State of European Public Cloud 2026

European providers' share of their own cloud market has been stuck around 15% since 2022, roughly half its 2017 level [1]. The market itself grew roughly 24% in 2025 on Synergy Research's reckoning, to something over €75 billion, and European providers captured almost none of that growth. AWS, Microsoft and Google alone take around 70%.

Year	European providers' share	Market size (Synergy)
2017	~27-29% (baseline varies across Synergy's own releases)	not broken out
2020	just under 16% [54]	~€5.9B/quarter
2022	13-15% [55]	~€10.4B/quarter
2024	15% [1]	€61B/year
2025	~15% [1]	>€75B/year (projected)

The share halved, then froze, while the market kept compounding. As a second source on size: IDC puts the EMEA public cloud market (IaaS+PaaS+SaaS, a wider scope) at $203 billion in 2024, heading for $415 billion by 2028 [56]. No second analyst publishes a European-provider share split, so the 15% cannot be independently cross-checked; treat its error bars as unknown. Every debate in this report lives inside that table.

One caution before reasoning about the 15% as a single number: it aggregates at least three businesses that barely compete with each other. Self-service developer cloud (Hetzner, entry-level Scaleway) is roughly holding its ground; enterprise hosting and private cloud is where the halving happened; AI infrastructure is a new race entirely. The blended figure hides the only actionable fact.

This report writes down where European public cloud actually stands in August 2026: real prices, real revenues, real migrations, and the claims that do not survive contact with the numbers. "Europe" here follows Synergy's definition and includes the UK, which alongside Germany is one of the two largest national markets; note that the hyperscalers' single biggest 2025 European commitment, Microsoft's $30 billion for the UK, sits outside every EU policy lever discussed below. How we gathered and checked the numbers is described in the Methodology section at the end; numbered references throughout point to the source list. Where a number cannot be known, we say so instead of guessing.

Key takeaways

European providers hold 15% of their own market, flat since 2022, and the flagship OVHcloud just posted its first profit in 26 years: €0.4 million on €1.08 billion of revenue [2]. Twenty-six years of competing with hyperscalers buys a 0.04% margin.
Egress is the biggest lock-in tax: past the free allowance, one terabyte out of AWS costs about $92; out of Hetzner, €1. A company pushing 500 TB a month pays AWS roughly $30,000 for traffic that costs about €480 at Hetzner. Not a price, an exit tariff.
Hetzner, Europe's price benchmark, raised x86 cloud prices up to 175% in a single June day, blaming DRAM up 171% year over year. "Europe is the cheap option" is no longer a safe assumption.
Microsoft admitted under oath, before the French Senate, that it cannot guarantee French data will never be handed to US authorities. Nothing legally new, but procurement teams can now cite sworn testimony instead of a legal memo.
The migrations that actually happened in 2025-26 were mostly Microsoft 365 exits, not IaaS moves. The big exception: France's Health Data Hub is leaving Azure for Scaleway.
The AI layer is the one segment where Europe competes from day one: the cheapest H100s accessible from Europe are sold by a European provider, Mistral is filling its own datacenter near Paris, and the EU has put €10 billion on the table for "AI gigafactories". Every GPU in that sentence is Nvidia's.

Eighteen months in ten lines

When	What happened
Feb 2025	Microsoft completes the EU Data Boundary; US sanctions hit ICC prosecutor Karim Khan
Jun 2025	Microsoft France admits under oath it cannot shield French data from US law
Sep 2025	Data Act switching rules apply; Nebius signs its $17.4B Microsoft deal; JUPITER, Europe's first exascale computer, comes online
Oct 2025	OVHcloud posts its first-ever net profit; the ICC moves to OpenDesk; the Commission publishes its cloud sovereignty framework (SEAL)
Dec 2025	S3NS (Thales + Google) wins SecNumCloud across IaaS, CaaS and PaaS
Jan 2026	AWS European Sovereign Cloud goes live from Brandenburg; Cybersecurity Act 2.0 proposed
Apr 2026	Commission awards its €180M cloud tender to European-led lots; Health Data Hub announces Azure to Scaleway; Cohere announces Aleph Alpha acquisition; Hetzner's first price round (+30-43%)
Jun 2026	Hetzner's second round: +113-175% on x86 lines; the Cloud and AI Development Act is proposed
Jul 2026	EU opens the €10B AI gigafactories call; Evroc goes live; Infomaniak announces its Swiss listing
Aug 2026	This report's snapshot

What counts as "European cloud"

Definitions decide conclusions here, so ours is explicit. A European cloud provider is a company that operates its own cloud control plane and whose ultimate parent is both headquartered in Europe and not controlled by a non-European owner. Headquarters alone is not enough: a Frankfurt datacenter owned from Seattle answers to Seattle, and the same logic applies in reverse to letterbox structures. By that test the list includes OVHcloud, Scaleway, Hetzner, IONOS, Infomaniak, Exoscale, UpCloud, Aruba, StackIT, T-Systems, and, uncomfortably, Nebius (Amsterdam-headquartered and US-listed, with dispersed ownership). It excludes the "sovereign hyperscaler" joint ventures (Bleu, S3NS, Delos Cloud), which are European-operated shells around Microsoft or Google technology: they matter, and they come back below, but calling them European cloud providers would be scorekeeping fraud.

Where they sit:

Country	Providers
France	OVHcloud, Scaleway
Germany	Hetzner, IONOS, StackIT, T-Systems
Switzerland	Infomaniak, Exoscale
Italy	Aruba
Finland	UpCloud
Netherlands	Nebius
Sweden	Evroc

The market map: almost no one publishes real numbers

Provider	HQ	Owner	Latest revenue (cloud share noted)	Jurisdiction of parent
OVHcloud	France	Listed (Euronext)	€1,084.6M FY2025, +9.3% (all cloud) [2]	EU
Scaleway	France	Iliad Group	Not disclosed separately	EU
Hetzner	Germany	Private (family)	Not disclosed (last filing: €367M in 2021)	EU
IONOS	Germany	Listed (FSE)	€1,317M FY2025, +5.5% (cloud unit: €187M, public cloud growing ~48%) [3]	EU
Infomaniak	Switzerland	Employee/founder-owned; Swiss listing announced Jul 2026 [4]	CHF 54.2M 2025 (group)	CH
Exoscale	Switzerland	A1 Group (Austria)	Not disclosed	EU/CH
UpCloud	Finland	Private	Not disclosed	EU
Aruba	Italy	Private	Not disclosed	EU
StackIT	Germany	Schwarz Group (Lidl)	Not disclosed	EU
T-Systems (T Cloud Public)	Germany	Deutsche Telekom	Not disclosed	EU
Nebius	Netherlands	Listed (Nasdaq)	$529.8M FY2025, +351% (all cloud) [5]	EU HQ, US-listed, ex-Yandex
Evroc	Sweden	VC-backed	Pre-revenue (live July 2026)	EU

And the capability matrix, because "who sells what" is the table nobody maintains. Cells marked n/v were not verified for this edition.

