Scaleway in 2026: an honest review for teams leaving AWS or Azure

Scaleway is the public cloud of the French group iliad, which the Niel family group controls, with regions in Paris, Amsterdam, Warsaw and Milan. Of the European-controlled providers we checked, it has the widest catalogue of AWS-style managed services, and it charges nothing per gigabyte for traffic leaving its instances and load balancers: our production workload costs EUR 582.07 a month there, 54% of the AWS price. The weak points are redundancy and contract cover. The managed database keeps its standby in the same data centre, one router change in March 2025 cut public connectivity to the whole Paris region, and SLA credits are vouchers, capped at 50% for most services.

This article is one of a series comparing 24 European cloud providers with AWS and Azure; the overview puts them side by side. Everything below was checked on 15 and 16 October 2026 against Scaleway's documentation, legal terms, price data and status page, iliad's annual report and user reports, each at its source. We did not open an account or run workloads.

At a glance

What you can build

Status is general availability unless noted, from the lifecycle field of Scaleway's public price catalogue; "Paris only" means the products availability page lists no other region.

The AWS-compatible APIs are partial. Queues "is not currently compatible with Scaleway VPC", holds at most 100 MB per Project and ignores parameters such as DelaySeconds and MessageGroupId. Topics and Events delivers only to HTTP and HTTPS clients, Serverless Functions and Queues in the same Project. Object Storage has versioning and Object Lock but no bucket replication or access logging, and bucket notifications are "in development".

Missing outright: regions outside Europe, a managed backup service, an API gateway, an event bus with rules like EventBridge, a serverless key-value database like DynamoDB and a customer identity service like Cognito. Workflow orchestration is Data Orchestrator, in beta, and Autoscaling Groups have been in public beta since 25 August 2026. Inbound DDoS mitigation for instances is not documented: the load balancer FAQ points to the Edge Services WAF, and the security page claims "native DDoS protection" without describing it.

The tooling is Scaleway's own: a Terraform provider (2.86.0, published 7 October 2026), the scw CLI, an OpenAPI 3.1 schema per API, and a public catalogue API, still an alpha version, that returns all 5,655 SKUs with prices and no login. Some documentation trails the product: the Instances FAQ still quoted the IPv4 price from before June 2026.

What it costs

Scaleway list prices in EUR, VAT excluded, from its catalogue API and pricing pages, checked 15 and 16 October 2026, in Paris (fr-par-1) at 730 hours a month.

Workload A, a starter setup: one VM with 2 vCPU and 4 GB, a 40 GB disk, one IPv4 address, 1 TB of egress and 100 GB of object storage.

AWS charges EUR 112.37 and Azure EUR 116.15 for the same in Frankfurt; on AWS the 1 TB of egress alone is USD 81.00 of a USD 125.95 bill. PLAY2-NANO brings Scaleway's total to EUR 29.16 and the Arm-based BASIC2-A2C-4G to EUR 25.84, but PLAY2 has no SLA.

Workload B, a small production system: three VMs of 4 vCPU and 16 GB with an IPv4 address each, 300 GB of SSD block storage, a load balancer, PostgreSQL with 2 vCPU, 8 GB and 100 GB as primary plus standby, 1 TB of object storage with 1 million PUT and 10 million GET requests, a Kubernetes control plane and 2 TB of egress.

With three dedicated-vCPU STANDARD3-X4C-16G instances the total is EUR 676.78. AWS charges EUR 1,082.61 for workload B on m7i.xlarge and Azure EUR 1,091.88 on D4s_v5, so Scaleway costs 54% of AWS and 53% of Azure on shared vCPUs, and 63% of AWS on dedicated ones. Two lines make most of the gap: AWS bills the 2 TB of egress at USD 171.00 and its HA database at USD 320.13 with storage, where Scaleway charges nothing for the egress and EUR 170.09 for the database. One line is dearer at Scaleway: the dedicated Kubernetes control plane, EUR 80.30 against EUR 65.13 for EKS. The free shared control plane would make it zero, with one API server and no SLA.

Two assumptions sit in the total: the database storage is priced once (if the standby's volume is billed again, add EUR 9.93), and the load balancer's first IP is taken as included, since the price list names only an "Additional IP address". The dedicated control plane carries a 30-day commitment. Worker nodes are not included, as everywhere in this series.

Egress and IPv4. For instances, "List prices include egress and IPv6 addresses"; bandwidth is capped per type instead (350 Mb/s for BASIC3-X2C-4G, 1 Gb/s for STANDARD3-X4C-16G). Load balancer traffic is "Unlimited", and we found no fair-use clause in the Instance Specific Conditions. Object Storage egress costs EUR 10 per TB beyond the free 75 GB, against EUR 80.30 per TB of egress on AWS and EUR 77.62 on Azure beyond their free 100 GB; requests are free. A zonal IPv4 address costs EUR 0.005 an hour (EUR 3.65 a month), the same number AWS and Azure charge in dollars.

