EU Cloud Provider News Roundup: Late February – May 2026

This is the second edition of our monthly EU Cloud Provider News Roundup, covering everything that happened across our tracked providers between February 17 and May 20, 2026. The previous edition ended at February 16, so nothing here repeats what we already covered.

Key Takeaways

Pricing went up across the EU, and providers had to explain themselves. Three of the biggest EU providers raised prices in a single quarter. Octave Klaba published a long-form justification of OVHcloud’s Public Cloud, Bare Metal and VPS increases. Scaleway followed with a “transparent update on pricing” post. Then Hetzner — historically the discount benchmark — announced an across-the-board hike effective April 1, with cloud servers up 30-37% and storage up to +53%, applied to existing contracts. Common stated cause: hardware costs (DRAM +171% YoY, NAND doubling, HDDs sold out for the year) and energy. The discount-cloud era in the EU just got measurably more expensive.

Sovereignty stopped being a slogan and became a product roadmap. Cyso publicly migrated off OpenStack and is moving from VMware to CloudStack, explicitly framing the move as a sovereignty play. Scaleway dedicated two posts to SecNumCloud. Civo published three posts in May tearing into hyperscaler “sovereignty washing” and the marketing of “European Cloud” labels on US-owned infrastructure. STACKIT, OVHcloud, and Exoscale kept hammering the same theme. When this many providers spend this much editorial budget on one narrative, the budget is coming from somewhere — probably from sales pipelines that now actually close on sovereignty grounds.

A Linux kernel zero-day forced everyone to patch K8s at once. Copy.Fail (CVE-2026-31431) hit Kubernetes nodes across providers in late April / early May. Thalassa, OVHcloud, and Azure all published mitigation guidance within days. If you’re running self-managed K8s on any EU IaaS, this is the kind of incident that justifies paying for managed K8s.

AI inference workloads are now the centre of gravity for K8s product development. Azure shipped a five-part AI-on-AKS series, NVIDIA Dynamo / MIG / DRA integrations, and per-app token rate limiting. AWS shipped Trainium DRA support, multi-account observability, and EKS Hybrid Nodes Gateway GA. STACKIT shipped MLflow Tracking (AI Model Experiments), Telemetry Router, and a GPT-OSS 20B model. Civo previewed NVIDIA Vera Rubin coverage and signed a UK sovereign-LLM partnership. The bulk of new managed-K8s features this quarter is AI-shaped.

Hyperscalers still ship more, faster. GKE pushed Kubernetes 1.36 to Rapid in May, GA’d Pod Snapshots and Accelerator Network Profile, and added concurrent node-pool upgrades. AKS shipped Azure Container Storage v2.1 GA with Elastic SAN. AWS Hybrid Nodes Gateway hit GA. EU providers move slower on infrastructure primitives, but they’re winning on positioning.

---

Hetzner

The headline this quarter: Hetzner joined OVHcloud and Scaleway with a major price hike. Plus continued API hardening and TLS modernization.

Price adjustment as of April 1, 2026 (announced Feb 23, effective Apr 1): Across-the-board increase covering Cloud, dedicated servers (EX/AX/RX/SX/GPU/BRANDS), Storage Box, Object Storage, Storage Share, web hosting, Load Balancers, SSL, and domains. Applies to both new orders and existing contracts — no grandfathering. Reported ranges: cloud servers +30-37%, object storage +30-53%, dedicated servers ~+20% (AX42 €47.30 → €57.30; DX293 €305.60 → €355.60). Entry-level CX23 went €2.99 → €3.99. Hetzner cites “dramatically” higher hardware acquisition costs — DRAM up 171% YoY, NAND flash doubling, HDD supply consumed by AI buildout. See Hetzner’s price-adjustment docs page for the per-product table. This is the third major EU price hike in three months and the largest one — Hetzner customers should re-cost any auto-scaled or commitment-free workloads.

