Managed Kubernetes for teams who already run Kubernetes.

What is and isn't in these numbers

• Two configurations, and which one the totals price: Every total in tables 1 and 2 prices the 4 vCPU / 16 GB configuration: one worker of 4 vCPU / 16 GB Linux, on-demand, no commitment; 100 GiB of block storage at the compared worker size, or the smallest volume the provider requires to reach the performance target where its product scales performance with size (OVHcloud Gen2: 489 GiB) — the performance target (≥10,000 sustained IOPS and ≥250 MiB/s) is otherwise priced separately, per provider; one named L4 load balancer with a public IPv4; and the named paid control-plane tier where one exists — on-demand and with no commitment, except where noted. That is a price comparison at a fixed shape, and it is NOT a claim that the performance target is met — on AWS, Azure, Google and both Scaleway rows the compared worker does not reach it, and Aether publishes no figure that could show it either way. The configuration meeting the performance target — ≥10,000 sustained IOPS and ≥250 MiB/s, or the smallest configuration reaching them — is stated per provider under "The configuration meeting the performance target": a different, dearer configuration on AWS, Azure, Google and Scaleway POP2 (never folded into a total); the row's own total where the provider already reaches the target at the compared size (OVHcloud, DigitalOcean); not priced at all where no such configuration can be shown (Aether publishes no figure; Scaleway shared vCPU defers to the POP2 row). Regions: Aether eu-fi, AWS eu-west-1, Azure West Europe, Google europe-west1, Scaleway fr-par-1/2, OVHcloud GRA on the French catalogue, and DigitalOcean's region-independent list (AMS3).
• Boot, root and OS disks: Every row states its own treatment, and the four that exclude a separately billed disk say so rather than leaving it unsaid. Aether: the worker's own 50 GB node disk is a separate rate-card line and IS inside its total. Bundled, so nothing is missing: DigitalOcean (50 GiB of Droplet storage), OVHcloud b3-16 (100 GB of local NVMe), Scaleway GP1-XS (up to 150 GB of local SSD). EXCLUDED from the total, because the provider bills the disk separately and no line for it was read from an official source: AWS (m6i.xlarge is EBS-only — its EBS root volume is not priced here), Google (n4-standard-4 has no bundled instance storage — its boot disk is not priced here), Azure (D4s v5 has no local temporary disk — its managed OS disk is not priced here), and Scaleway POP2-4C-16G (no local disk at all — the block volume a cluster needs to boot from is not priced here). Read the compute columns as compute, not as like-for-like storage.
• What the totals exclude: VAT, egress and every other transfer charge, load-balancer processing and capacity fees, NAT and gateway products, support plans, backups, and any committed-use or reserved discount. Transfer is in its own table precisely because leaving it inside a total would hide the largest variable in a real bill.
• On-demand, except where noted: Every price here is the provider's on-demand, no-commitment rate, with one exception, and it is on a control plane rather than on any worker, volume or load balancer. It is marked in the "Control-plane tier" cell of both Scaleway rows in table 1, which is the only control-plane column this page has — so read it as applying wherever that rate is counted, INCLUDING inside the table 2 totals and the target configuration priced beside them, neither of which breaks the control plane out. Scaleway's Kapsule Dedicated-4 control plane — the tier BOTH Scaleway rows use — carries a 30-calendar-day commitment: "When selecting a dedicated environment, you are committing to this plan for 30 calendar days." A downgrade is blocked for its duration. The alternative that carries no commitment is the free Kapsule Mutualized tier, which lists its SLA as N/A — which is why the paid tier is the one priced here. Every other selected tier — the control plane, worker, block volume and load balancer of Aether, OVHcloud, DigitalOcean, AWS, Azure and Google, and both Scaleway workers — was re-examined for a term in the same source captures the prices came from, and none of those captures records a commitment, a minimum term or a required reservation. Where a monthly figure is used instead of hourly × 730, it is a postpaid or consumption rate, not a committed one: OVHcloud's load balancer, floating IP and volume prices come from its "month.consumption" and "monthly.postpaid" catalogue entries, and DigitalOcean's 672-hour cap is a cap, not a term. Committed-use, reserved and savings-plan discounts are excluded from every total, as "What the totals exclude" says.
• Hours and currency: Hourly rates are multiplied by 730 unless the provider publishes a monthly price or cap, in which case the published figure is used: DigitalOcean caps bundled worker nodes at 672 hours, and OVHcloud's published monthly prices for its load balancer, floating IP and volumes are below hourly × 730. Prices are shown in the currency the provider publishes; USD rows are converted at 1 EUR = 1.1614 USD, the ECB euro reference rate of 2026-09-08.