Provider	Managed K8s	Managed DBs	S3-compatible storage	GPU offer	Egress included	Parent under US law
OVHcloud	Yes (free tier)	Yes	Yes	Yes	Yes (ex-APAC)	No
Scaleway	Yes (free tier)	Yes	Yes	Yes	Yes	No
Hetzner	No	No	Yes	Partial (dedicated GPU servers)	20 TB, then €1/TB	No
IONOS	Yes	Yes	Yes	n/v	2 TB, then €0.03/GB	No
Infomaniak	Yes (free tier) [58]	Yes [58]	Yes	n/v	10 TB (overage not published)	No
Exoscale	Yes (SKS)	Yes (Aiven-powered)	Yes (SOS)	Yes	Per-instance allowance	No
UpCloud	Yes	Yes	Yes	n/v	Pooled allowance	No
Nebius	Yes	Yes	Yes	Yes (core business)	n/v	No (Dutch parent; US-listed)
AWS	Yes (paid)	Yes	Yes	Yes	100 GB, then ~$92/TB	Yes
Azure	Yes (free tier; paid standard)	Yes	No (proprietary API)	Yes	100 GB, then per-GB	Yes
GCP	Yes (free zonal; paid standard)	Yes	Interop mode	Yes	Free tier, then per-GB	Yes

Control-plane and database prices live in the price audit below; this matrix only answers "does the product exist".

Three things jump out of these tables:

Only three of these companies publish audited cloud-company revenue: OVHcloud, IONOS, and Nebius (Infomaniak reports a group figure). Europe's cloud sector is mostly financially opaque. That means the headline 15% itself rests on Synergy modeling companies that have published nothing for years; the error bars are unknowable.
The two listed pure-players on completely different trajectories tell the whole story: OVHcloud at 9% growth, Nebius at 351%.
Nobody in the left column sells anything resembling the 200+ managed services of a hyperscaler. Bert Hubert's framing is the honest one: Europe sells lumber, hyperscalers sell furniture [6].

The price audit: the discount is narrowing, the traffic gap is not

The rules: list prices, no committed-use discounts, EU regions where available, August 2026 unless noted, ex-VAT, EUR for European providers and USD for US providers (treat sub-10% currency differences as noise). † marks prices read from a third-party comparison [7] rather than an official pricing page.

General-purpose VM, ~4 vCPU / 16 GB

Provider	Instance	vCPU type	vCPU / RAM	Monthly price	Region
IONOS	Memory Cube M	Shared	4 / 16 GB	€23 †	Germany
OVHcloud	b3-16	Shared	4 / 16 GB	~€75 †	France
Scaleway	PRO2-XS	Shared	4 / 16 GB	€81.90	Paris
Hetzner	CCX23	Dedicated	4 / 16 GB	€85.99	Germany
Exoscale	Extra-Large	Shared	4 / 16 GB	~€92 †	CH/EU
GCP	e2-standard-4	Shared (oversubscribed)	4 / 16 GB	~$108 †	Netherlands
AWS	m6i.xlarge	Non-oversubscribed (shared host)	4 / 16 GB	~$140-155 †	Frankfurt
Azure	D4s v5	Non-oversubscribed (shared host)	4 / 16 GB	~$140+ †	EU regions

Infomaniak publishes a close config rather than this exact shape: 4 vCPU / 12 GB at €24.92/month on annual billing [57], which would sit near IONOS at the cheap end (its 16 GB price exists only in an interactive configurator, so we do not print it).

"Shared" means the physical CPU cores are oversubscribed between tenants: fine for most workloads, but throughput can dip under neighbor load. "Dedicated" pins physical threads to your VM; the hyperscaler rows sit between the two (no oversubscription, no pinned cores). Note also that m6i is not AWS's current price-performance line: newer x86 and Graviton shapes price lower, so if anything this row flatters the European comparison. The columns exist so you can compare like with like.

Two winners, one per class: IONOS wins shared at €23 for an oversubscribed cube; Hetzner wins dedicated at €85.99 with pinned threads even after its increases. Ranking the two classes in one list would flatter shared instances, so we print one winner per class. Loser in both: Azure, at six times the IONOS price for the same shape.

Before quoting this table, know that the classic conclusion ("Europe is 3-5x cheaper") no longer holds everywhere. Hetzner's CCX23 cost €31.49 in May 2026. Today it costs €85.99.

The increases came in two waves [8]:

April: +30-43% on cloud servers, +30-53% on object storage.
Mid-June: up to 175% on the x86 cloud lines, both dedicated (CCX) and shared (CPX).

Hetzner blamed component costs, and the receipts back it up: DRAM spot prices rose roughly 171% year over year in the AI memory squeeze [9]. The cheap shared ARM instances survived with milder increases (CAX11: €4.49 to €5.99), so the entry-level story holds. But the "Hetzner is always 4x cheaper than AWS" reflex, built on a famous 2024 comparison [10], is now outdated for anything with x86 cores. And this is not one company's problem: Exoscale raised its list prices in May 2026 as well [11]. The AI boom is repricing European infrastructure from below, through the memory and GPU supply chain, faster than any competitor could from above.

Three caveats, stated plainly. First, real enterprises do not pay list: with three-year commitments and negotiated discounts, hyperscaler compute lands 40-60% below the numbers above, which compresses the VM gap to roughly 1.5-2x. Control-plane and per-service fees barely move under discounting; egress is negotiable at large scale (more below), but the meter itself never goes away.

Second, the AWS premium does buy real things: mature multi-region failover, deep compliance tooling, enterprise support with a named contact. European providers have documented reliability failures: OVHcloud's 2021 Strasbourg fire destroyed a datacenter along with backups some customers believed were elsewhere [12], and Hetzner has a documented pattern of terminating accounts through automated abuse-detection with minimal explanation. If you have never priced what those risks cost you, the hyperscaler premium is not obviously irrational.

