• Tallinn06:03
  • Ljubljana05:03
  • Novi Sad05:03

By Kristijan Sekereš

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

Server racks in a data centre

Gcore is a Luxembourg company that began with a content delivery network and now sells virtual machines, bare metal, managed Kubernetes and GPUs behind its edge network. Traffic leaving VMs and GPU clusters is free and unlimited, the highly available Kubernetes control plane costs nothing, and our production workload comes to EUR 384.60 a month in Luxembourg-2, about a third of the AWS price. That total includes a managed PostgreSQL that is a free beta, and the gaps are real: no availability zones, no queues, key management or fine-grained IAM, compute SLA credits that stop at 20% and count downtime from your own emailed report, and certificates we could not see. Gcore does not publish its shareholders; Northern Data, now about 85.2% owned by the US-listed Rumble, holds two seats on its board and in 2025 secured an option on a majority, whose status is not disclosed.

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 Gcore's documentation, legal terms, price data and status page, filings by Northern Data and Rumble, press reports and user reports, each at its source. We did not open an account or run workloads. AWS and Azure are priced in Frankfurt, but Gcore's Frankfurt-2 region sells neither managed Kubernetes nor object storage, so we priced Luxembourg-2, the only EU region that sells VMs, Kubernetes, S3 Standard object storage and GPUs together.

At a glance

Gcore
Headquarters and controlContern, Luxembourg: G-Core Labs S.A., under G-Core Holding S.A. Shareholders are not published; press names Wargaming's founder. Northern Data AG, about 85.2% owned by US-listed Rumble since June 2026, took two board seats and an option on a majority in 2025 and holds 2.13%; after amended arrangements in July 2026 it says it no longer has significant influence. The option's status is not disclosed. Not listed
Regions and zones12 cloud regions in the EU (three in Luxembourg, Frankfurt-2, Amsterdam, three in Paris, Warsaw, Helsinki, two in Sines, Portugal), 33 in all. No availability zones
Workload A (starter)EUR 40.70 a month in Luxembourg-2
Workload B (production)EUR 384.60 in Luxembourg-2, on dedicated vCPUs. The managed PostgreSQL in it is a free beta (versions 13 to 15, 100 GB limit), priced at zero and not a production service
EgressFree and unlimited from VMs and GPU clusters. Object storage EUR 0.02 per GiB, bare metal EUR 0.0013 per GB
Compute SLA99.95% for VMs in "Core" regions, 99.9% in "Edge" regions, none for shared Basic VMs. Credits of 2.5% to 20%, counted from your emailed defect report
Included supportBasic: chatbot, email and tickets, English only, no response time
Who can sign upAnyone with an email address, a payment method and EUR 4 of credit; the signup docs describe no ID check, but the MSA makes every contract subject to KYC. Default quotas are not published; bare metal always needs an approved quota request

What you can build

Gcore labels almost nothing as beta, so status is general availability (sold on the price page without a beta label) unless noted.

What you needGcoreAWSAzure
Virtual machinesIntel, AMD EPYC and Ampere Arm flavours, "Dedicated vCPUs" per the product page; shared Basic VMs only in Amsterdam within the EUEC2Virtual Machines
Bare metalHourly Xeon, Ryzen and EPYC servers, after a quota requestEC2 metalBareMetal Infrastructure
GPUs8-GPU nodes with H100, H200, A100 or L40S; Virtual GPU with 1 to 8 H100s; spot; B300 and GB300 NVL72 through salesEC2 P and GND and NC series
Block and file storagethree volume types with triple replication; NFS file sharesEBS, EFSManaged Disks, Azure Files
Object storageS3 Standard, in the EU only in Luxembourg-2; S3 Fast in SinesS3Blob Storage
Relational databasesPostgreSQL 13 to 15, beta: free, 100 GB limit, three-node HARDSAzure Database for PostgreSQL
Other data servicesnoneElastiCache, DocumentDB, MSK, OpenSearch ServiceAzure Cache for Redis, Cosmos DB, Event Hubs
KubernetesVM, bare-metal or GPU node pools, CNCF conformant for 1.27 to 1.36, free HA control planeEKSAKS
ServerlessFunctions, Container as a Service; FastEdge (WebAssembly on the CDN edge)Lambda, App Runner, Lambda@EdgeFunctions, Container Apps
MessagingnoneSQS, SNSService Bus, Event Grid
EdgeCDN, Managed DNS, DDoS protection (basic free), WAAPCloudFront, Route 53, Shield, AWS WAFFront Door, Azure DNS, DDoS Protection, Azure WAF
Securityfixed roles; Secrets Manager, documented only for load balancer certificates; no KMSIAM, Secrets Manager, KMSEntra ID RBAC, Key Vault
AIEverywhere Inference on L40S, A100 or H100, OpenAI-compatible endpointsSageMaker endpoints, BedrockAzure ML endpoints, AI Foundry
Backupsnapshots; no backup serviceAWS BackupAzure Backup