Billing. CPU instances are billed by the hour, GPU instances by the minute, and IP addresses and volumes while allocated, powered on or not, with no monthly cap stated. Savings Plans promise "up to 25%" off compute. There is no general free tier, and we could not confirm a hard spending cap. The General Terms allow a deposit of the last 60 days' average consumption after a payment incident, or when paying by anything other than direct debit or card.

GPUs, per GPU-hour: H100 PCIe EUR 2.87, H100 SXM EUR 3.17 (8-GPU size), B300 SXM EUR 7.50, L40S EUR 1.47 and L4 EUR 0.79. There is no H200 or B200. AWS does not sell the H100 in Frankfurt and charges EUR 6.57 in Stockholm; its L40S is EUR 2.08 and its L4 EUR 0.90 in Frankfurt. Every H100 and B300 size needs a support request, and the contract lets Scaleway terminate an on-demand H100 PCIe instance "with a minimum notice period of ten (10) minutes", deleting "all Content stored on them". That makes it closer to a preemptible machine than a reserved one.

Price changes. On 27 April 2026 Scaleway announced increases for 1 June 2026, 35 days' notice where its terms require one month. It said prices had been flat for three years and cited inflation, storage and memory hardware costs and IPv4 scarcity, plus a "strategic repositioning" for load balancers and gateways. Selected lines, EUR per hour:

We found no other general price change in the last 24 months.

Reliability

What a zone is. Scaleway's engineering post of April 2021 defines a zone as one or more data centres in an area of about 5 km, "situated at least 50km from another availability zone in the same region"; in Paris, fr-par-1 was DC2 and DC3, fr-par-2 was DC5 and fr-par-3 was DC4. We could not confirm the 50 km figure for today's zones. The same post sets the limit: "The physical fault domain is the availability zone and the network failure area is the region." The zones are not equal either: fr-par-3 had 117 live SKUs against 400 in fr-par-1.

The database standby. Managed PostgreSQL and MySQL replicate synchronously to a hot standby, and the FAQ says the two "are indeed located in the same data center but in two separate racks". The pair survives a failed rack, not the loss of the building. Only read replicas can go to another zone (EUR 0.0608 an hour at the size we priced), while AWS and Azure put the standby itself in another zone. The Kapsule control plane keeps three etcd replicas across zones, and Multi-AZ Object Storage spans them; instances and block storage are zonal.

The SLAs:

• Credits are vouchers that "cannot under any circumstances be reimbursed in cash". The credit is "an all-inclusive payment for all damages", and the client "waives its right to any other claim".
• You claim "no later than thirty (30) days following the month" of the breach, with "the log files indicating the Unavailability Periods": your monitoring makes the case.
• An instance counts as down only after 4 continuous minutes without external connectivity; unavailability "due to a stock shortage" is excluded.
• The Kubernetes SLA counts an outage only when all API servers are lost "without the possibility to launch a replacement cluster in any region". Read as written, a regional outage does not count while another region could host a cluster.
• Outside the SLAs, the General Terms promise "a general best efforts obligation", limit liability to direct damages and bar claims after one year.

AWS commits to 99.99% for EC2 across two or more zones and 99.5% for a single instance, with credits up to 100% and claims until the end of the second billing cycle; Azure gives 99.99% across zones and 60 days to claim. Scaleway's best instance commitment, 99.5%, is AWS's single-instance level.

Status page and incidents. status.scaleway.com has history back to late 2019. From 1 August 2024 to 15 October 2026 it lists 15 critical and 129 major entries across every product, Dedibox and web hosting included. The two largest outages have detailed public postmortems.

Two of these bear on design: in May 2026 a failure in one Paris zone spilled into the others, and in March 2025 one change took the whole region's public network with it. Both limit what a three-zone design buys you. Milan, the newest region, had the densest run of major incidents in 2026.

Support

The targets are for a first response; no resolution time is committed, and Scaleway "may confirm or refute" the priority the client sets, a decision the client cannot contest. Support works in English and French, and Scaleway says its customer care team is in its Paris office. Even the free plan has a target, which AWS Basic (no technical cases) and Azure Basic (no technical response targets) do not. A one-hour target for a production outage costs the greater of USD 29 or 9% of spend on AWS Business Support+, or USD 100 a month on Azure Standard.