Third, and this audit does not price it: people. Replacing managed databases, queues, identity and observability with self-managed equivalents costs platform engineers, and ten extra SRE salaries erase a lot of egress savings. What that gap really costs, and when a managed-ops partner closes it cheaper than either extreme, is a modeled-TCO exercise we are keeping for a dedicated follow-up. Until then, treat every "pocket the savings" in this report as gross, not net.

The European price advantage that did survive 2026 untouched is elsewhere.

Egress: the exit tax

Provider	Included traffic	Overage per TB
AWS	100 GB/month	~$92 first tier, tapering to ~$50 at scale [13]
Hetzner	20 TB per server	€1.00
Scaleway	Included	€0
OVHcloud	Included (ex-APAC)	€0
Infomaniak	10 TB	Not published
IONOS	2 TB	~€30 (€0.03/GB)

Azure and GCP price egress on comparable per-GB tiers to AWS; the shape is the same. "Included" on the European side means capacity-managed, not infinite: OVHcloud and Scaleway apply fair-use policies, and sustained multi-gigabit flows prompt a commercial conversation. The honest contrast is a soft cap against a hard meter.

Past the free allowance, moving one terabyte out of AWS at list costs about ninety times more than moving it out of Hetzner. From the buyer's side it is a tax on leaving, and it compounds: it applies to every running workload, not just the migration, and it constrains architecture directly, because at ~$92 per TB any multi-cloud or high-egress design is priced out before the first line of Terraform. Worked example: a company serving 500 TB a month pays AWS roughly $30,000 at list for that traffic (tiered), versus about €480 of overage on top of a €86 server at Hetzner. Large accounts negotiate the rate down hard (Direct Connect runs near $0.02/GB, CDN deals lower still, which can turn that $30,000 into something nearer $10,000); the meter and its logic remain. The Data Act outlaws switching charges, egress included, from 12 January 2027 [14], but the ban covers leaving, not operating, so the tax stays on every running workload.

Egress is not the only exit tax. Microsoft's licensing makes its own software more expensive to run on rival clouds than on Azure. CISPE's antitrust complaint over it was settled in 2024 for about $21.7M plus concessions that pointedly excluded AWS and Google [15]; Google filed its own complaint [16]; a £1 billion UK class action is pending [17]. Cloud, meanwhile, still sits largely outside the DMA's designated services. The Data Act freed your data; your licenses stayed put.

The commit: the entry tax

Egress is the visible lock-in. The invisible one is the committed-spend agreement. AWS EDPs and Microsoft MACCs bind three to five years of budget with growth clauses, and marketplace software purchases burn down the commit, so even third-party tools deepen the tie. By the time anyone compares VM prices, the money is already allocated, and the real decision window is not chosen by the buyer at all: it is the renewal date of the agreement. The Data Act bans exit fees; it does not touch commits. Any European provider strategy that ignores the renewal calendar is shouting at closed doors.

Object storage and Kubernetes: primitives cheap, orchestration free

Provider	Object storage per TB/month	Managed K8s control plane
AWS	~$23.55 (S3 Standard)	$0.10/hour, ~$73/month, SLA-backed (EKS) [18]
Scaleway	€16.06 Multi-AZ / €8.03 One Zone	Free (mutualized, no SLA, up to 150 nodes); SLA-backed dedicated tiers from €0.11/hour [59]
Infomaniak	~€10 †	Free control plane, SLA up to 99.9% [58]
OVHcloud	n/v (egress-free since Dec 2025)	Free tier (99.5% SLO); Standard 3-AZ tier with 99.99% SLA at $0.099/hour [65]
Azure / GCP	Comparable to S3	Free tiers exist (no SLA); standard tiers ~$0.10/hour (not re-verified this edition)

Both sides now sell both free and paid control planes; the difference is the default. At the hyperscalers the paid tier is the norm and the free tier is the exception; at the European providers the free mutualized plane covers most real estates and the paid tiers buy an SLA. The broader pattern holds across the catalog: European providers compete on the primitives and give away the orchestration; hyperscalers monetize every layer.

GPUs, where the table flips

Provider	H100, on-demand per GPU-hour	Source
Scaleway	€2.87 (H100 PCIe)	Official pricing page [60]
OVHcloud	€3.10 (h100-1-gpu, FR, ex-VAT)	Official pricing page [61]
Verda (Finland, ex-DataCrunch)	$3.25 (H100 SXM5)	Official pricing page [62]
Lambda (US, for reference)	$3.29 (H100 PCIe)	Official pricing page [64]
Nebius	$3.85 (HGX H100)	Official pricing page [63]
AWS (US reference)	~$6.88 (P5, per GPU) †	Third-party read of list price [19]

Caveats first. These are on-demand list prices, snapshotted August 2026 from official pages except where marked †; serious training buyers use reserved capacity at a fraction of them. Form factor matters: PCIe cards (Scaleway's €2.87, Lambda's $3.29) suit single-GPU and inference work, while multi-node training needs SXM/HGX systems with NVLink, where the spread narrows but holds (Verda $3.25 and Nebius $3.85 against AWS ~$6.88). And by August 2026 the H100 is the commodity tier, not the frontier (Mistral is buying GB300s). With all that said: the cheap end of the table is European in both form factors, Scaleway on PCIe and Verda on SXM.

On AI compute, the newest market, the cheapest H100 rates accessible from Europe are European: Scaleway, OVHcloud and Verda all sell the GPU hour at less than half AWS's list price. There was no legacy market share to defend, no twenty-year-old enterprise agreements. Where the race started fresh, Europe is competitive on day one.

