EU AI Gigafactories vs. Deploying on Scaleway or OVHcloud Today: Which One Is Actually Realistic?

Key points:

• No, a growing company cannot realistically apply for an EU AI Gigafactory. The call for tenders accepts consortia or special purpose vehicles that will build and operate facilities with tens of thousands of top-end AI chips. A 20-200 person SaaS or AI company is a future customer of that capacity, not a credible applicant.
• The 12 November deadline has moved. On 5 November 2026 EuroHPC extended Lot 1 to 4 February 2027, kept Lot 2 at 3 December 2026, and asked every bidder to resubmit on new links (EuroHPC JU). Awards are expected in early 2027, with operations up to 18 months after contract signature.
• The real SME door is the EuroHPC AI Factories. Playground access is granted within 2 working days, Fast Lane within 4, and both are free for AI SMEs and startups (EuroHPC JU). The EIC Accelerator covers cash.
• For production, deploy on a European commercial cloud this week. Scaleway and OVHcloud give you EU-jurisdiction GPUs and managed Kubernetes with no application process. Expect a narrower service catalogue than AWS, GCP or Azure.
• Sovereignty comes down to architecture. Containers, Kubernetes and Terraform or OpenTofu turn "where do we run" into a configuration choice. That keeps you free to move between Scaleway, OVHcloud and the hyperscalers today, and onto public EU capacity later.

Each AI Gigafactory is meant to hold around 100,000 latest-generation AI chips, about four times the AI Factories being built now (European Commission). That scale tells you who the programme is for. I have spent 15+ years in cloud infrastructure and interviewed more than 200 CTOs. Every few years Europe announces a big compute programme, and every time founders ask me whether they should apply. For this one my answer is no. Applying is unrealistic for a growing company. The EuroHPC AI Factory access calls are the real SME door, and a European commercial cloud is where production should run today.

What exactly is the EU AI Gigafactories programme, and what stage is it at now?

The AI Gigafactories programme is a European Commission and EuroHPC Joint Undertaking procurement. Through it, the EU and Member States buy compute access time from up to seven industry-built AI data centres. As of 9 November 2026, the formal call for tenders is open. The Lot 2 deadline is 3 December 2026 and the Lot 1 deadline is 4 February 2027 (EuroHPC JU).

Here is how the money breaks down. In February 2025, InvestAI set out to mobilise €200 billion for AI, including a €20 billion fund for gigafactories (European Commission). The tender launched in July 2026 is smaller and more concrete. It offers up to €10 billion in EU and national funding, which the Commission expects to bring in at least €20 billion of private investment (European Commission). Add the two and you get the "over €30 billion" figure, though only a third of it is public money.

The structure shows who it targets. Lot 1 funds up to four projects, each with up to €100 million of EU money in phase one and up to €400 million more in phase two. Lot 2 funds up to three projects at €200 million plus up to €800 million. Participating Member States match the EU share (European Commission). Most of the public money comes from the next EU budget, 2028-2035 (EuroHPC JU).

The 2025 call for expression of interest was a market-sounding exercise and awarded nothing. It drew 76 expressions of interest for 60 sites in 16 Member States, planning at least three million GPUs in total (European Commission). The Commission did not publish a combined investment figure. The respondents were data centre operators, telecoms, power suppliers and financial investors. SaaS companies were not among them.

The timeline is the key fact for a CTO. Award decisions are expected by early 2027 and construction should start in 2027. Selected sites then have up to 18 months from contract signature to begin operations (European Commission). The formal call was first planned for Q4 2025 and opened on 30 July 2026, and Lot 1 has already slipped once. So I would not plan on Gigafactory capacity before 2028.

Can a growing European company realistically apply for AI Gigafactory funding?

No, unless you are a data centre, energy, chip or financing player joining a consortium. The call accepts "Consortia or Special Purpose Vehicles bringing together companies, public entities, investors and other partners" that will build and operate the facility (European Commission). A scaling software company is the intended customer of this capacity.

The power math alone settles it. One NVIDIA H100 SXM draws up to 700W (NVIDIA). A site of 100,000 chips therefore needs about 70 MW for the GPUs alone, before you add CPUs, networking and cooling. The call also requires bidders to show "adequate power capacity infrastructure" and water-efficiency practices (EuroHPC JU).

Writing a serious bid takes months of founder, legal and partner-search time, and you end up with zero running infrastructure. The one credible role for a smaller company is a supporting one: anchor tenant, workload partner, or a European hardware or software supplier. The Commission says part of hardware procurement "may be dedicated to European start-ups and scale ups" (European Commission). That opportunity is real if you sell chips, cooling or a scheduler. If your training queue is blocked this quarter, it does nothing for you.

What EU funding and compute routes are actually open to startups and SMEs right now?

Three EU routes accept single companies today and decide within days or months: the EuroHPC AI Factories access modes, the EuroHPC supercomputer access calls, and the EIC Accelerator. None of them asks you to co-finance a data centre.