• Pending cells: Every figure was read from an official source on 2026-09-09, from the provider's own pricing page, order catalogue or price-list API — never an aggregator, and never a summarised fetch. Where a figure could not be read, the cell says "Pending" and no total is completed from it: what remains pending is the published SLA percentage for AWS, Azure and Google, Scaleway's per-GB instance transfer rates, DigitalOcean's cross-region rate, DigitalOcean's NAT or gateway rate, and Azure's NAT Gateway meter. One figure is DERIVED rather than read and is labelled so where it appears: n4-standard-8's europe-west1 price, taken as 2 × n4-standard-4 because that is the exact ratio Google's published table shows. Google's pricing pages are per SOURCE REGION and switch client-side, so every Google figure here — instance, disk and the transfer tiers — was read with the region selector on Belgium (europe-west1); the page's default Iowa tab publishes different transfer tiers, and reading it by mistake is how a US row gets attributed to Europe. No cell is filled with an estimate or with a figure carried over from an earlier version of this table.
• The performance target is priced, never assumed: Where the 4 vCPU / 16 GB configuration does not reach the performance target, the row says so and prices the configuration meeting the performance target beside it. OVHcloud needs 489 GiB rather than 100 GiB because Gen2 performance is size-derived (+€33.45), and that larger volume is already inside its total; its Gen2 performance scales per decimal GB while its price is per GiB, so the sizing uses 489 GiB (525 GB) — taking the smaller figure is the reading that favours OVHcloud. AWS, Azure and Google all reach the target on the volume but not on the instance: the smallest compliant instance adds 51% to AWS, 53% to Azure and 50% to Google. Scaleway reaches it only on POP2-32C-128G, €975.06/month. DigitalOcean reaches it at the compared size, but as a ceiling rather than a provisioned figure. Those target-meeting figures are stated per row. Where they name a different configuration from the compared one they are deliberately NOT in the total column; where a provider already reaches the target at the compared size (OVHcloud, DigitalOcean) the row says so explicitly and the figure IS that row's total. The total column always prices the 4 vCPU / 16 GB configuration, nothing more.
• Two things Aether does not claim: Aether publishes no IOPS or throughput figure for aether-block, so its storage cell states that rather than claiming the target is met. And an Aether load balancer is a dedicated proxy VM per Service — a single instance, not a managed multi-instance service like AWS's, Azure's, Google's or DigitalOcean's. Aether also publishes no SLA at all; of the six others, three publish a percentage we could read (OVHcloud, Scaleway, DigitalOcean) and the rest are pending. On Aether: every cluster currently requires a separately billed public API endpoint, so the €4 public API endpoint is included in Aether's total; no other provider is charged a control-plane IPv4 here. Private-network connectivity, which would make the public endpoint optional, is a future capability.
• Control planes are not free once you want an SLA: None of the seven offers a control plane that is both free and covered by an SLA. DigitalOcean and Scaleway and OVHcloud each publish a free tier, and each free tier drops the SLA (Scaleway lists it as N/A; OVHcloud offers an SLO, not an SLA; DigitalOcean’s 99.95% applies only with HA enabled at $40/month). AWS charges a flat $0.10/h with no free tier. Google charges a flat $0.10/h across every mode — its free-tier credit is explicitly excluded for the regional clusters an HA comparison uses. Aether charges no control-plane fee across any available tier; these competitors have paid control-plane tiers. Aether publishes no SLA either.
• Why Scaleway has two rows: Scaleway sells two 4 vCPU / 16 GB tiers and a buyer meets both on the same pricing page, so both are shown. They are not interchangeable. GP1-XS is €0.09282/h (€67.76/month) and Scaleway's own catalogue flags it sharedCpu: true with an SLA of 0, on AMD EPYC 7281 (or equivalent) at 2.1 GHz with 500 Mb/s; POP2-4C-16G is €0.147/h (€107.31/month) with dedicated cores on EPYC 7543 at 2.8 GHz with 800 Mb/s. The cheaper row is the one to read carefully: shared cores, no instance SLA, and 300 MiB/s of block bandwidth against POP2's 800, which puts it further from the storage target — though it does bundle up to 150 GB of local SSD that POP2 lacks. With the same Kapsule Dedicated-4 control plane it reaches €148.06/month, against Aether's €92.00 for worker plus endpoint. Both rows carry the same control plane, storage and load-balancer rates; the difference is the instance.

Sources, all read 2026-09-09: Aether rate card, OVHcloud order catalogue (FR), OVHcloud block storage, Scaleway instance pricing, Scaleway storage pricing, Scaleway network pricing, DigitalOcean Droplet pricing, DigitalOcean volumes, DigitalOcean load balancers, AWS Price List Bulk API (eu-west-1), Azure Retail Prices API (westeurope), GKE pricing, Compute Engine general-purpose pricing, Google Cloud disk pricing, Hyperdisk per-VM performance limits, Google Cloud VPC network pricing. Check current rates before deciding on price alone.

Full example bills, with storage, load balancers and registry →