The gap ledger: the capex line ends most arguments

Gap	Europe	US hyperscalers	Ratio
Share of the European market	~15% (all European providers combined)	~70% (AWS+Azure+GCP)	~1:5
Largest single player's share	SAP and Deutsche Telekom, ~2% each	Big three combined ~70%	~1:10 per company
2026 capex	OVHcloud: €361M/yr; all Europeans: low single-digit billions	~$630-725B big-tech AI capex [20]	~1:100+
Revenue, largest pure-player	OVHcloud, €1.08B/yr	AWS, ~$120B+/yr	~1:100
Managed services catalog	Dozens	200+ each	~1:10
Net profit of Europe's flagship	€0.4M (OVHcloud, first ever)	AWS operating income ~$40B+/yr	not a ratio, a category difference
Migration funding and co-sell channel	No structured programs; partner base is mostly hosting resellers	MAP-class funding covers 15-25% of first-year spend; 100,000+ incentivized partners each	a different sales model

A note on that 2% row. Europe's two largest cloud vendors by Synergy's count, SAP and Deutsche Telekom, barely appear in this report's market map, and that is the point. SAP's cloud is software running largely on hyperscaler infrastructure, and its sovereign play, Delos, is a Microsoft stack operated by Arvato, reportedly up to 20% more expensive than plain Azure [21]. Europe's leaders at the infrastructure layer are, on inspection, distribution and packaging of American platforms.

Big tech's AI capex in 2026, even discounting the share that is not cloud at all, will outspend the combined annual revenue of every European cloud provider roughly fifty times over. Whatever European strategy exists cannot be "outbuild them". It has to be asymmetric: jurisdiction, price on primitives, egress, GPUs, and the public sector's wallet.

The frozen share also has a quieter arithmetic behind it: expansion, not acquisition. Hyperscalers grow inside existing accounts, with net revenue retention above 110% as each account adopts new managed services. European providers sell primitives with little to expand into, so they grow only as fast as their customers do. Flat share in a compounding market is what a retention gap looks like.

Demand is not the problem: 52.7% of EU enterprises bought paid cloud services in 2025, up 7.4 points in two years [22]. The country spread behind that average returns below, as one of the four levers that could actually move the 15%.

What changed in 2025-26: sovereignty became a product

Microsoft admitted it under oath

"No, I cannot guarantee that." That is Anton Carniaux, Microsoft France's director of public and legal affairs, under oath before a French Senate inquiry in June 2025, asked whether he could guarantee that French citizens' data would never be transmitted to US authorities without French consent [23]. He added that it had never happened, and that if compelled by a valid US order, Microsoft hands over the data.

Nothing in that answer was legally new. The CLOUD Act has attached to US companies, wherever their servers sit, since 2018, and every lawyer in the room knew it. What changed is that the caveat moved from footnotes in compliance PDFs to sworn testimony, quotable in every procurement meeting in Europe. The EU Data Boundary, Microsoft's program guaranteeing that European customer data is stored and processed inside the EU, was completed in February 2025 [24]. It guarantees where data is stored. It does not, and cannot, change who the company answers to.

Three facts keep the risk honest. First, the base rate: provider transparency reports show enterprise-content disclosures from EU datacenters at or near zero over years, and Carniaux said as much ("it has never happened"); FISA gag orders make the public numbers a floor rather than a ceiling, but the floor is low. Second, the nexus is often yours, not your provider's: a group with US subsidiaries or a US listing can be reached through its own corporate structure, whatever flag the datacenter flies. Third, European providers answer to European access regimes (French intelligence law, Germany's G10): "European provider" means a different government's access, not no government's. The test later in this report stands precisely because of this: what you can actually buy is not immunity, it is reversibility.

The abstract risk also acquired a face this year. Karim Khan is the chief prosecutor of the International Criminal Court, the body that issues arrest warrants for sitting heads of state; when US sanctions hit him in early 2025 over the Netanyahu warrant, he lost access to his Microsoft email and moved to Proton Mail [25]. Microsoft denies it blocked the account and says the ICC moved the mailbox itself; the factual record is genuinely contested. Either version is instructive: whether the provider pulled the plug or the customer fled preemptively, a US sanctions decision ended a European institution's use of a US cloud service within weeks. The ICC has since moved to OpenDesk, the German open-source suite.

The hyperscalers moved inside the sovereignty tent

The strategic response was not to fight the sovereignty narrative but to sell it. AWS launched its European Sovereign Cloud in January 2026: a structurally separate cloud, first region in Brandenburg, EU-resident staff, its own EU-incorporated parent entity ultimately owned by Amazon, €7.8 billion committed to Germany through 2040 [26].

Microsoft shipped a stack of "sovereign solutions" and reported its European capacity up 40% across 16 countries [27]. Its French joint venture Bleu (with Orange and Capgemini) is still awaiting its SecNumCloud qualification.

Google's play runs through S3NS, its joint venture with Thales. In December 2025, S3NS became the first provider to hold France's strictest certification, SecNumCloud, across IaaS, CaaS and PaaS simultaneously [28]. About nine providers, OVHcloud included, hold it for narrower scopes.

Stop on that last point: the most sovereignty-certified cloud platform in France runs Google technology. By this report's own framework, that is a defeat for the original project. SecNumCloud existed to answer the jurisdiction question, and jurisdiction follows the parent, not the keys; S3NS is French-controlled operations wrapped around a product roadmap set in Mountain View. It is also a real improvement in operational guarantees, and procurement officials quoting the certification should understand they are buying the operations, not the independence.

The test: can you leave, can you keep running when the provider's home government changes its mind, can you replace the provider without rebuilding your stack. Three noes means you bought residency, not sovereignty, whatever the label on the region says. By that test, "sovereign regions" are resilience products, and the only sovereignty actually for sale anywhere is reversibility: open interfaces, portable workloads, and a rehearsed exit.

Gartner puts European sovereign-cloud spending at $6.9 billion in 2025, tripling by 2027 [29]. The label sells; the open question is who qualifies to capture that spend, and right now the hyperscalers are winning the qualification: CISPE, the cloud trade body whose members include AWS alongside European hosts, spent the year warning that the Commission's opaque sovereignty scoring could let hyperscaler "sovereign" offers pass as European. Its word for it: "sovereignty washing" [30].

The sharpest critique of the year came from inside the movement. Cristina Caffarra, the economist behind the EuroStack initiative, wrote in late July 2026 that regulation-as-sovereignty has not changed hyperscaler conduct or market structure, called the mega-deals "horrifying", and argued for a private-capital buy-fund-build strategy instead of Brussels processes [31].