Missing outright: managed MySQL, Redis or Valkey, MongoDB, Kafka and a general OpenSearch service; queues; a key management service; IAM policies or resource-level permissions; a backup service; and availability zones. Object storage documents Object Lock and lifecycle expiration but not replication or versioning, and Gcore publishes no list of unsupported S3 features. The function runtimes are all past or near their upstream end of life.

One REST API covers every product, and the Python and Go SDKs (0.59.0, 12 October 2026) are generated from its OpenAPI specification. Gcore's Terraform provider has two lines: v0 at 0.37.1 (12 October 2026), and v2, a rewrite still at release candidate, which needs a manual re-import and has no resource yet for snapshots, S3 storage, functions or PostgreSQL. The repository had 6 open issues on 16 October 2026, the oldest from September 2023. There is an official MCP server too.

What it costs

Gcore list prices in EUR, VAT excluded, from the price API behind gcore.com's calculator and from the price pages, checked 15 and 16 October 2026, in Luxembourg-2 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.

LineChoiceEUR a month
VMg2-standard-2-4, Intel, 2 vCPU, 4 GB30.44
Disk40 GB Standard volume4.00
Public IPv4external IP2.53
1 TB egressfree0.00
100 GB object storageS3 Standard, 93.13 GiB at EUR 0.043.73
Total40.70

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. Gcore costs 36% of the AWS price, on a dedicated vCPU against burstable ones, since Luxembourg-2 sells no shared-vCPU plan. With Frankfurt-2's AMD g3a-standard-2-4 (EUR 20.76) and the object storage bought in Luxembourg-2, the total is EUR 31.02.

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.

LineChoiceEUR a month
3 VMsg2-standard-4-16, Intel, dedicated vCPU, 3 x 85.19255.57
3 public IPv43 x 2.53317.60
300 GB block storageHigh IOPS SSD at EUR 0.22 per GB, boot volumes included66.00
Load balancerlb1-1-2, active-standby pair15.18
PostgreSQL HAthree nodes, synchronous replication; beta, free0.00
1 TB object storageS3 Standard, 931.32 GiB at EUR 0.0437.25
1 million PUTEUR 0.03 per 10,0003.00
10 million GETnot charged, per the price page0.00
Kubernetes control planeHA, free0.00
2 TB egressfree0.00
Total384.60

AWS charges EUR 1,082.61 for workload B on m7i.xlarge and Azure EUR 1,091.88 on D4s_v5, so Gcore comes to 36% of AWS and 35% of Azure. That flatters Gcore, because its database is a beta at zero: take AWS's highly available database out (USD 320.13 with storage) and AWS still costs EUR 796.98, of which Gcore is 48%. Most of the remaining gap is two lines Gcore does not charge: 2 TB of egress (USD 171.00 on AWS) and the Kubernetes control plane (USD 73.00 for EKS).