Post-quantum key agreement on Load Balancers (Mar 9): X25519MLKEM768 hybrid for TLS 1.3. Hetzner is ahead of most EU providers on PQ-readiness.

Datacenter fields deprecated for Locations (Apr 1): datacenter.server_types / recommendation removed October 1, 2026. Replaced by server_types.locations.available / .recommended. Audit your tooling now.

Primary IP API rework (Apr 27): assignee_type optional on POST /v1/primary_ips. From August 1, unassigned IPs return unassigned instead of server. Prerequisite for attaching Primary IPs to non-server resources.

Resource-scoped action GET endpoints deprecated (Apr 30): Use unified GET /actions/{action_id} instead.

Configurable Load Balancer idle timeout (Apr 30): http.timeout_idle now 30-300s (was fixed 50). Useful for long-polling and SSE.

Deletion of assigned IPs blocked (May 1): Last edition’s deprecation went live. Returns must_be_unassigned. Terraform Provider v1.60.0+ required.

cloud-init Private Network auto-config (May 18): Images with cloud-init 25.3+ auto-configure Private Networks. Removes hc-utils need on Ubuntu 22.04/24.04/26.04 and Fedora 44. Ubuntu 26.04 and Fedora 44 images also now available.

Minor: SHA-1 TLS dropped Apr 22, LB target cleanup Apr 13.

OVHcloud

The big news isn’t on the engineering blog — it’s Octave Klaba’s pricing post.

Pricing evolution of Public Cloud, Bare Metal and VPS (Mar 5): Klaba walks through why prices have to go up: energy, hardware costs, EU regulatory burden, and infrastructure modernization. It’s an unusual move — most cloud price hikes ship as a footnote. This one came with a forum-style explanation from the founder. If you’re an OVHcloud customer, read it; if you’re a competitor, watch the comment thread.

Running an Ethereum Node on OVHcloud Public instances (Mar 9): Practical guide. Niche but solid.

VMware Cloud Foundation 9: Evolving Private Cloud In One Go (Mar 20): Follow-up to February’s VCF 9.0 announcement, focused on the upgrade path. OVHcloud continues to position aggressively for Broadcom refugees.

DNS records: SVCB/HTTPS developments (Apr 21): Explainer of SVCB and HTTPS DNS records. Useful background for anyone running their own DNS.

How Mia Experts Is Reinventing Medical Software with AI and Sovereign Cloud (Apr 22): Customer story on regulated AI on a French sovereign cloud. The sales pitch is obvious; the case is real.

Copy.Fail (CVE-2026-31431) on Managed Kubernetes (Apr 30): Rapid-response guide for protecting MKS clusters against the Linux kernel zero-day. Same root vulnerability that Thalassa and Azure patched within days. The MKS team’s transparency on this is worth noting.

Enterprise File Storage with Trident CSI on MKS (May 11): Integration guide for OVHcloud EFS as a CSI-backed PV class on Managed Kubernetes. Real shared-filesystem story for K8s workloads, finally.

Devoxx France 2026 recap (May 19): The recap itself is conference-coverage filler, but the editorial quote stood out: “Three years ago, people were saying: ‘I don’t care about sovereignty, I’ll just choose the cheapest option.’ This year, the tone has clearly changed: ‘How can we use your sovereign products?’” If you believe the OVHcloud sales team’s read on customer conversations, the European market actually shifted in 2025-2026.

Scaleway

A pricing post, a data warehouse, and a SecNumCloud doubleheader.

Data Warehouse for ClickHouse: Power without compromise (Apr 20): Managed ClickHouse data warehouse. Slots in next to their existing MongoDB and Serverless SQL offerings. Pricing was not in the announcement, which is the usual Scaleway pattern.

A transparent update on Scaleway pricing (Apr 27): Three weeks after OVHcloud’s pricing post, Scaleway published their own. Same general framing: energy, hardware, sovereignty compliance costs. The pattern repeating across two French majors in one month is telling.