Brussels tried again

In mid-2026 the Commission proposed the Cloud and AI Development Act: datacenter capacity targets and, for the first time, sovereignty requirements for sensitive public-sector workloads [32]. It is the third attempt at organizing European supply. The first, Gaia-X, promised production services by 2021, shipped none, and survives as a standards body [33]. The second, EUCS, the EU cloud certification scheme, is unadopted six years in because France will not accept a top tier without immunity-from-US-law criteria [34]; a January 2026 Cybersecurity Act 2.0 proposal reworks the certification machinery and may eventually unblock it [35]. The new act starts with both failure modes unfixed: no shipped product (the Gaia-X path) and a certification blocked on the US-immunity fight (the EUCS path). Whether it clears either is checkable by 2027.

The public sector started moving, but at the SaaS layer

The migrations that actually completed in 2025-26 share a pattern almost nobody predicted: they are Microsoft 365 exits, not AWS exits.

Schleswig-Holstein moved about 80% of its ~30,000 government workstations to LibreOffice, claiming €15M+ in annual savings, with the Linux rollout still ongoing [36].
Denmark's Ministry of Digitalisation, then Copenhagen and Aarhus, began phasing out Microsoft, explicitly citing geopolitics. Aarhus cut one department's costs from 800,000 to 225,000 DKK on a Nextcloud stack [37].
Lyon announced in June 2025 it is dropping Microsoft for OnlyOffice, Linux and PostgreSQL [38].
The European Commission awarded its €180M sovereign cloud tender in April 2026 to four European-led lots: Post Luxembourg with OVHcloud and Clever Cloud, StackIT, Scaleway, and a Proximus-led consortium that includes S3NS and Mistral [39].

The logic is sound even if the narrative missed it: the political exposure lives in email, documents and identity, not in virtual machines. Roughly 80% of EU corporate software-and-cloud spend goes to US vendors per a European Parliament briefing, most of it at that layer [40], and an office suite has drop-in substitutes while a DynamoDB estate has a migration project.

The one big IaaS-layer move is French: after six years of litigation, the national Health Data Hub is leaving Azure for Scaleway, with migration due by early 2027 [41]. It is the reference case: if Scaleway carries it, the "Europeans cannot run critical workloads" claim loses its best evidence; if it slips badly, every RFP for years will cite it.

The counter-story keeps everyone honest. The Netherlands' own domain registry SIDN tried to move .nl zone administration to AWS. A unanimous parliament halted it. The compromise still sent part of the workload to AWS Frankfurt, because the alternatives could not match the operational bar [42]. And the European Commission itself, after being found in breach over Microsoft 365 by its own data protection supervisor, fixed the contract and stayed on Microsoft [43]. Sovereignty rhetoric is free; migrations are paid for in ops teams.

Bert Hubert calls this the demand problem: the constraint is buyer behavior, not missing supply [6]. His proposed fix is regulators, not subsidies: "Regulators can likely force banks to do the right thing, and thus forge a path for governments to follow." The 2025-26 record supports him; the levers that moved anything were procurement rules and DORA, not new datacenters.

The AI layer is a different race

Everything above describes a mature market where the positions froze years ago. AI infrastructure is not that market.

Nebius grew revenue 351% to $529.8M in 2025, guides to $3.0-3.4 billion for 2026 [5], and in September 2025 signed a contract with Microsoft worth $17.4 billion through 2031, expandable to $19.4 billion [44]. On current guidance it overtakes OVHcloud in revenue during 2026, mostly serving American demand from New Jersey. Every taxonomy of "European cloud" chokes on this company, which is exactly why it belongs in the table.

Mistral raised $830M in debt to fill a datacenter near Paris with 13,800 Nvidia GB300 systems, targeting 200MW of its own European compute by end-2027, with a 1.4GW campus planned behind it [45]. It is not alone:

DeepL skipped the cloud entirely and runs its own DGX SuperPODs in Sweden [46].
Fluidstack raised an $830M Series A (coincidentally the same figure as Mistral's debt round) and signed an MoU for a 1GW machine in France [47].
The EU opened its AI gigafactories call in July 2026: €10 billion public money, up to 7 facilities, targeting €30 billion with private co-investment [48].
The public track is not purely paper: JUPITER in Jülich, Europe's first exascale system at €500M, came online in September 2025 and ranks fourth on the TOP500 [49].

The pattern across all of it: when the workload is new, European players compete at the global frontier, immediately. The lesson cuts against the dominant policy instinct, which keeps trying to subsidize a European AWS fifteen years too late instead of funding the layers where the race is still open.

One asterisk the size of Nvidia itself: every GPU in this section is Nvidia's. Mistral's sovereignty, Nebius's margins, the gigafactories' schedules, all of it depends on allocation decisions made by one US company. Europe did not escape dependency in the AI build-out; it moved the dependency down one layer, from the cloud API to the chip.

The other physical constraint is the grid. Datacenter connection queues in Frankfurt, Amsterdam and Dublin now run five years and more, and the hyperscalers locked long power purchase agreements early: capacity sellable in 2028 was contracted in 2024. The next edition of this report will add a contracted-power column to the market map; it may predict share better than the price tables do.

Meanwhile, consolidation quietly accelerated

More than 30 European hosting and cloud acquisitions closed in the first half of 2026 alone. Kyndryl is buying Solvinity. team.blue was underwritten at €4.8 billion [50]. And in the most symbolically loaded deal of the year, Canada's Cohere announced in April it is acquiring Aleph Alpha, the company that spent three years as Germany's sovereign-AI flagship, with the Schwarz Group putting $600M into Cohere instead of into a German champion [51]. Europe's answer to concentration risk is, so far, more concentration, under other people's flags.

Reality vs. hype

Five widely repeated claims, checked.

"Sovereign regions solve the jurisdiction problem." Mostly hype. A sovereign region operated by an EU entity with EU staff genuinely raises the practical bar for foreign access, and dismissing that as marketing is lazy. But the ultimate parent remains subject to US law, and no corporate structure yet tested in court removes CLOUD Act reach over an American parent. The honest label is "resilient residency". Buy sovereign regions for the operational guarantees, not the flag on the box.

"The Data Act frees you from lock-in." Partial. It abolished the exit tariff, which is real money: AWS waived around $250k of egress for 37signals' celebrated cloud exit alone [52]. But nobody's workload is trapped by egress fees; it is trapped by IAM policies, proprietary managed databases, and the three engineers who understand the Terraform. The 37signals exit, $10M in projected savings and a deleted AWS account, remains the most quoted repatriation story in Europe, and it is American. There is still no audited European equivalent at that scale, which after two years of sovereignty discourse is itself a data point.

