Cheaper Alternatives to AWS in 2026: What Each One Cuts

There is no single cheapest alternative to AWS, because AWS bills are not shaped the same way. A bill dominated by EC2 has a different answer from one dominated by egress, RDS or GPU-hours. And several of the providers routinely listed as cheap alternatives are not cheaper than AWS at all once committed-use pricing is applied — they are simpler, which is a different product.

This is a survey of what is genuinely cheaper in 2026, organised by the line item each provider addresses, at prices verified in August 2026.

Start with which line item you are cutting

Before comparing providers, pull the cost breakdown and find what actually dominates. Four patterns cover most bills:

• Compute-dominated. A fleet of always-on instances. This is where the raw-compute providers below apply, and where the largest multiples are available.
• Egress-dominated. Media serving, data exports, cross-region replication. Often the single largest correctable line, and the one where alternatives win by the widest margin.
• Managed-service-dominated. RDS, ElastiCache, MSK, OpenSearch. The hardest to move, because the alternative is not a cheaper managed service, it is operating the thing yourself.
• GPU-dominated. Training or inference. A separate market with a separate set of providers.

A comparison built on the wrong line item produces a migration that saves very little. It is worth ten minutes in Cost Explorer before reading any further.

The 2026 repricing wave

Any comparison written before mid-2026 is stale, because the discount hosting market repriced upward this year, reportedly under DRAM cost pressure:

• OVHcloud repriced on 1 April 2026. The entry VPS-1 moved from roughly €3.50 to €7.60, with mid-tier and bare metal rising more modestly at 10–20%.
• Hetzner repriced on 15 June 2026. The shared-AMD CPX and dedicated-vCPU CCX lines rose by roughly 2.4–2.75×, while the CX and CAX lines rose about 1.3×.

The effect is uneven within each provider, not just between them. At Hetzner specifically, the repricing made the Arm line the value tier and left CPX close to AWS committed pricing. Check the current price of the exact plan rather than the provider’s reputation.

Cheapest on raw compute

The comparison shape below is 8 vCPU with 16 GB or more, Linux, monthly, at list price. AWS figures are us-east-1, its cheapest major region. Euro conversions use approximately $1.17.

Provider	Plan	vCPU / RAM	Storage	Traffic	Monthly
AWS	c7g.2xlarge on demand	8 Graviton3 / 16 GB	EBS extra	metered	~$212
AWS	c7g.2xlarge, 3-year	8 Graviton3 / 16 GB	EBS extra	metered	~$93
Contabo	Cloud VPS 8	8 / 24 GB	300 GB SSD	unmetered, 600 Mbit/s	€16.80 (~$20)
Hetzner	CX43	8 shared x86 / 16 GB	160 GB NVMe	20 TB	€15.99 (~$19)
Hetzner	CAX31	8 Arm / 16 GB	160 GB NVMe	20 TB	€20.99 (~$25)
OVHcloud	VPS-4	8 / 24 GB	200 GB NVMe	unmetered, 3 Gbit/s	$23.37

Hetzner is the reference point for this category: EU and US locations, an official Terraform provider, and 20 TB of included traffic per server. Post-repricing, buy the CX or CAX lines rather than CPX. Covered in detail in Hetzner vs AWS: The Real Cost Difference in 2026.

Contabo posts the lowest headline number and the most RAM per euro. The trade-offs are visible in the specification: SSD rather than NVMe on the core plans, a 600 Mbit/s port at this tier, and a reputation for more variable performance under contention than the others here. Suitable for non-production, batch and internal workloads; test before putting latency-sensitive production traffic on it.

OVHcloud is the largest European provider of the three, with a wider footprint including North America and India, its own bare metal range, and unmetered traffic at a 3 Gbit/s port. Worth checking if data residency in a specific country is a requirement, since the location list is broader than Hetzner’s.

For sustained, predictable workloads all three land between roughly $19 and $25 for a shape that costs $93 on a three-year AWS commitment and $212 on demand. That is a real four-to-eleven-times difference, and it is the largest gap available anywhere in this post.

Cheapest on egress

This is where the alternatives win by the widest margin, and it is frequently the line that decides the whole comparison.

AWS bills outbound transfer at roughly $0.09 per GB for the first 10 TB and about $0.085 per GB beyond, after 100 GB free. Twenty terabytes therefore costs around $1,740 per month.

The same 20 TB is included in a single Hetzner server at no additional charge in EU and US locations, with overage at €1 per TB. Contabo and OVHcloud both advertise unmetered traffic subject to a port speed cap. Any of the three converts a four-figure monthly line into zero.

Two things to keep in mind. AWS egress into CloudFront is free and CloudFront’s own rates are well below EC2’s, so a correctly fronted static or media workload never pays the raw EC2 rate — check what you are actually being billed before assuming the saving. And included-traffic allowances are regional: Hetzner’s Singapore location bills at €7.40 per TB rather than €1.