• AI Factories, Industrial Innovation track. There are 19 AI Factories plus 13 AI Factory Antennas, which extend AI Factory services to further countries (European Commission). Playground access runs first-come-first-served, with access in 2 working days. Fast Lane offers up to 50,000 GPU hours for up to 3 months, with access in 4 working days. Large Scale covers more than 50,000 GPU hours, with approval within 10 working days of a cut-off and two cut-offs a month. All three are free for AI SMEs and startups.
• EuroHPC supercomputer access calls. Development and Benchmark access are continuously open, give resources within two weeks, and allocate "a small number of node hours" (EuroHPC JU). The AI and Data-Intensive Applications call explicitly targets SMEs and startups, with one-year allocations starting within a month of a cut-off.
• EIC Accelerator. It offers grants up to €2.5 million and investment from €1 million to €10 million. The 2027 budget is €428 million for Accelerator Open and €150 million for Challenges (EIC). Feedback on a short proposal takes roughly 4-6 weeks, followed by a full proposal and a jury interview.

These allocations go through a batch scheduler. They suit training runs, fine-tuning and benchmarks, and they do not suit an always-on API with an SLA. Demand is also high. MareNostrum 5 and LUMI-G were closed to new Large Scale allocations when I checked (EuroHPC JU).

Should you just deploy on Scaleway or OVHcloud today?

Yes, for nearly every growing European company. A European provider gives you EU-jurisdiction GPUs and managed Kubernetes this week, which beats waiting for capacity that has not been built. The trade-off is a narrower managed-service catalogue and tighter large-scale GPU supply than AWS, GCP or Azure.

Scaleway, part of France's Iliad group, runs regions in Paris, Amsterdam and Warsaw. Kubernetes Kapsule includes a free shared control plane (Scaleway). In fr-par-2 the GPU line-up runs from L4 and L40S through H100 PCIe and SXM to B300. List prices are about €0.79/hr for L4-1-24G, €1.47/hr for L40S-1-48G and €2.87/hr for H100-1-80G, before tax (Scaleway).

OVHcloud is a listed French company. Its Managed Kubernetes Service has a free control-plane tier and a paid Standard tier spread across three availability zones (OVHcloud). Its SecNumCloud-qualified scope covers the SNC Cloud Platform, Managed VMware vSphere and Bare Metal Pod, hosted in three regions with dedicated SNC zones (OVHcloud). Outbound public traffic is included with instances everywhere except its Asia-Pacific regions (OVHcloud).

Other credible options include Exoscale (Swiss, simple managed Kubernetes), IONOS Cloud (German, strong in public sector), Hetzner (very cheap bare metal and VMs, little managed tooling), UpCloud (Finnish, fast block storage) and StackIT (Schwarz Group, aimed at German enterprises).

The weaknesses are real. European providers hold about 15% of their home market, down from 29% in 2017, against 70% for Amazon, Microsoft and Google (Synergy Research Group). In practice that means fewer managed services, fewer regions outside Europe, thinner Terraform modules and fewer third-party integrations. Multi-node H100 clusters may also need a sales conversation instead of a console click.

The hyperscalers have an answer too. AWS launched its European Sovereign Cloud on 15 January 2026 in Brandenburg, with more than €7.8 billion committed (AWS). Microsoft completed its EU Data Boundary in February 2025 (Microsoft), and Google sells sovereign controls. All three are still US-parented, so legal teams keep asking about the US CLOUD Act.

Many teams land on a hybrid: an EU provider for regulated or data-resident workloads, a hyperscaler for the rest, and one deployment workflow across both.

What does "sovereign" actually mean for your architecture, and how do you keep the option open?

Sovereignty depends on your architecture and your contracts more than on the provider name on your invoice. The move that lasts is to keep workloads portable, so that changing where you run means changing configuration instead of doing a six-month rewrite.

Separate three things people conflate. Data residency is where the bytes sit. Operational sovereignty is who can touch the control plane. Legal jurisdiction is which courts can compel access. A hyperscaler EU region gives you the first, a sovereign offering adds part of the second, and an EU-owned provider covers all three.

A portability checklist you can apply this week:

• Containerise every workload, and deploy with Kubernetes-native manifests or Helm.
• Keep infrastructure in Terraform or OpenTofu, with one module per provider behind the same interface.
• Put object storage behind an S3-compatible API. Scaleway, OVHcloud and the hyperscalers all speak it.
• Keep provider-only serverless and proprietary queues out of the critical path, or wrap them behind your own interface.

Lock-in costs real money when you leave. AWS charges around $0.09/GB in the first egress tier after 100 GB free (AWS), and Azure charges $0.087/GB in Europe (Azure). The EU Data Act limited switching charges to cost-based fees from 12 September 2025 and removes them entirely from 12 January 2027 (European Commission).