"Europe is always the cheap option." No longer safe to assume: two rounds of Hetzner increases and an Exoscale repricing in one spring, all traced to the AI memory squeeze. The advantages that survive are egress, free control planes, and simpler pricing. The raw compute discount is narrowing.

"European clouds cannot run serious workloads." Disproven at the low end and unproven at the high end, which is not the same as false. The Commission's €180M tender, the Health Data Hub migration, and Nebius's Microsoft contract are serious workloads by any definition. What remains genuinely unproven is a European provider running a hyperscaler-grade, multi-region, 200-managed-services enterprise estate, because no European provider sells one. The furniture gap is real. It is also, on the evidence of the GPU market, a product decision rather than a law of nature.

"Procurement is decided by sovereignty frameworks." Mostly hype. Shortlists are copied from analyst quadrants where no European provider appears as a Leader, and budgets sit inside hyperscaler commit agreements signed years earlier. SEAL scores the finalists; the commit and the quadrant chose them. Fixing analyst presence and the renewal calendar is cheaper than fixing capex.

What would actually move the 15%?

The honest answer: only four levers are big enough, and each comes with a catch.

The public wallet. The $6.9 billion of sovereign-cloud spending Gartner counts for 2025, tripling toward 2027 [29], is worth several points of share if European providers capture it. The catch: they will not capture all of it, because "sovereign" now includes AWS's Brandenburg cloud and S3NS running Google technology. The sovereignty label was supposed to be the moat around European providers; the hyperscalers walked over it. The share impact depends entirely on how procurement frameworks like SEAL, the sovereignty-scoring framework the Commission published in October 2025, score jurisdiction. That is precisely the fight CISPE is having with the Commission.

The AI numerator. European AI-cloud revenue is real money against a €75 billion market, but most of the fastest-growing part of it is earned serving American customers from American soil, and that will not move the European market share number. Mistral Compute and the gigafactories, if they fill with European demand, are the version of this lever that would actually register.

The second Europe. 17.8% of Bulgarian enterprises buy cloud against 79.2% in Finland [22]. The next decade of European cloud demand comes disproportionately from countries and companies with no incumbent hyperscaler relationship to defend. Greenfield is the one segment where distribution, price and jurisdiction can beat a 200-service catalog. The catch: nobody is visibly organizing to win it.

The channel. Cloud in Europe is sold by integrators and MSPs compensated through hyperscaler co-sell programs, and by hiring markets full of AWS and Azure certifications; the 15% is a distribution outcome as much as a product outcome. A European co-sell fund plus a serious certification track would cost a rounding error of one gigafactory and would touch the share faster than any datacenter subsidy. The catch: no European provider has built either.

Absent those four, the arithmetic says the first expectation below is safe: procurement documents change faster than market structure.

What we expect by 2027

Falsifiable and dated; we will grade ourselves publicly in next year's edition. For calibration, the analyst consensus is cautious to the point of emptiness: Forrester's 2026 call was that no European enterprise will shift entirely off US hyperscalers this year, a bar set so low it cannot miss [53].

European providers' share of the European market will still be between 14% and 17% in Synergy's mid-2027 reading. The sovereignty wave changes procurement documents before it changes market share. [80%]
At least two more EU member states announce a contracted migration of a named national system (registry, health, tax, or identity scale) to a provider in this report's market map. Announcements without a signed contract do not count. [70%]
The Health Data Hub migration to Scaleway slips past its early-2027 target. Flagship migrations always slip. [60%]
At least one AI gigafactory contract is awarded by mid-2027, and zero gigafactories are operational by August 2027. [85%]
Hetzner does not revert its June 2026 increases, and no provider in this report's market map undercuts Hetzner's new CCX pricing by more than 20% at equivalent dedicated-vCPU specs. Memory costs are everyone's costs. [70%]

Our recommendations

These are the calls we make with clients today, by situation, including the cases where the right call is to stay put.

Your situation	Our recommendation	First action this quarter
Startup, no compliance constraints	European primitives (Scaleway, Hetzner, OVHcloud); pocket the gross savings. The DX gap at this level has narrowed to tolerable	Price your current stack on two European providers, stress-tested against a Hetzner-scale repricing; decide in one meeting
Regulated enterprise (bank, insurer, health)	Split the estate: sensitive and continuity-critical systems on SecNumCloud-grade or European infrastructure where the provider's resilience evidence supports it; commodity stays put	List the workloads whose seizure or suspension would stop the business; those are the ones DORA will eventually force you to justify. Check your commit renewal date: that is your real decision window
Public administration	Start where migrations actually succeed: the office and collaboration layer. Schleswig-Holstein published the playbook	Pilot one department on LibreOffice/Nextcloud-class tooling before touching any IaaS
Company deep in proprietary managed services	Stay. A migration you cannot operate is more dangerous than a subpoena you will probably never receive	Write the substitution plan per proprietary service; do not execute it out of sentiment
AI company	Train wherever the GPUs are cheap and available; today that is often a European provider	Requote your training and inference on two GPU clouds outside your current provider

And for everyone, the exit checklist. From January 2027 leaving is legally free; make it architecturally cheap too:

Storage S3-compatible, not proprietary-API (compatibility is a spectrum: object lock, event notifications and consistency semantics differ between implementations, so test the features you actually use).
Workloads Kubernetes-shaped, or at least container-shaped.
A written substitution plan for every proprietary managed service you adopt.
Your monthly egress volume known, and priced at two other providers.
The exit rehearsed once, even partially, on a non-critical workload.

Do all five and switching providers becomes a project you can scope and price, not a rewrite. The checklist is necessary, not sufficient: IAM, data gravity and delivery pipelines still take real engineering. And keep Bert Hubert's caveat in mind: being locked into a European cloud behemoth is not the goal either. Independence from any single vendor is.

How we can help

This report is written by Dixence. We operate cloud infrastructure for European companies that do not have, or do not want, a dedicated ops team: Kubernetes, deployments, and SLA-backed operations on European providers, built on the reversibility principles above (the exit checklist is how we design, not just what we recommend).

If you want to know where your own infrastructure actually stands on jurisdiction, cost, and exit options, we run a free 30-minute Infrastructure Checkup: one call, no slides, and you leave with a short written summary of the three things to fix first. Book it at dixence.dev.

