European champions: building demand across borders

Two announcements on a new scale

On 31 August, the European joint undertaking EuroHPC and the LUMI AI Factory consortium — Finland, the Czech Republic, Denmark, Estonia, Norway and Poland — selected Bull to design and deliver LUMI-AI, a supercomputer dedicated to artificial intelligence. The value: €387.8 million. It is the sixth system commissioned under Europe’s ‘AI Factories’ programme, and the largest contract Bull has ever signed. It comes five months after the company returned to public ownership: on 31 March 2026, the French state acquired its entire share capital from Atos. The machine will be assembled in Angers, Europe’s only supercomputer factory, for delivery in 2027. In the meantime, Bull had already secured the Mimer AI factory in Sweden and delivered a machine to Airbus in Germany.

On 8 September, Mistral announced a €3 billion Series D led by Samsung Electronics, alongside the Scaleup Europe Fund — a European vehicle managed by EQT — and PSG Equity. The company’s valuation rises from €11.7 billion to more than €21 billion. Its stated priorities are research, model training and, above all, computing power, with a target of one gigawatt by 2030. Mistral says it has more than 125 customers in around twenty countries.

From a distance, the case seems proven: Europe knows how to produce champions.
Seen through the discussion on 30 June, the picture is far less clear-cut.

The supply has been there for some time; demand is what is missing

This was the position put forward by Anne-Marie Calmeil, President of T-Systems France.
Her diagnosis was not about gaps in the catalogue: European solutions exist, and many of their building blocks are mature. What is missing is demand that truly gets moving. In her view, public spending remains the main lever to activate in the short term. Public procurement generates revenue, funds R&D, builds references and makes private purchasing possible. She supported her argument with two elements: the Draghi report and the European text adopted in June 2026. Two further accelerators should encourage this investment: artificial intelligence, which is reshaping the landscape faster than any regulation, and the Franco-German partnership, without which European scale remains theoretical. The LUMI-AI contract is an almost perfect illustration of this mechanism: a European public order coordinated between EuroHPC and six member states. Mistral illustrates the other route: private capital. Effective, fast — and international.

Buying European does not necessarily mean buying sovereign

A closed solution, locked into a single supplier, even a European one, offers only a limited gain: the applicable law. The flag changes; the dependency remains.
Applying this test to the two recent announcements is instructive. Samsung is taking a significant equity stake in Mistral. The company has signed a multibillion-dollar contract with Microsoft to use its computing power in Europe. Its data centres run on American chips. LUMI-AI, meanwhile, will rely on AMD accelerators, and Bull, wholly owned by the French state, remains dependent on American semiconductors. This shows that sovereignty must be measured across a chain, rather than by a flag. The useful question is not ‘Is it European?’ but ‘What do I control, and what can I easily move away from?’

A champion, or a common foundation?

Amandine Le Pape, co-founder of Matrix and Element, stressed the need to build a common foundation. Her argument: for critical building blocks such as messaging and collaboration, a credible alternative means building an open, standardised foundation that each player can then specialise for its market, rather than selecting a single European champion in each category.
The distinction is not rhetorical. A closed champion becomes another point of dependency — just geographically closer. What provides lasting protection is reversibility and interoperability: the ability to change direction without rebuilding everything. An ecosystem of interoperable champions cannot be managed in the same way as a national champion. The two perspectives are compatible. We need players capable of reaching critical mass — Bull has just demonstrated this through a contract, and Mistral is securing the means to do so — alongside standards that prevent that scale from becoming a lock-in mechanism.

Dependency is systemic, not isolated

Nouamane Cherkaoui, author of ‘Sous l’emprise du code’ and Deputy Chief Executive Officer of BPCE SI, offered an uncompromising user perspective: a cloud hosted in France is not enough if SaaS software, hardware and energy remain beyond your control. Dependency must be measured across the whole chain, rather than through a single component.
Two dimensions received particular attention:

• The physical reality of digital technology. Data centres, energy and hardware: the cloud has an address, consumption and a factory. The session gave a name to the associated risk — ‘technological surrogacy syndrome’, in which infrastructure is built on European soil but the added value is captured elsewhere.
• Training data. Boris Lecoeur, Chief Executive Officer of Qwant and Shadow, pointed out that the fresh, real-time data needed by all models is currently supplied by only a handful of American players. A European model trained on foreign-controlled data is only halfway sovereign.

In this respect, LUMI-AI ticks several boxes that are often overlooked: assembly in Europe, direct liquid cooling developed by Bull, renewable electricity and the recovery of waste heat for Kajaani’s district heating network.

The real test will be the order book

The roundtable’s central theme could be summed up in one sentence: European sovereignty is a prerequisite for French sovereignty. On its own, a country such as France carries little weight against the major technology powers. At continental scale, the equation changes — provided orders follow.
That is precisely the challenge of the next twelve months. Mistral will have to turn €3 billion into lasting contracts, with one milestone already identified: the €1 billion in annual recurring revenue targeted for the end of 2026. Bull will have to deliver in 2027 and turn a record contract into a regular stream of business rather than a one-off achievement. And European organisations — public first, then private — will have to do what no one can do for them: buy. Champions do not create demand; they reveal it. A record funding round and a record contract do not mean the battle is won; they show that there is now something to buy. What remains is the least spectacular part of the work, and the only part that really counts: placing orders, demanding reversibility and checking what you actually control, building block by building block.