Cheapest on GPU-hours

Accelerators are a separate market. The general-purpose hosts above are not competitive here — Hetzner’s own GPU line is monthly-billed dedicated servers topping out at a single 96 GB card — and the relevant providers are GPU specialists.

Per H100-hour, at current on-demand rates:

Option	Per H100-hour
AWS p5.48xlarge on demand	~$6.88 (8 × H100, $55.04/hr)
E2E Networks, single H100 node	$3.77
RunPod Secure Cloud	$2.89
AWS p5.48xlarge spot	~$2.60
RunPod Community Cloud	$1.99

RunPod also lists L40S at $0.79–0.99 per hour and A100 80 GB at $1.19–1.59, against AWS g6e.xlarge at $1.861 per hour for a single L40S. Lambda and CoreWeave occupy similar ground with a more enterprise posture and longer commitment terms. E2E Networks is worth knowing about specifically for Indian data residency, since it is MeitY-empanelled.

Three qualifications that decide whether these numbers apply to you:

• Community or spot tiers are interruptible. RunPod’s Community Cloud is capacity from third-party hosts, and AWS spot can be reclaimed. Both are appropriate for checkpointed training and batch inference, and inappropriate for a synchronous serving path.
• Multi-GPU topology is not equivalent. A p5.48xlarge is eight H100s with NVLink between them and EFA between nodes. For distributed training that fabric is the product. For single-GPU inference it is capacity you are paying for and not using.
• Data gravity applies. If training data sits in S3, egress to an external GPU provider is a per-run cost that needs to be in the comparison.

Cheaper is not what every alternative offers

Two providers appear on most “cheaper than AWS” lists without being cheaper than AWS.

DigitalOcean. A CPU-Optimized droplet with 8 dedicated vCPU and 16 GB is $168 per month, against roughly $93 for a three-year committed c7g.2xlarge. Managed PostgreSQL starts at $15.15 per month for 1 vCPU and 1 GB, which is broadly RDS-comparable. What DigitalOcean sells is a smaller product surface, predictable flat pricing and managed Postgres and Kubernetes that a small team can actually operate. That reduces engineering time rather than invoice size.

Scaleway. The POP2 dedicated-vCPU range runs about €0.21 per hour for 8 vCPU and 16 GB, roughly €155 per month. Its arguments are European sovereignty, a free Kubernetes control plane and a genuinely good developer experience — not price.

Both are reasonable choices. Neither belongs in a comparison whose objective is a lower bill on comparable compute.

What all of them leave you to build

The price difference is the managed layer, priced separately. On AWS, RDS bills for the database plus automated backups, point-in-time recovery, failover and patching. On Hetzner, Contabo or OVHcloud, a server bills for the server. The same applies to identity, managed logging and tracing, secrets, autoscaling and WAF.

DigitalOcean and Scaleway sit in between, offering managed Postgres and Kubernetes without the rest of the catalogue.

For the bare-VM providers, the platform work is a real and sizeable scope item: network segmentation, an access plane, a container runtime, CI/CD, observability, secrets management, host hardening, database replication and a tested restore path. The full inventory and the order to build it in is covered in Leaving AWS for Hetzner: What You Have to Rebuild. Cost it alongside the server saving. Below roughly $2,000 a month on AWS it will exceed the first-year saving; above roughly $20,000 it pays back in weeks. For Hetzner specifically we deliver that layer as a standardised package rather than a bespoke build: the Hetzner landing zone.

Choosing

If your bill is dominated by	Look at	Expected direction
Always-on compute	Hetzner, OVHcloud, Contabo	4–11× lower, plus platform work
Egress	Hetzner, OVHcloud, Contabo	A four-figure line goes to zero
GPU-hours, interruptible	RunPod Community, AWS spot	2.5–3.5× lower
GPU-hours, dedicated	RunPod Secure, Lambda, CoreWeave, E2E	1.8–2.4× lower
Managed databases	Staying, or DigitalOcean	Little to no saving; the move is operational
Engineering time	DigitalOcean, Scaleway	Higher invoice, smaller surface

The hybrid is the outcome the numbers most often favour and the one least often put on the table: managed data and inference stay on AWS, while stateless compute, CI and non-production environments move to a cheaper host. That captures the majority of the saving without relocating the parts that carry the most operational risk.

Summary

• There is no single cheapest alternative. Identify which line item dominates the bill first, then pick a provider against that line item.
• On raw compute, Hetzner, OVHcloud and Contabo land between roughly $19 and $25 for a shape that costs $93 on a three-year AWS commitment.
• Egress is usually the widest gap. Twenty terabytes costs around $1,740 per month at AWS list and is included at all three.
• On GPU, RunPod, Lambda, CoreWeave and E2E run 1.8–3.5× below AWS on demand, with interruptible tiers cheaper still and AWS spot competitive with them.
• DigitalOcean and Scaleway are not cheaper than committed AWS pricing. They sell a smaller operational surface.
• Prices moved twice in 2026, at OVHcloud in April and Hetzner in June, unevenly across each provider’s own lineup. Verify the specific plan rather than trusting the provider’s reputation.