Regulation forces the residency decision earlier than any funding timeline will. GPAI obligations under the AI Act have applied since 2 August 2025, and the Act became generally applicable on 2 August 2026. After the AI Omnibus, high-risk rules in areas such as employment and critical infrastructure apply from 2 December 2027 (European Commission). DORA has applied since 17 January 2025 (EUR-Lex), and NIS2 had to be transposed by 17 October 2024 (EUR-Lex).

How do you run Kubernetes on a European cloud without hiring a platform team?

Qovery deploys and operates your applications inside your own cloud account: AWS, GCP, Azure, Scaleway, or an existing Kubernetes cluster, including on-prem or an EU provider's managed cluster. You get Kubernetes-grade deployments without a dedicated DevOps engineer, and the contract, the bill and the jurisdiction stay in your name.

This is BYOC, bring your own cloud. Qovery never resells compute, so your committed-use discounts and your data residency stay intact. Scaleway is directly supported alongside AWS, GCP and Azure. OVHcloud, Exoscale, IONOS and other EU providers are covered through bring-your-own-Kubernetes on their managed clusters.

What you get: git-push deployments, a preview environment per pull request, auto-stop rules for non-production, managed cluster upgrades, per-environment RBAC, and databases backed by managed cloud services. Your workflow stays the same whichever provider or region you deploy to, so moving between them is a configuration change.

Compare that with hiring. The median DevOps engineer earns €76,748 in total compensation in Germany and €51,411 in France (Levels.fyi, Levels.fyi). Those figures exclude employer contributions, and on-call realistically takes two people.

Be fair to the alternatives. If you already have three platform engineers, Terraform plus ArgoCD is a great choice. Coolify or Dokku is enough for a single VM. Northflank and Porter are good picks if you will never leave a hyperscaler.

How do I deploy to a European cloud with Qovery from Claude Code or Cursor?

Install the Qovery Agent Skills once, then ask your coding agent in plain language to deploy to your own Scaleway, AWS, GCP or Azure account, or to an existing cluster. The agent proposes each step and waits for your approval, and the MCP Server stays read-only until you enable write access.

Install globally for Claude Code, Cursor, Codex, OpenCode and Gemini CLI:

curl -fsSL https://skill.qovery.com/install.sh | bash

Or scope the install to one project:

curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project

Add the MCP Server to Claude Code (OAuth, read-only by default):

claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242

Enable write access only when you want it. It also has to be enabled in the Qovery Console:

claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242

The skills qovery-onboard, qovery-deploy, qovery-troubleshoot, qovery-optimize, qovery-speedup, qovery-preview and qovery-terraform do the deploying, and the MCP Server operates what already exists. Then ask: "deploy this app to my Scaleway account with Qovery" or "set up a preview environment on our existing Kubernetes cluster". Details are in the AI agent quickstart and the Claude Code integration.

Frequently asked questions

Is the EU AI Gigafactory tender open to startups and SMEs, or only to large consortia?

Only consortia or special purpose vehicles can bid, bringing together companies, public entities and investors that will build and operate the facility. Startups and SMEs are the intended users of the finished capacity. A smaller company can join a consortium as a tenant or supplier, but it cannot realistically bid alone.

How much funding is behind the EU AI Gigafactories, and where does the money come from?

The tender offers up to €10 billion in EU and national funding and expects at least €20 billion of private investment, which is where the "over €30 billion" figure comes from. Member States match the EU share, and most of the public money comes from the 2028-2035 EU budget. The €20 billion InvestAI gigafactory fund announced in February 2025 is the broader political envelope.

When will AI Gigafactory compute capacity actually be available to use?

Award decisions are expected by early 2027, construction should start in 2027, and sites have up to 18 months from contract signature to begin operations. The Lot 1 deadline was already extended to 4 February 2027. Realistically, plan on 2028 at the earliest.

How is an EU AI Gigafactory different from a EuroHPC AI Factory?

An AI Factory is built around AI-optimised EuroHPC supercomputers and gives SMEs free access in days through Playground, Fast Lane and Large Scale modes. A Gigafactory is a new, industry-built facility targeting around 100,000 chips, roughly four times an AI Factory. Today, 19 AI Factories and 13 Antennas exist, and no Gigafactory has been selected yet.

Are Scaleway and OVHcloud good enough for GPU and AI workloads today?

Yes, for inference, fine-tuning and mid-size training. Scaleway lists L4, L40S, H100 and B300 instances in Paris, and OVHcloud offers GPU instances, bare metal and a SecNumCloud-qualified scope. They fall short of the hyperscalers on managed AI services and on reserving very large multi-node clusters at short notice.

How do I get EU-sovereign compute without locking myself into one European provider?

Containerise your workloads, deploy on Kubernetes, keep infrastructure in Terraform or OpenTofu, and use S3-compatible storage. Then the provider becomes a configuration choice. A BYOC platform like Qovery keeps one deployment workflow across Scaleway, the hyperscalers and any existing Kubernetes cluster.