Closing: the anger is real, the 15% is realer

The summary of 2026 fits in two sentences. Europe finally got angry about cloud dependency, and the market share number did not move. Both halves matter. The anger produced sworn testimony, regulation with enforcement dates (Data Act 2027, DORA), a €180M tender won by European-led lots, and the first flagship national workload heading to a European provider. One of those winning lots includes S3NS, the Thales-Google venture; nothing in this story is pure. The frozen 15% proves that none of it has yet touched the layer where the money is, and the hyperscalers' response, selling sovereignty themselves, is working.

The question for the next two years is not whether OVHcloud beats AWS. It will not. The open questions: does the Health Data Hub migration land, does the gigafactory money become racked machines, does the "compete globally on the new workload" model get copied, and does the 15% move once the egress ban and DORA enforcement hit in 2027. This report will be back next year, same tables, same rubric, with this year's expectations graded in public.

Methodology

Desk research conducted July-August 2026 across primary sources: company filings and results releases, official pricing pages, regulator and Commission publications, and analyst estimates. Every load-bearing claim carries a numbered reference below.
Prices are list prices snapshotted August 2026, ex-VAT, EUR for European providers and USD for US providers. † marks a price read from a third-party comparison rather than an official pricing page; n/v marks a cell not verified this edition. No committed-use or negotiated discounts are reflected unless stated.
Market share and market size figures are analyst estimates (Synergy Research unless noted), built partly on modeling of private companies that publish no financials. Treat them as directionally solid, not precise.
The draft went through several adversarial review rounds before publication: an independent fact-check of every figure against the source corpus, an editorial pass against the best "State of" reports, and domain-expert challenges (startup reader, enterprise CTO, European provider executive). Corrections are welcome and will be listed in a changelog.
Known gaps, planned for the next edition: a modeled TCO including platform headcount, a provider resilience audit (AZ topology, SLA terms, incident history), and contracted-power data per provider.