Getting an account

A valid payment method comes first, and Scaleway validates each subscription. The identity check that unlocks more is a government photo ID plus a live face check in the browser within 15 minutes; company IDs, credit cards and driving licences are not accepted.

With a payment method alone you get one POP2-2C-8G, one PRO2-XXS and five PLAY2-NANO instances, 500 GB of block storage, 5 database instances and one dedicated Kubernetes control plane. Identity validation unlocks POP2-4C-16G, PRO2-XS and larger, one L4 or L40S GPU, 5,000 GB of block storage and the SMTP ports 25, 465 and 587. Every H100 and B300 size, and every multi-GPU L4 or L40S size, needs a support request on top.

Workload B on day one: the storage, the database and one dedicated control plane fit the payment-only quotas. The three 4 vCPU, 16 GB instances probably need the identity check, because their older equivalents do and the quota table did not yet list the BASIC3 and STANDARD3 families. No support request is named for any of it, so our reading is yes after the identity check, though the documents do not settle it. On AWS the default of 5 vCPUs for standard instances is below workload B's 12; Azure's 20 cores per region fit on paper.

Two clauses in the General Terms matter later. Above "a certain volume of Service consumption", Scaleway may ask for the legal representative's ID and a company register extract under three months old. And it may suspend a service "immediately and without prior notice" for reasons including non-payment, fraud, "legitimate suspicion of fraud" or a security risk, with billing continuing during the suspension.

Compliance and data protection

The HDS list covers instances (CPU and GPU), Block and Object Storage, Dedibox, Elastic Metal, VPC, Kapsule and Load Balancer, but not the managed databases or serverless products. It means hosting in France, a signed HDS contract through sales and a Business or Enterprise support plan.

For SecNumCloud, ANSSI accepted Scaleway's application without reservation at the end of December 2024, according to Next.ink, and Scaleway aimed for qualification by the end of 2025. That date has passed with no later milestone found and no new date given. The announced scope is a single-zone configuration in Paris-region data centres of the iliad group, covering instances, storage, networking, IAM, Key Manager and Elastic Metal, with Kapsule later.

On 17 April 2026 the European Commission named Scaleway one of four winners of its sovereign cloud tender, worth up to EUR 180 million over six years. Scaleway was rated SEAL-3 ("EU actors exercising meaningful but not full influence"), as were STACKIT and a Post Telecom, OVHcloud and Clever Cloud partnership.

The Data Processing Agreement of 1 June 2024 says services "are located within the European Union by default", the client chooses the region, and Scaleway will not store or transmit personal data outside the EU "without having expressly informed the Client in advance". The sub-processor list sits in the Trust Center and did not render for us.

Who controls it

iliad SA owned 99.56% of Scaleway SAS at the end of 2025; the rest is employee and executive share grants. iliad SA is 98.88% held by iliad Holding SAS, which iliad's annual report describes as over 95% controlled by Maya, "wholly owned by the Niel Family Group". iliad was delisted on 14 October 2021. The report shows no non-European parent or significant shareholder.

You contract with Scaleway SAS under French law and the Paris commercial court, or, with an Italian billing address, with Scaleway Italia S.R.L. under Italian law.

The data centres are run by OpCore, the group's data centre company, of which iliad sold 50% to InfraVia on 31 March 2025; the May 2026 postmortem names "the OpCore PAR5 datacenter" behind fr-par-2. The API, instance and Kubernetes products are Scaleway's own.

Two facts readers ask about. iliad's 2024 list of consolidated companies includes Scaleway US Corporation in Chicago, marked "N.D." for 2025, so a US subsidiary existed; its current status is not stated. And Scaleway's pages say it is "Free from the US CLOUD Act", which is Scaleway's claim, not our finding. Scaleway's own revenue and headcount are not published.

Leaving

Instance and load balancer egress has no per-GB charge. Exporting 10 TB from Object Storage costs about EUR 99.25, or close to nothing pulled through an instance in the same region, since traffic inside a region is free. Glacier objects must first be restored, at EUR 0.009 per GB. Scaleway documents the AWS CLI and rclone for buckets and pg_dump and mysqldump for databases; we did not establish whether logical replication to an outside target is allowed.

Lock-in sits in the integrations. Kapsule is CNCF-conformant, but load balancers and volumes come through Scaleway's cloud controller manager annotations and CSI driver. IAM scopes and permission sets, serverless triggers and Cockpit's alerting, which uses Scaleway's own alert manager, have to be rebuilt elsewhere. The S3, partial SQS and SNS, and OpenAI-compatible APIs ease a move in either direction.

When a contract ends, content "shall be automatically and permanently deleted" and Scaleway "undertakes not to retain any copy". The 2026 terms promise at least three months' notice before a service's end of sale; beta services can be "definitively discontinued by Scaleway without notice". The Public Gateways API v1, for example, was removed on 1 October 2025, about six months after the changelog announced it.