Top-Level Domain price updates (Apr 27): TLD-by-TLD price changes, published same day as the broader pricing update. At least they batched the bad news.

SecNumCloud: Understanding the trusted cloud standard (May 11) and SecNumCloud: The Strategic Challenges of The Qualification (May 18): Two-part explainer on ANSSI’s SecNumCloud framework. Scaleway is positioning hard around the French-state-trusted cloud designation, which excludes most US providers by construction.

What is File Storage? (May 11): SEO content. Skip.

STACKIT

Most active EU provider in our tracking — ~20 substantive release notes in three months. They are clearly investing.

Kubernetes 1.32 deprecated in SKE (Feb 20): EOL upstream Feb 28, removed from SKE Apr 15. Auto-upgrade to 1.33 in maintenance window. SKE lifecycle now aligned with upstream patchdays.

IaaS-API v1 deprecated (Feb 26): Sunsets March 1, 2027. v2 required for multi-region resource creation.

STACKIT Logs GA (Mar 3): Managed Grafana Loki, production SLAs, 1-180 day retention, native Grafana/Alloy/Fluentd/OTEL Collector support.

AI Model Serving: GPT-OSS 20B replaces Mistral-Nemo and Llama 8B (Mar 4): openai/gpt-oss-20b with 4-bit MXFP4. Migration deadline June 4. RAG/agentic workloads need re-validation.

STACKIT Pipelines: managed runners (Mar 17): ~90% GitHub Actions compatible, in-Docker isolation, native Container Registry and Secret Manager integration. Goal: full SDLC stays inside their network.

EU02 region buildout (Mar-Apr): Windows Server (Mar 17), RHEL images (Mar 23), KMS (Apr 7). Multi-AZ HA without crossing regions now plausible.

OTLP log ingestion (Mar 24): Native OTLP endpoint alongside legacy URL.

Unified Firewall (Preview) (Mar 25): Single pane for ACLs, public IPs, security groups. Free for all customers, Terraform on roadmap.

Object Storage: Object Lock (Apr 2): Compliance immutability up to 365 days.

SKE: IDP users for kubeconfig (Apr 2): STACKIT IDP user/service accounts directly for cluster access. Also version expiry dates in API — automate forced-upgrade warnings.

AI Model Experiments (Preview) (Apr 7): Managed MLflow Tracking. Free during preview, pay only for artifact Object Storage. Real gap-filler in EU ML platform space.

Telemetry Router (Preview) (Apr 7): Centralized OTLP audit-log ingestion. Legacy Audit Log API decommissions June 2 — migrate now if you depend on it for compliance.

Incremental Block Storage backups deprecated (Apr 22): Maintenance mode now, permanent discontinuation October 31, 2026. Full backups remain.

CDN: Image Optimizer + traffic/security controls (Apr 24/28): WebP conversion, smart resizing, JS/CSS minification, per-distribution stripResponseCookies, forwardHostHeader, TLS 1.0/1.1 controls.

VPN API (May 5): IPsec IKEv2, active-active HA, BGP, site-to-site/multisite, 100 Mbit/s - 1 Gbit/s.

Run Command & Server Agent GA (May 18): Server Agent pre-installed on all public images by default. Shell/PowerShell via API. Right primitive for fleet management.

Civo

Civo continues to publish at high volume. The editorial is consistently sharper than the average provider blog.

An Introduction to CivoStack Enterprise (Mar 6): The Civo platform packaged for on-prem and partner deployments. Pairs with February’s Konstruct launch — Civo is building a real platform business beyond their public cloud.

Introducing hosted control planes on Konstruct (Mar 23): Hosted K8s control planes on Konstruct. The vCluster-style architecture is getting commercialized fast across providers.

The UK’s first sovereign LLMs are here (Apr 14): Partnership with LocAI Labs to host UK-sovereign large language models. The “sovereign LLM” framing is going to matter as procurement teams formalize requirements.

