS built. Nothing in this article comes close, and if you depend on it you’d be rebuilding a security model, not migrating servers.

You actually use the managed services. If you’re running Lambda, DynamoDB, SQS, Kinesis and Step Functions and they’re doing real work, your bill isn’t for compute — it’s for not operating those things. Replacing them means running Postgres, Redis, a queue and a scheduler yourself. That can be the right call, but price the operational load honestly before you start.

Your traffic is small. If you egress under 100 GB a month, the entire premise of this article doesn’t apply to you. Everything above the free tier is where the argument lives.

The pattern: leave for cost if you’re running VMs and moving data. Stay if you’re using AWS as a platform rather than a datacentre.

What migration actually costs

If Hetzner is your destination specifically, we costed that migration end to end — 78% on current prices rather than the 90% older case studies quote, plus the operational work you re-insource.

Two things people underestimate, in opposite directions.

Data transfer out of AWS is charged at the rates above. Moving 5 TB of S3 data to another provider costs roughly $452 in egress, once. That’s a real migration line item and it’s charged by the provider you’re leaving.

The engineering effort is usually smaller than feared for VM-based workloads — a Linux box is a Linux box. It’s much larger if you’ve built on managed services, because you’re not moving code, you’re replacing dependencies.

If you’re partway between, the sensible move is incremental: shift the bandwidth-heavy parts first, since that’s where the entire saving is, and leave anything using managed services where it is.

How we checked this

AWS data transfer pricing — the first 100 GB per month free globally, then $0.09/GB up to 10 TB, $0.085/GB from 10–50 TB, $0.07/GB from 50–150 TB and $0.05/GB above — is from published documentation of AWS’s egress rates in August 2026. The dollar figures in the table are our arithmetic on those tiers.

DigitalOcean’s included transfer allowances (4 TB on the $24 4GB droplet, 5 TB on the $48 8GB) and overage of roughly $0.01/GB, and RunPod’s GPU rates, are those providers’ published figures read in August 2026. IONOS, Cloudzy and Hetzner compute pricing is from our VPS price index, verified in the same period.

On Hetzner’s traffic specifically, one limitation. Its overage rate of about €1.00 per terabyte at European locations is consistent across its published plan data, and that’s the figure we’ve quoted. The included allowance varies substantially by plan family and by region — its data shows markedly smaller allowances at its US and Singapore locations than in Europe, and different figures between its CPX and CCX lines — and we could not confirm the per-plan allowances on the rendered pricing page. So we’ve compared the overage rates, which we’re confident in, rather than claiming a specific included figure. Check the allowance for the exact plan and region you’re buying, particularly outside Europe, where it appears to be a fraction of the European figure and the overage rate is higher.

What we did not do: we haven’t migrated a workload off AWS, measured a real bill before and after, or tested any provider’s delivered network performance. Egress pricing is one line of an invoice — a complete comparison would include storage, requests, support plans and reserved-instance commitments, which vary far too much by account to model usefully.

The provider links above are affiliate links. The section telling you not to leave AWS is the one we’d most want a reader to take seriously.

FAQ

Why is AWS so expensive for small teams?

Usually egress. AWS charges $0.09/GB for data transfer out after the first 100 GB, which is about $92 per terabyte — where flat-rate hosts include several terabytes and charge a few dollars beyond that.

What is the cheapest AWS alternative?

Hetzner for VMs, on both compute and bandwidth, though check plan availability and regional traffic allowances. DigitalOcean is the middle path with better documentation and generous included transfer.

How much cheaper is Hetzner than AWS for bandwidth?

On overage rates, roughly 90×: about €1 per terabyte against AWS’s $92 per terabyte in its first tier. Included allowances differ by plan and region, so check both.

What’s the best AWS alternative for containers?

Railway for the developer experience, with the caveat that its metered billing charges for residency — anything that must stay awake is billed as always-on. DigitalOcean’s App Platform is the more predictable option.

What’s the best AWS alternative for GPUs?

RunPod, with published hourly rates from $0.34 for an RTX 4090 and no sales call. Budget separately for volume storage, which bills whether the pod is running or not.

When should I not leave AWS?

If you’re regulated and multi-region, if you depend on IAM’s fine-grained access model, if you genuinely use the managed services you’re paying for, or if you egress under 100 GB a month.

Does it cost money to migrate off AWS?

Yes — data transfer out is charged at AWS’s normal egress rates, so moving 5 TB costs roughly $452 once. Engineering effort is modest for VM workloads and substantial if you’ve built on managed services.