References

Synergy Research Group, "European cloud providers' local market share holds steady at 15%", July 2025: srgresearch.com/articles/european-cloud-providers-local-market-share-now-holds-steady-at-15
OVHcloud, FY2025 results, October 2025: corporate.ovhcloud.com/en/newsroom/news/financial-results-fy25/
IONOS, FY2025 results (EQS release), 2026: tradingview.com/news/eqs:36e1dd274094b:0-ionos-reports-successful-2025-fiscal-year-profitability-at-record-levels/
Infomaniak, Swiss listing announcement, July 2026: news.infomaniak.com/en/infomaniak-announces-its-planned-listing-on-the-swiss-stock-exchange/
Futurum, Nebius Q1 FY2026 earnings, 2026: futurumgroup.com/insights/nebius-q1-fy-2026-earnings-show-ai-cloud-capacity-scaling/
Bert Hubert, "The European Cloud Situation at the end of 2025", December 2025: berthub.eu/articles/posts/the-european-cloud-2025/
From Europe With Love, "European cloud providers vs AWS, 2026 comparison", March 2026: fromeuropewithlove.eu/en/blog/european-cloud-providers-vs-aws-comparison-2026
Hetzner, price adjustment documentation, June 2026: docs.hetzner.com/general/infrastructure-and-availability/price-adjustment/
Northflank, "Why Hetzner raised prices", 2026: northflank.com/blog/hetzner-cloud-server-price-increases
Hacker News, "AWS and Azure are at least 4x-10x more expensive than Hetzner", October 2024: news.ycombinator.com/item?id=41902103
EU Cloud Cost, "EU cloud price update", May 2026: eucloudcost.com/blog/eu-cloud-price-update-may-2026/
The Register, "OVH Strasbourg fire", March 2021: theregister.com/2021/03/10/ovh_strasbourg_fire/
AWS, EC2 on-demand pricing (data transfer): aws.amazon.com/ec2/pricing/on-demand/
Kemp IT Law, "The end of switching charges", 2026: kempitlaw.com/insights/the-end-of-switching-charges-commercial-impact-and-compliance-priorities/
TechCrunch, "Microsoft settles with European cloud trade body", July 2024: techcrunch.com/2024/07/10/microsoft-settles-with-european-cloud-trade-body-over-antitrust-complaints
TechCrunch, "Google files antitrust complaint against Microsoft in Europe", September 2024: techcrunch.com/2024/09/25/google-files-antitrust-complaint-against-microsoft-in-europe-over-cloud-licensing-practices
WinBuzzer, "Microsoft faces £1B UK lawsuit over cloud practices", December 2024: winbuzzer.com/2024/12/03/microsoft-faces-1b-uk-lawsuit-over-anticompetitive-cloud-practices-xcxwbn/
AWS, EKS pricing: aws.amazon.com/eks/pricing/
GetDeploying, Nvidia H100 price index, August 2026: getdeploying.com/gpus/nvidia-h100
Futurum, "AI capex 2026: the $690B infrastructure sprint", 2026: futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/
Julien Simon, "The EU Cloud and AI Development Act (CADA)", 2026: julsimon.medium.com/the-eu-cloud-and-ai-development-act-cada-a-last-shot-at-cloud-sovereignty-or-another-expensive-74ccc1a4007c
Eurostat, "Cloud computing services used by 52.7% of enterprises", February 2026: ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20260203-1
The Register, "Microsoft admits it cannot guarantee data sovereignty", July 2025: theregister.com/2025/07/25/microsoft_admits_it_cannot_guarantee/
Microsoft, "EU Data Boundary complete", February 2025: blogs.microsoft.com/on-the-issues/2025/02/26/microsoft-completes-landmark-eu-data-boundary-offering-enhanced-data-residency-and-transparency/
heise, "ICC: Microsoft's email block a wake-up call for digital sovereignty", May 2025: heise.de/en/news/Criminal-Court-Microsoft-s-email-block-a-wake-up-call-for-digital-sovereignty-10387383.html
AWS, "AWS launches the European Sovereign Cloud", January 2026: press.aboutamazon.com/aws/2026/1/aws-launches-aws-european-sovereign-cloud-and-announces-expansion-across-europe
Microsoft, "One year on: progress on our European digital commitments", April 2026: blogs.microsoft.com/on-the-issues/2026/04/29/one-year-on-progress-on-our-european-digital-commitments/
Thales, "S3NS announces SecNumCloud qualification", December 2025: thalesgroup.com/en/news-centre/press-releases/s3ns-announces-secnumcloud-qualification-premi3ns-its-trusted-cloud
Data Center Dynamics, "Europe's sovereign cloud spending to triple 2025-2027 (Gartner)", 2025: datacenterdynamics.com/en/news/europe-spending-on-sovereign-cloud-infrastructure-to-triple-from-2025-2027-gartner/
The Register, "CISPE: sovereignty washing", March 2026: theregister.com/2026/03/18/cispe_sovereignty_washing/
Cristina Caffarra, "The 'sovereignty' narratives that don't help Europe", TechPolicy.Press, July 2026: techpolicy.press/the-sovereignty-narratives-that-dont-help-europe/
Inside Global Tech, "The EU Cloud and AI Development Act, in depth", June 2026: insideglobaltech.com/2026/06/11/the-eu-cloud-and-ai-development-act-in-depth/
EuroStack Directory, "Gaia-X: chronicle of a failure foretold", February 2025: euro-stack.com/blog/2025/2/gaia-x-failure
cep, "EU cloud certification at an impasse", April 2025: cep.eu/fileadmin/user_upload/cep.eu/cepItalia/cepInput_EU_Cloud_Certification_at_an_Impasse.pdf
Netguardia, "Cybersecurity Act 2.0 explained", 2026: netguardia.com/cybersecurity-intelligence/regulatory-updates/cybersecurity-act-2-0-explained-the-january-2026-commission-proposal-and-what-changes-for-enisa/
The Document Foundation, "Schleswig-Holstein moving to LibreOffice", March 2025: blog.documentfoundation.org/blog/2025/03/13/updates-on-schleswig-holstein-moving-to-libreoffice/
Euronews, "Two Danish city governments move away from Microsoft", June 2025: euronews.com/next/2025/06/12/two-city-governments-in-denmark-are-moving-away-from-microsoft-amid-trump-and-us-big-tech-
The Register, "Lyon leaving Microsoft", June 2025: theregister.com/2025/06/26/lyon_leaving_microsoft/
European Commission, "Commission advances cloud sovereignty through strategic procurement", April 2026: commission.europa.eu/news-and-media/news/commission-advances-cloud-sovereignty-through-strategic-procurement-2026-04-17_en
European Parliament, research briefing on EU software and cloud spend, 2025: europarl.europa.eu/RegData/etudes/ATAG/2025/780413/ECTI_ATA(2025)780413_EN.pdf
Euronews, "France moves public health data from Microsoft to French cloud provider", April 2026: euronews.com/health/2026/04/24/france-moves-public-health-data-from-microsoft-to-french-cloud-provider
TechPolicy.Press, "The dangers of moving key internet governance functions to Amazon's cloud: the case of the Netherlands", 2025: techpolicy.press/the-dangers-of-moving-key-internet-governance-functions-to-amazons-cloud-the-case-of-the-netherlands/
EDPS, "European Commission brings use of Microsoft 365 into compliance", July 2025: edps.europa.eu/press-publications/press-news/press-releases/2025/european-commission-brings-use-microsoft-365-compliance-data-protection-rules-eu-institutions-and-bodies_en
Data Center Dynamics, "Microsoft to use Nebius GPU data centers in deal worth $17.4bn", September 2025: datacenterdynamics.com/en/news/microsoft-to-use-nebius-gpu-data-centers-in-deal-worth-174bn-over-five-years/
Euronews, "Mistral raises $830M for datacenter near Paris", March 2026: euronews.com/next/2026/03/30/europe-needs-ai-cloud-infrastructure-mistral-raises-830m-for-data-centre-near-paris
PR Newswire, "DeepL first to deploy Nvidia DGX SuperPOD with DGX GB200 in Europe", June 2025: prnewswire.com (DeepL DGX GB200 announcement)
SDxCentral, "Fluidstack signs MoU for 1GW AI supercomputer in France", 2026: sdxcentral.com/news/gpu-cloud-provider-fluidstack-signs-mou-for-1gw-ai-supercomputer-in-france/
EuroHPC JU, "AI gigafactories call", July 2026: eurohpc-ju.europa.eu/eurohpc-joint-undertaking-launches-ai-gigafactories-call-2026-07-30_en
EuroHPC JU, "JUPITER: launching Europe's exascale era", September 2025: eurohpc-ju.europa.eu/jupiter-launching-europes-exascale-era-2025-09-05_en
Webhosting.today, "Thirty deals in six months: the H1 2026 hosting consolidation map", July 2026: webhosting.today/2026/07/28/thirty-deals-in-six-months-the-h1-2026-hosting-consolidation-map-and-the-two-clocks-driving-what-sells-next/
CNBC, "Cohere to acquire Aleph Alpha", April 2026: cnbc.com/2026/04/24/cohere-aleph-alpha-germany-ai-europe-expansion.html
The Register, "37signals cloud repatriation savings", May 2025: theregister.com/2025/05/09/37signals_cloud_repatriation_storage_savings
Forrester, "European predictions 2026": forrester.com/press-newsroom/forrester-european-2026-predictions
Synergy Research Group, "European cloud providers struggle to reverse market share losses", January 2021: srgresearch.com/articles/european-cloud-providers-struggle-reverse-market-share-losses
Synergy Research Group, "European cloud providers continue to grow but still lose market share", September 2022: srgresearch.com/articles/european-cloud-providers-continue-to-grow-but-still-lose-market-share
IDC, "EMEA public cloud services market to reach $415 billion by 2028", November 2024: my.idc.com/getdoc.jsp?containerId=prMETA52761324
Infomaniak, VPS Cloud pricing, August 2026: infomaniak.com/en/hosting/vps-cloud/prices
Infomaniak, managed Kubernetes and DBaaS launch, January 2026: news.infomaniak.com/en/launch-service-kubernetes-manage/
Scaleway, containers pricing (Kapsule and Kosmos), August 2026: scaleway.com/en/pricing/containers/
Scaleway, GPU pricing, August 2026: scaleway.com/en/pricing/gpu/
OVHcloud, Public Cloud pricing (GPU), August 2026: ovhcloud.com/fr/public-cloud/prices/
Verda (ex-DataCrunch), GPU pricing, August 2026: verda.com/pricing
Nebius, GPU pricing, August 2026: nebius.com/prices
Lambda, GPU cloud pricing, August 2026: lambda.ai/service/gpu-cloud
OVHcloud, Managed Kubernetes Service tiers, August 2026: ovhcloud.com/en/public-cloud/kubernetes/