The EU Is Building an AI Supercomputer Network: How Growth Companies Get Access to LUMI-Class Computing Power

The EU currently runs 19 operational AI Factory hubs, one of them in Finland. The programme opens publicly funded supercomputing capacity to companies too — not just research institutions. Yet few growth companies know that an alternative to renting compute from the big cloud providers already exists.

In daily business, the problem shows up concretely: fine-tuning or training a model on a large dataset requires computing power a standard cloud account cannot deliver cost-effectively. The alternatives are vendor lock-in with a single cloud provider or investing in your own hardware — both expensive, slow paths for a growth company that should be focused on its business, not its infrastructure.

The EU's answer is the AI Factories initiative. It links the EuroHPC Joint Undertaking's supercomputers, data, and talent into one European network that is also open to SMEs and midcap companies. This article covers what AI Factories means in practice and when a growth company should actually consider training its own model on the public compute network.

What is the AI Factories programme?

AI Factories is a network built by the European Commission and the EuroHPC Joint Undertaking (EuroHPC JU) that opens up European supercomputing capacity for developing trustworthy AI. The programme stems from the Commission's January 2024 AI Innovation Package and was reinforced later by the AI Continent Action Plan (European Commission, digital-strategy.ec.europa.eu, 2026).

The network currently comprises 19 operational AI Factory hubs and 13 connected "Antenna" units across member states. EuroHPC JU selected the first seven consortia in December 2024, involving 15 member states, with hubs deployed in Finland, Germany, Greece, Italy, Luxembourg, Spain, and Sweden (European Commission, digital-strategy.ec.europa.eu, 2026). The programme has since expanded in several waves, most recently in October 2025.

Finland's hub is called the LUMI AI Factory, built around the LUMI supercomputer. The hardware sits in CSC's data centre in Kajaani, Finland, and is operated by the international LUMI consortium (LUMI-supercomputer.eu, 2026). LUMI is among Europe's most capable machines, and its capacity keeps expanding: the LUMI consortium announced in July 2026 that IQM had been selected to deliver the LUMI-IQ quantum computer as part of the setup.

Why does this matter to a growth company, not just researchers?

Compute capacity across the AI Factories network is open to European users across sectors — industry, research, academia, and public administration (European Commission, digital-strategy.ec.europa.eu, 2026). The programme specifically prioritises access for startups and SMEs, which makes it relevant for a Finnish growth company that has valuable data but no data centre of its own.

In practice, this opens three scenarios where a growth company should consider the public compute network instead of just scaling up its cloud bill. First: the company wants to fine-tune an existing language model on its own industry data without routing that data repeatedly through a foreign cloud provider. Second: the company needs a short burst of massive compute, for example to train a computer-vision model on a large image dataset. Third: the company wants to test whether building a specialised model is even worthwhile before committing to a major investment.

This does not mean every growth company should build its own model. Most businesses do not need their own foundation model — integrating existing, proven models at the Master Layer level is usually enough. AI Factories becomes relevant only once a company's data is specialised or valuable enough that a general-purpose model falls short, and full training or extensive fine-tuning becomes worthwhile.

When should a growth company actually consider its own model?

Training or extensively fine-tuning your own model is rarely the right first step. It becomes worthwhile based on three factors: the volume and quality of your data, how distinctive your competitive edge is, and whether the model delivers measurable business value repeatedly — not just as a one-off experiment.

If your company recognises itself in the right-hand column, the AI Factories network offers a path that doesn't require building a data centre. Access runs through EuroHPC JU's resource access calls; terms and application processes vary by hub, and are worth reviewing in advance together with a technology partner.

How does a growth company prepare to use AI Factories compute?

Preparation always starts with data, not compute. A public supercomputer doesn't fix the fact that your company's data is scattered across your CRM, ERP, and disconnected documents. The Master Layer connects these systems securely before the data is even ready for training — the same reason we previously covered why AI fails when company data is scattered.

The next step is figuring out whether you need your own model at all. Master Plan is an AI strategy sprint that maps where AI creates the most value for your business — measured in euros. Only after this mapping do you know whether the business value justifies the public supercomputer's application process, or whether integrating an existing model is enough.

The third step is execution through the sprint model: custom AI solutions are delivered in 3-day development cycles, which keeps the cost of even compute-heavy experiments predictable — the first sprint proves the concept before any larger commitment.

What does the LUMI AI Factory mean for a Finnish growth company?

For a Finnish growth company, the LUMI AI Factory means access to a European, EU-rules-compliant supercomputer without investing in your own hardware. The domestic location (Kajaani) and European governance model can be a deciding factor for companies whose data cannot leave the EU, or that want to avoid depending on a single US-based cloud provider.

Should a growth company wait or build its data architecture now?

Don't wait. The AI Factories network keeps expanding, but access to it doesn't solve the underlying problem of messy data. A growth company that gets its data integrations and governance in order now will be ready to use growing public compute capacity as the programme scales — instead of starting the data work only once the opportunity is already in front of it.

Frequently asked questions

Questions and answers about the AI Factories network and how growth companies can use it.

Can a Finnish growth company apply for access to the LUMI supercomputer?

Yes. EuroHPC JU's resource access calls are open to European users across sectors, and the programme specifically prioritises access for startups and SMEs. The application process and requirements vary by use case, so it's worth checking in advance whether your need actually fits this route.

Does a growth company need its own AI model?

Rarely as a first step. Most business value comes from integrating existing models with your own data through the Master Layer. Your own model or extensive fine-tuning becomes worthwhile only once your data is highly specialised and the business value is recurring, not one-off.

What does the AI Factories network cost a company?

The AI Factories infrastructure is publicly funded by the EU and member states as part of EuroHPC JU. A company's own costs come mainly from preparing its data, integration, and model development work — these follow the predictable cost structure of the sprint model.

The first step isn't applying for supercomputer access. It's figuring out where AI creates the most value in your business — and whether you even need your own model. Book a free Master Mind analysis, and we'll go through whether pursuing the public compute network makes sense for your case at all.