What users report

We read Hacker News comments mentioning Scaleway since October 2024; Reddit gave no usable threads, and review sites do not count as evidence here. Comments are cited by their Hacker News item number.

The closest match to AWS, with fewer services. Users report moving production from AWS to Scaleway's managed Kubernetes, Postgres, Redis, serverless containers and queues, while saying it has "much fewer managed services" (46836273, January 2026) and "still lags behind the big three" (48340477, May 2026). Four more report the same, among them 46837275; together the reports run from March 2025 to May 2026.

Support: mixed. Users report "excellent support" (43361754) and quick answers to Terraform provider bugs (46678419); others report no reply to an email during signup (46060566) and "site and support is horrible" (46061189).

Reliability. Users report "a couple big outages every year" (43153568, February 2025) and reliability that "was terrible last time we used them" (48817760, July 2026); one counters with "simplicity and reliability" (47085681). The status record above confirms the pattern.

Database HA in one data centre. A five-year customer calls the feature "cosmetic" because the HA instances "are in the same DC" (46837788, January 2026), which Scaleway's FAQ confirms.

Identity checks depend on what you order. One user was asked for ID (48544740, June 2026), two others were not (42398985; 48125822). The quota table explains why.

Payment faults. One user reports a failed first payment followed by dunning emails over an invoice under EUR 10 (47716333, April 2026). The status page shows major incidents with SEPA payments (June 2025), adding a card (September 2025) and resources locked after an invoice was paid (December 2025).

For Kapsule we found no two first-hand reports on reliability. The status page lists 7 major and 19 minor Kubernetes entries from October 2024 to October 2026.

Who it suits, and who should look elsewhere

Scaleway suits teams that want AWS-style building blocks (Kubernetes, PostgreSQL, serverless containers, queues, S3) under a French or Italian contract, services that send a lot of traffic out, startups on its credit programmes, open-model AI inference in Paris, and French health-data projects that fit the HDS product list.

Look elsewhere if you need database failover across zones, or several regions in one provider under strong SLAs; if you need anything outside Europe; if you are a public buyer who needs SecNumCloud today; if you depend on managed backup, an API gateway or an event bus; or if you need H100-class GPUs on demand without talking to sales.

Sources

Checked on 15 and 16 October 2026.

Pricing

• Public product catalogue API (5,655 SKUs)
• Instances pricing, network pricing, storage pricing, Managed Databases pricing, containers pricing
• Support plans
• "A transparent update on Scaleway pricing", 27 April 2026
• AWS and Azure figures: AWS Price List bulk API and Azure Retail Prices API, converted at the ECB rate of 15 October 2026

Terms and SLA

• General Terms of Services, v07/04/2026
• Instance Specific Conditions, v07/04/2026
• Specific Conditions, Support Services, v07/04/2026
• Data Processing Agreement, 1 June 2024
• SLA index, with the SLAs for Instances, Block Storage, Object Storage, Managed Databases, Kubernetes Dedicated Control Plane and Load Balancer

Documentation

• Products availability by region
• "Design considerations and our recommendations for data protection", 20 April 2021
• Managed Databases for PostgreSQL and MySQL FAQ
• Kubernetes control plane offers
• Instances FAQ
• Organization quotas and Verify your identity
• Documentation source and changelog
• Terraform provider and CNCF Kubernetes conformance results

Status and incidents

• Status page and history
• Postmortem, Paris region, 10 March 2025
• Postmortem, fr-par-2, 22 May 2026

Compliance

• Security and compliance and Trust Center
• HDS
• SecNumCloud and Scaleway's post on its SecNumCloud scope
• European Commission, sovereign cloud tender, 17 April 2026
• Next.ink, "SecNumCloud : Scaleway passe J0 sans aucune réserve et vise une qualification fin 2025", 13 January 2025

Company and ownership

• Legal notice
• iliad Group, 2025 Annual Financial Report
• Milan region launch, 19 March 2026

User reports

• Hacker News comments, by item number at news.ycombinator.com: 42398985 (12 December 2024), 43153568 (23 February 2025), 43361754 (14 March 2025), 43459311 (24 March 2025), 45977397 (19 November 2025), 46060566 and 46061189 (26 November 2025), 46678419 (19 January 2026), 46836273, 46836757, 46837275 and 46837788 (31 January 2026), 47085681 (20 February 2026), 47716333 (10 April 2026), 48125822 (13 May 2026), 48340477 (30 May 2026), 48544740 (15 June 2026), 48817760 (7 July 2026)