Three assumptions sit in the total. The price page charges only write requests to object storage, while the docs say "Requests"; if GET requests are billed, add EUR 30.00. Gcore does not say whether the load balancer's automatic public address is billed as an IP (EUR 2.53 if so). Together they give an upper bound of EUR 417.13. And the database will cost something once the beta ends, at a price nobody can budget yet. Worker nodes are not included, as everywhere in this series.

Egress and IPv4. "Network traffic for all VMs is free and unlimited", and GPU cluster egress is free too. Bare metal pays EUR 0.0013 per GB. Object storage egress is EUR 0.02 per GiB on the price page, about EUR 18.63 per TB, against EUR 80.30 on AWS and EUR 77.62 on Azure beyond their free 100 GB; the docs say "per Gb", which would be eight times dearer if it meant gigabits. A public IPv4 address costs EUR 2.53 a month, below the USD 3.65 AWS and Azure charge.

Billing. Cloud resources are "charged per minute for the time they are in use", from a prepaid balance of at least EUR 4, topped up from the saved card. VMs bill until "completely stopped", bare metal and GPU servers until deleted, even when powered off. There is no free tier for cloud, and no spending limit is documented beyond the prepaid balance.

GPUs, EUR per GPU-hour on 8-GPU bare-metal nodes (Gcore prints the node price divided by eight):

GPUOn demandSpotReserved 12 / 36 monthsRegions
H100 80 GB2.731.641.85 / 1.75Luxembourg-2 and -3, Helsinki, Sines-2 and -3, London-2, Incheon
H200 141 GB3.182.272.09 / 2.00Chester, UK, only
A100 80 GB1.36none1.05 / 0.91Luxembourg-2 and -3
L40S 48 GB1.27none0.91 / 0.82Luxembourg-2 and -3, Manassas, Singapore

These are self-service, subject to stock and GPU quota. The B300 (EUR 6.00) and the GB300 NVL72 rack go through sales; there is no B200 or L4. A single H100 comes as a Virtual GPU at EUR 2.82 an hour. AWS does not sell the H100 in Frankfurt and charges EUR 6.57 per GPU-hour in Stockholm; Azure's H200 costs EUR 12.29 in Germany West Central. Spot nodes "can be reclaimed with 24 hours' notice", and the contract lets Gcore deliver part of a committed GPU order or substitute hardware of an "equivalent or superior tier".

Price changes. An undated banner announces "New lower prices" in eight locations, Frankfurt-2 among them, where AMD plans cost about half of Luxembourg-2's (EUR 54.77 against EUR 108.00). We found no increase in the last 24 months. The Master Services Agreement of 17 June 2026 lets Gcore "increase the Fees ... annually" by at most the higher of 5% or Luxembourg CPI, and adjust them "at any time" for demonstrable cost increases on 30 days' notice.

Reliability

No zones. Gcore has no availability zones. Inside a region you get placement groups that put VMs on separate physical servers, load balancers that run as an active-standby pair, and volumes on a distributed file system "with triple data replication". Where the load balancer pair or the three PostgreSQL HA nodes sit (rack, room or building) is not stated, nor how long a database failover takes; AWS and Azure put a synchronous standby in another zone and fail over in 60 to 120 seconds. Surviving the loss of a site means a second region, and the choice is narrow: Luxembourg (ED-9) sells managed Kubernetes, Luxembourg-3 and Frankfurt-2 do not, and none of the three sells S3 Standard.

The SLA, from the legal page of 17 June 2026:

ServiceMonthly targetCredit cap
VMs and load balancers, Core regions99.95%20%
VMs and load balancers, Edge regions99.9%20%
GPU Cloud, bare metal, Managed Kubernetes, Functions99.9%20%
Object Storage99.9%25%
Basic VMnone
Managed PostgreSQLno level given
  • Credits run from 2.5% of the affected service's monthly fee (below 99.95%, Core regions only) to 20% below 90%. They go against the next invoice, never cash, and none is issued under EUR 25 a month.
  • The clock starts "from Gcore receiving a valid Defect Report from Customer via email and Gcore's subsequent confirmation of the Defect": an outage nobody reports by email is not downtime. For Object Storage it starts when Gcore opens its own incident ticket. Claims go by email within 30 days.
  • Kubernetes counts as down only when the API is unreachable for all clusters and no replacement workers can be launched. Its docs promise "a 99.99% SLA"; the contract says 99.9%.
  • Which regions are "Core" is on a list we could not read. Gcore's cloud map API tags Amsterdam, Helsinki, London, Paris and Warsaw as edge locations but not Luxembourg or Frankfurt; our reading, not Gcore's statement, is that those two are Core.
  • Large GPU clusters lose SLA cover unless a "Spare Reserve" of at least 8% of nodes is kept. Maintenance never earns credits.
  • Liability is capped at 50% of the fees paid under the affected Service Orders in the previous 12 months, plus that month's fees; claims are barred 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. Gcore's 99.95% for a single VM is a higher target than AWS's single-instance 99.5%, but its largest credit is a fifth of AWS's, and it depends on your report.

Status page and incidents. status.gcore.com keeps only the latest 50 entries in its public feed, which on 16 October 2026 reached back to 1 August. None of the entries we read gives a root cause, and we found no postmortems. In those 11 weeks the Luxembourg locations had five incidents, each confined to one location, and the customer portal failed everywhere once:

DateScopeDuration
4 Aug 2026S3 storage, Luxembourg: partial outage4 h 57 min
26 Aug 2026Luxembourg-2 control plane: create, start, stop and resize failing; running VMs unaffected2 h 16 min
13 Sep 2026Luxembourg-2: slow or failing VM and load balancer creation2 h 19 min
15 Sep 2026Luxembourg: VM drives "entering a read-only or unavailable state"1 h 26 min
7 Oct 2026Customer Portal, global: unavailable, login problems39 min
15 Oct 2026Luxembourg-2: "partial outage of all Platform Services"42 min

Elsewhere in the EU the feed shows a partial SFTP storage outage in Eschborn of nearly three days (1 to 4 August), a Warsaw connectivity outage and three Amsterdam entries in October, on bare metal and networking. Only the portal outage spread across regions.

Support

PlanEUR a monthP1P2Channels
Basicincludedbest effortbest effortchatbot, email, tickets; English only
Standard29not coverednot coveredP3 under 4 hours, P4 under 8
Advanced200under 1 hourunder 2 hoursadds calls; several languages
Enterprise500under 15 minutesunder 30 minutesadds Slack, Teams, Telegram

The docs add that "Gcore may lower a request's priority if the issue is not caused by a Gcore service error" and that workload "can extend response times beyond a plan's guarantee". A one-hour target for a production outage costs EUR 200 a month at Gcore, against the greater of USD 29 or 9% of spend on AWS Business Support+ and USD 100 a month on Azure Standard.

Getting an account

Signup takes an email address (or a Google or GitHub login), a payment method with a billing address and at least EUR 4 on the balance. The signup docs describe no identity check, but the contract says "all contracts with Gcore are contingent upon successful completion of the KYC check", which the privacy policy says may cover "proof of identity, proof of address, PEP information". Gcore screens the customer's "beneficial owners, or directors" against sanctions lists.

Default quotas for a new account are not published. Raising one is a request that Gcore reviews, with at most ten open at once; bare metal always needs an approved request, and Managed PostgreSQL a one-time activation per project and region. Whether workload B fits the defaults on day one we could not establish without an account. 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.

If a payment fails, every product on the account is suspended. VMs, volumes, snapshots and bare metal are deleted 48 hours later; Managed Kubernetes, functions, containers and GPU spot clusters at once; PostgreSQL and logging after 14 days. Gcore may also suspend "without prior notice" to prevent fraud or "potential material liability or a third-party claim".

Compliance and data protection

FrameworkStatus
ISO/IEC 27001Claimed on the website; certificate and scope not seen
PCI DSSClaimed: "Our Cloud has passed the annual QSA audit"; attestation not seen
SOC 2Claimed on newer pages; type and period not stated, report not seen
ISO 27017, ISO 27018, BSI C5, SecNumCloud, HDS, TISAX, ENSNone found

Gcore's trust centre renders only with JavaScript and appears to release reports on request; we could not read it.