How are hyperscalers misleading the cloud industry? (May 13): Sequel to February’s “sovereignty washing” post. More pointed, more specific examples. Civo is the most aggressive voice in EU cloud right now on this topic.

The benefits of decentralized AI infrastructure (May 14): Argument for distributing AI inference across regional providers vs. centralizing on hyperscaler GPU clusters.

Multi-cloud vs. hybrid cloud (May 15): More positioning content, but with clearer-than-average distinctions.

What Vera Rubin means for AI infrastructure in 2027 (May 18) and NVIDIA Vera Rubin: What is it, when you can get it (May 20): Two posts on NVIDIA’s next-gen platform. Civo is positioning to be on the supply list early.

Exoscale

A mix of one strong technical post, one customer story, and one storage primer.

How to deploy Matrix on Kubernetes: a complete guide (Mar 13): Production deployment of Matrix on SKS with EU data residency. Genuinely useful for organizations migrating off Slack/Teams for sovereignty reasons.

What is Object Storage? (Apr 7): SEO content. Decent quality.

The Best Discord Alternatives for Companies (Apr 8): Same playbook as the Matrix post — Exoscale is leaning into the “you should not be on US chat platforms” angle.

Platform Engineering on Exoscale: Building an IDP (guest post) (Apr 27): Gravitek case study on building a self-service developer platform on Exoscale. Concrete details on autoscaling and security.

The 9000-to-1500 cliff: an IPv6 PMTUD field note (May 4): 22-minute network debugging deep-dive. ECMP, jumbo frames, missing TCP MSS clamping for IPv6. This is the best technical post out of any EU provider this quarter. If you operate networks, read it.

Thalassa Cloud

One post, but it’s a security one.

CVE-2026-31431 (Copy.Fail): patched Kubernetes images (May 1): Kernel-level vulnerability patched in Thalassa Kubernetes images v1.34.7-1 and v1.35.4-1. Same vulnerability addressed by OVHcloud’s MKS team on April 30 and Azure’s AKS team on May 11. The speed of EU-wide response was decent.

Cyso

Cyso is repositioning hard, and they’re being unusually open about the engineering reasoning.

Why “European Cloud” means (almost) nothing if your cloud provider is American (Mar 3): Argument that EU regions of US-owned clouds don’t deliver sovereignty. Standard talking point delivered well.

From OpenStack to CloudStack: why we are moving away from OpenStack after 10 years (Mar 18) and Beyond VMware: Cyso’s road to CloudStack (Mar 22): Two-part series on Cyso’s migration away from both OpenStack and VMware to Apache CloudStack. The transparency about why OpenStack didn’t work after a decade is unusual. CloudStack is having a quiet moment as the open-source survivor in the post-Broadcom VMware landscape — this is one of several similar migration stories landing this year.

Meet Cyso Cloud at Cloud & Cyber Security Expo Frankfurt 2026 (Apr 7): Event announcement.

Cyso Cloud Engineer Becomes Kubestronaut – CKS certified (Apr 19): Team news. Worth flagging because the CKS pipeline at smaller EU providers is a real differentiator — if you care about K8s expertise depth from your provider, certifications like CKS and Kubestronaut are signal.

Webdock

Big private interconnects + API v1.1 (Apr 20): New direct paths into Facebook, Google, Amazon, and Microsoft networks. API v1.1 adds custom profiles, script deployment, SSL certificate management, and removes auth requirements for some endpoints. Webdock is a small VPS provider, but the direct-interconnect investment is sensible — peering into hyperscaler networks is where most user traffic actually goes.

netcup

Minecraft Server: the quick overview for beginners (Feb 18): Tutorial content. Skipping.

VPS for small teams: self-host internal tools (Apr 23): Positioning post for SMBs consolidating off SaaS.

WordPress hosting: which option suits your project? (May 8): Hosting selection guide.

Root Server or VPS: which fits your project? (May 13): Useful comparison if you’re new to netcup’s lineup.

