Europe is finally paying for its own AI compute, and Denmark has put 750 million kroner on the table
Brussels opened bidding for up to seven AI gigafactories, Denmark committed money to buy capacity in one, and Mistral closed the largest funding round in European tech history. None of it changes where your AI runs this year, but it changes the options.
On July 30 the European Commission opened the call for up to seven AI gigafactories, large data centers built to train the next generation of models. On August 3 Denmark said it would help pay for one. On September 8 the French lab Mistral announced it had raised €3 billion, partly to build compute of its own. European AI policy has mostly been about regulation. These three decisions put public and private money into European computing capacity.
Seven gigafactories, €30 billion, running in 2028
The gigafactory call aims at around €30 billion of total investment. The EU budget is meant to cover about €5 billion, member states another €5 billion, and private investors the remaining €20 billion. Only €1 billion of the EU share is committed so far. Construction is planned to start in early 2027, with the facilities running by mid-2028, and each one has to become financially self-sustaining by selling commercial services. Denmark, Finland, Germany, France, Italy, Poland, Spain and several others have signalled interest.
The Danish government committed to buying up to DKK 750 million of computing capacity over five years and is interested in a mid-sized facility with about 14 times the power of Gefion, the Danish AI supercomputer that went live in 2024. Minister Christina Egelund said Europe is far too dependent on countries outside the EU. The commitment follows the June government platform, which promised more supercomputer capacity funded together with the foundations.
Mistral sells sovereignty
Mistral’s Series D raised €3 billion at a €21 billion valuation. Samsung led, EQT’s Scaleup Europe Fund and PSG Equity co-led, and Nvidia, a16z, BlackRock and the state of Luxembourg were among the other investors. Mistral calls it the largest equity round ever raised by a European technology company. The plan is a gigawatt of European compute by 2030 and expansion beyond the 20 countries it operates in today.
Mistral now sells control to governments and companies that want to decide where their AI runs and rely less on US providers. That includes hosting other labs’ open-weight models, Chinese ones among them, so customers can choose the region where their queries are processed. A customer can use Mistral’s infrastructure without using Mistral’s models.
The sums are small next to US spending
The four largest American cloud companies plan to spend about $725 billion on capital expenditure this year alone. Europe’s gigafactory programme is €30 billion over several years, two thirds of it still to be found from private investors, and the facilities will not run before 2028. Mistral’s gigawatt target for 2030 is comparable to a single large US data-center campus.
Europe will not match American capacity with this. The realistic aim is to have some capacity of its own when the next Fable 5-style export directive or trade dispute hits, and to keep European researchers and companies from renting everything abroad. That is a sensible goal, and a more modest one than the press releases suggest.
What it means for a Danish business now
For the next two years, the frontier models most companies use will keep running in American-owned data centers, many of them physically located in Europe. The gigafactories do not change that before 2028.
What has changed is the choice for workloads where location matters. Providers like Mistral now offer capable models, open-weight ones included, processed in EU regions under European ownership. If you handle health data, public-sector data or anything a customer’s procurement team asks hard questions about, that option exists today. Price it against the US providers instead of assuming it is worse.
Danish companies with heavy compute needs, particularly in life sciences, should follow how the Danish capacity purchase gets allocated. Gefion has already been opened to companies as well as researchers, and a facility fourteen times larger will need users.
And ask your AI vendors where your queries are processed, who owns the machines, and what happens to your service if a government on either side of the Atlantic changes the rules. After this summer, the answers belong in your business continuity plan as well as your compliance file.