The Data Processing Agreement of 17 June 2026 gives Gcore a general authorisation for sub-processors and transfers data outside the EU and EEA "relying by default on Standard Contractual Clauses". The sub-processor list of 25 November 2025 names Gcore companies in Germany, Luxembourg, Cyprus, the United States, Serbia, the Philippines, Georgia and Uzbekistan, plus Microsoft Ireland, Zendesk Inc. in the US and HubSpot Ireland. It names no data centre operator. You pick the region, and Gcore markets five "dedicated EU regions": Luxembourg, Frankfurt, Paris, Amsterdam and Warsaw. We found no contractual commitment that support or operations staff reach customer data only from inside the EU.

Who controls it

Gcore publishes no list of its shareholders. The tables give each source and its type: P is the company, a registry or a filing; S is the press; U is a user's copy of a document. We draw no conclusion on which laws reach Gcore.

FactSourceType
You contract with G-Core Labs S.A., Contern, Luxembourg (RCS B184832), under Luxembourg law, with arbitration in LuxembourgMSA, 17 June 2026P
The group holding is G-Core Holding S.A., Luxembourg, with equity of EUR 52,973,549 at the end of 2025 and a 2025 loss of EUR 18,272,725Northern Data AG's 2025 accounts, in Rumble's prospectus of 14 April 2026P
Gcore called its USD 60 million Series A of July 2024 "the first external raise since the company was started over a decade ago". It was "led by Wargaming, with participation from Constructor Capital and Han River Partners"Gcore; Business WireP
Paperjam describes G-Core Labs as the company of Victor Kislyi, the Belarusian billionaire who heads Wargaming (undated, about 2020)PaperjamS
taz wrote that two billionaires from Belarus with Cypriot passports, Victor Kislyi and Nikolai Katselapov, stand behind G-Core, and that Wargaming gave it start-up financing in 2015 (14 July 2023)tazS
On 2 April 2025 Northern Data AG (Frankfurt am Main) took two Gcore board seats and "secured an option to acquire a majority of the outstanding capital stock of Gcore at a predefined valuation", subject to due diligenceNorthern DataP
Northern Data subscribed 2.13% of G-Core Holding S.A., treated Gcore as an associate from March 2025 because it "has significant influence through its governance rights and participation in key strategic decision-making", and lent Gcore EUR 10 million interest-free until August 2029Rumble's prospectus, notes 4.4, 4.6 and 5.10P
Northern Data funded a further EUR 5 million loan commitment to G Core Holding S.A. in May 2026. In July 2026 it "entered into amended arrangements"; because of "changes to the governance and decision-making rights", it concluded that it "no longer has significant influence", and Gcore ceased to be an associateNorthern Data's first-quarter 2026 statements, notes 4.2.2 and 4.5, filed by RUM Group on 28 August 2026P
Rumble Inc. (Nasdaq: RUM, Florida; renamed RUM Group Inc. on 18 June 2026) closed its purchase of about 85.2% of Northern Data on 17 June 2026RumbleP

Luxembourg's register of beneficial owners is not public, so neither press report could be checked at a registry. No source we found says whether Northern Data's option was exercised, has lapsed or is still open: neither Rumble's prospectus nor Northern Data's first-quarter 2026 statements mention it, and the July 2026 amendment is described only by its effect. Through Northern Data, a US-listed group holds the stake and the loans, and the option if it is still open; whether it keeps the two board seats is not stated.