New SCP UI: beta phase starts May 18 (May 18): Server Control Panel redesign in open beta. The current SCP is dated — this is overdue.

Hostinger

A lot of marketing volume, two interesting product items.

Hostinger Trust Center launched (Mar 27): A real Trust Center page covering security posture, compliance, and incident response. The reverse-engineering of the February Notepad++ incident clearly continues — this is the kind of follow-through that other providers should copy.

OpenClaw one-click AI agents (Mar 24): One-click AI agent deployment for OpenClaw. (Note: an earlier February announcement covered the same product; this is the GA / promotional iteration.)

Horizons integrated AI (Apr 21): AI now built into Horizons, no integration step required. The customer-facing message is “10% of you were already wiring in external models — we did it for you.”

Hostinger Agents: $7/month (Apr 9): “Seven-person business team for $7/month” — Hostinger’s bundled agent suite for SMBs. Aggressive price point for the segment.

Your VPS terminal now speaks human (May 11): Kodee (their AI assistant) in the VPS terminal. Natural-language-to-shell for the small-team VPS market.

plusserver

Quiet quarter publicly, no substantive announcements landed in the data we monitor.

Serverspace

High publication volume, mostly tutorials. One that stands out:

Best OVH Alternatives in Europe (May 16): Serverspace publishing a competitive comparison page targeting OVHcloud customers. Aggressive but useful — if you’re considering OVH and want to see the EU alternatives, this is a starting point (with the obvious caveat that Serverspace puts itself near the top).

AWS EKS

Steady cadence on the Containers blog. EKS leaning into AI inference and operational maturity.

Containerized hybrid nodes (Feb 24): Docker containers connecting to EKS as hybrid nodes. Edge/on-prem extension. Paired with EKS Capabilities deep dive same day on ACK + kro orchestration.

Production GenAI at the edge with Hybrid Nodes + NVIDIA DGX (Mar 18): Low-latency LLM inference on-prem with centralized EKS management.

Session policies for EKS Pod Identity (Mar 24): Dynamic per-pod-session IAM scoping. Real least-privilege win.

Navigating networking with EKS Auto Mode (Apr 14): VPC CNI, pod density, hybrid connectivity guidance.

SPIFFE/SPIRE authorization on EKS (Apr 27): Workload identity. Long-overdue formal guidance.

EKS Hybrid Nodes gateway GA (May 1): GA for any EKS workload bridging cloud and on-prem.

Inter-AZ and NAT gateway traffic tracking (May 5): Visibility on the two AWS line items that surprise people most. Cost-optimization gold.

Cross-region DR using AWS Backup (May 6): End-to-end DR walkthrough with persistent data restoration.

Centralized observability for multi-account EKS (May 12): Hub-and-spoke monitoring.

Trainium + EFA with Kubernetes DRA (May 18): DRA for Trainium accelerators and EFA. AWS doubling down on Trainium-as-NVIDIA-alternative.

Also: PCI DSS architectures (Apr 1), knowledge graphs via AWS DevOps Agent (Apr 9), Velero backup/restore (May 12), Auto Mode cost governance (May 13), KubeCon EU recap (Mar 12).

GCP GKE

A heavy quarter. GKE is shipping faster than the other hyperscalers on raw feature volume.

HealthCheckPolicy validation tightened (1.34+) (Apr 13): Stricter validation on the Gateway API HealthCheckPolicy CR. Existing mismatched resources are exempted but can’t be updated. Mostly catches latent bugs.

Accelerator Network Profile GA (Apr 20): Automated networking config for GPU/TPU workloads is now GA. Major usability win for AI clusters.

C4A c4a-highmem-96-metal and Slurm Operator add-on (Preview) (Apr 21): Bare-metal Arm flavors and Slurm-on-GKE for HPC workloads.

AI Zones support (Apr 27): New zone config option for AI workloads.

Kubernetes 1.36 in Rapid channel (May 1): Mutating Admission Policies GA, L4 ILB improvements. GKE remains the fastest hyperscaler to land new K8s minor versions.

GKE Pod Snapshots GA (May 6): Generally available on v1.35.3-gke.1234000+. C3D Confidential Nodes now support live migration with AMD SEV. Cloud Storage FUSE fix for ARM64 64KiB-page nodes.

Concurrent node pool upgrades (Preview) (May 14): Configurable max-simultaneous node pool upgrades. Managed OpenTelemetry multimodal prompt/response capture for LangGraph and ADK agents (Preview). kubectl removed from /usr/bin in COS milestone 129+ — automation that shells into nodes may need adjustment.

Azure AKS

Five-part AI-on-AKS series anchored by Azure Arc, plus storage and observability GAs.

Scaling Anyscale Ray on AKS (Feb 13): Reference architecture for large-scale distributed ML.

GPU partitioning with DRA (Mar 3) and vGPU with DRA (Mar 6): MIG and NVIDIA vGPU via Kubernetes Dynamic Resource Allocation. Fine-grained GPU sharing for inference.

NVIDIA Dynamo Part 3 (Mar 16): Claimed >20x TTFT and >4x E2E latency wins. Verify yourself, but Dynamo is real.

AKS Control Plane Enhancements (Mar 30): Streaming LIST responses, etcd optimization. Real impact for clusters where API server bottlenecked.

DRANET RDMA optimization (Apr 1): Topology-aware GPU + NIC co-scheduling for HPC-grade AI networking.

AI Inference on AKS via Azure Arc — 5-part series (Apr 7-9): Edge/on-prem AI with Arc-enabled AKS. Covers ResNet-50 predictive, open-source LLM serving, Triton + TensorRT-LLM. Most coherent EU-relevant AI series from a hyperscaler in months — Arc matches the sovereignty workload pattern.

Azure Container Storage v2.1 GA + Elastic SAN (Apr 8): Changes the cost/perf tradeoff for shared block storage on AKS.

Per-app token rate limiting via agentgateway (Apr 17): Workload-identity-aware rate limits. FinOps primitive that makes GenAI costs governable.

Copy.Fail + DirtyFrag CVE patching at scale via Fleet Manager (May 11): Safe rollout of CVE-2026-31431 and CVE-2026-43284/43500 across many clusters. Fleet Manager getting real adoption-driven validation.

Also: NAP placement best practices (Mar 20), Agent Skills for AKS (Apr 8), Argo CD + Entra ID SSO (Apr 22).

What We’re Watching in June

• OVHcloud’s first Italian datacentre opens May 21 (just outside this window). Italy gets a major sovereign-EU option; expect heavy marketing through summer.
• STACKIT legacy Audit Log API decommissions June 2. If you depend on it, migrate to Telemetry Router now.
• STACKIT GPT-OSS 20B migration deadline June 4. Mistral-Nemo and Llama-3.1-8B workloads need to move.
• Hetzner old shared-vCPU server types went unavailable for new orders January 1; existing servers still run but the migration window is open. Worth budgeting time before something forces your hand.
• Apache CloudStack momentum. Cyso’s two-post series is one of several similar migrations underway. We’ll track which other EU providers move off VMware or OpenStack onto CloudStack through the rest of 2026.
• EU pricing wave. OVHcloud, Scaleway, and now Hetzner have all raised prices. Watching whether STACKIT, IONOS, UpCloud, or netcup follow — if hardware-cost pressure is real, smaller providers will have less margin to absorb it.
• NVIDIA Vera Rubin general availability. Civo is positioning early; expect every EU GPU-cloud provider to publish a Vera Rubin post in Q3.

---

This is the second edition of our monthly EU Cloud Provider News Roundup. The previous edition covered January and the first half of February 2026. Subscribe or check back monthly for the next one.

Compare pricing across all providers mentioned above with our Kubernetes price calculator.