The history with Russia, as the sources give it:

FactSourceType
2017: G-Core Labs bought the Russian CDN company Skypark CDN, through G-Core Holding, and renamed it G-Core Rus LLCComNews, 16 March 2026S
18 March 2022: G-Core Rus renamed itself EdgeCenter; in April 2022 G-Core Labs left its capital. Kommersant paraphrased Gcore's website as saying it would stop providing services in Russia and that part of its Russian team would continue as the independent EdgeCenter. We did not find Gcore's own statementKommersant, 4 June 2022; ComNewsS
Gcore told Correctiv it holds no shares in EdgeCenter and has no influence over itCorrectiv, 17 February 2023S
31 March 2023: Gcore emailed customers that "due to the current regulatory and legal restrictions, we will be unable to continue providing resources in Russia and Belarus". Cloud and hosting resources there were suspended that night, and data could be retrieved until 1 May 2023a customer's copy of the email on LowEndTalk, archivedU

Gcore does not say who operates its data centres; the contract refers to leased co-location space. Northern Data said the joint network is "powered by Northern Data's existing state-of-the-art infrastructure", but no source ties a named Gcore region to a Northern Data facility. Field names in the cloud API, such as keystone_id, point to OpenStack underneath; Gcore does not say so.

Leaving

Traffic out of VMs and GPU clusters is free; object storage egress is about EUR 18.63 per TB. The exits use standard tools: S3-compatible storage, CNCF-conformant Kubernetes, plain PostgreSQL with dump and restore documented, and NFS. Custom images can be uploaded, but downloading a VM image or a snapshot is not documented.

The EU Data Act addendum of 13 January 2026 sets the timeline: a Switching Request at least two months ahead, a transitional period of up to 30 days starting one month after receipt, then at least 30 days to retrieve data, with fees running throughout. Gcore "may apply standard data egress charges as permitted by applicable law", and a fixed-term contract still owes the early termination fee, the remaining monthly charges times the months left. Pay-as-you-go orders end on 14 days' written notice. "Customer remains solely responsible for backups."

Lock-in sits at the edge and in the AI products: FastEdge apps, Everywhere Inference deployments, CDN, WAAP and DDoS configuration, and GPU reservations of 12 or 36 months.

What users report

Independent reports on Gcore's cloud are thin, and mostly about CDN, DNS and DDoS rather than VMs; Reddit could not be read. Three themes meet our bar.

Quota increases are manual and can take days. SemiAnalysis' ClusterMAX review found quota decisions made by "a nameless faceless support team member", with a storage quota taking "three separate attempts (over the course of two working days)" (ClusterMAX 2.0, November 2025). Gcore's docs confirm that requests are reviewed (quotas).

Kubernetes on bare-metal and GPU nodes works, self-service. A user reports that "their managed kubernetes even supports bare-metal nodes, which is actually something I'm using" (41739402, October 2024), and ClusterMAX called it "one of the best purely self-service offerings we tried for kubernetes" (November 2025).

Node-pool handling has rough edges, and fixes are slow. In Terraform, node pools are replaced before the new nodes lose their taints, about 5 minutes of downtime when all are replaced (issue 88, opened March 2024, first maintainer reply December 2024, still open), and every other apply recreates all node pools (issue 266, opened September 2025, still open). ClusterMAX's first cluster was "stuck in a "provisioning" state for over two hours before ultimately failing".

Who it suits, and who should look elsewhere

Gcore suits teams that want CDN, DDoS protection, WAF and DNS at the edge with VMs and Kubernetes behind them on one bill; services that send a lot of traffic out of VMs; and AI teams that want self-service H100 nodes in Luxembourg, Helsinki or Portugal at published prices, or Kubernetes on bare-metal and GPU nodes.

Look elsewhere if you need a production managed database or other managed data services, failover across zones within one region, certificates in hand before you sign, or quota increases without a person reviewing them. Look elsewhere too if your definition of a European provider requires a published, fully European ownership chain and processing only by EU entities.

Sources

Checked on 15 and 16 October 2026.

Pricing

Terms and SLA

Documentation

Status and incidents

Compliance

Company and ownership

User reports

Data & Cloud

New guides and rule changes by email

When a rule changes, a deadline moves or we publish a new guide, we send it by email.

We send a confirmation link first. You can unsubscribe at any time. Privacy Policy