Five of the largest companies in Europe have recently acquired computing resources that Mistral has not developed.

      Mistral outlined its plan on 11 August, which encompasses three key aspects. Regional endpoints are now largely available, allowing customers to choose between Europe or the US for inference. A Priority Tier is currently in public preview, introducing custom rate limits and a commitment to uptime. The third aspect is the funding mechanism for the other two.

      The European Compute Unit is essentially a forward sale. Enterprises will invest money upfront, which will translate into multi-year access to computing resources that Mistral will develop in the future. Mistral designates these as European Compute Units, which customers can use across its computing products once capacity is made available.

      The agreement terms are long, with partners committing for approximately five years without the option for early termination, as stated by chief technology officer Timothée Lacroix in an interview with European Business Magazine. This arrangement allows Mistral to gauge demand that it can finance against, while the buyer assumes delivery risk for infrastructure that is primarily still in the design phase.

      According to Sifted, five companies compose what Mistral refers to as the anchor group: ASML from the Netherlands, Spain's Amadeus, France's Capgemini and Caisse des Dépôts, along with the shipping group CMA CGM. Each company is represented by its CEO in the announcement: Christophe Fouquet for ASML, Luis Maroto for Amadeus, Aiman Ezzat for Capgemini, Olivier Sichel for Caisse des Dépôts, and Rodolphe Saadé for CMA CGM. Fouquet emphasized the support for scalability, stating that Mistral “is taking on that challenge with the scale, ambition, and staying power.”

      Caisse des Dépôts merits special mention as a French state financial institution, making it one of the five anchor customers and effectively representing the state itself.

      ASML occupies a unique position in this scenario, not just as a customer but also as a leader in Mistral’s €1.7 billion funding round in September 2025, which valued the company at €11.7 billion. With this commitment, ASML is also set to purchase computing resources enabled by this financing round. Both pieces of information come directly from the companies involved and are not concealed. This relationship alters how the demand signal is interpreted; a large order from an investor offers different insights into the broader market compared to an order from an unrelated entity.

      Three weeks earlier, Microsoft made a distinct commitment, which funds Nvidia's Vera Rubin chips for Mistral’s European operations, intentionally without acquiring an equity stake.

      The aim is to achieve up to 1GW by 2030, with a target of approximately 200MW by the end of 2027. Arthur Mensch has previously estimated that a gigawatt of computing requires around $50 billion in investment.

      The initial facility is smaller than the headline figures suggest, at 44MW, located south of Paris, and secured by an $830 million loan. The difference between the current 44MW and the ultimate goal of 1,000MW comprises the scope of the project. Everything acquired by the anchor group will unfold beyond this gap.

      Mistral has additional projects underway, according to DatacenterDynamics. A 40MW GPU cluster is operating in the Paris area at an Eclairion facility hosted by Scaleway. They are also considering a larger campus through a joint venture involving Mistral, Bpifrance, the UAE fund MGX, and Nvidia, discussing the development of a 1.4GW site in the same region with a possible commencement in 2028.

      Mistral will also host third-party open models on the same infrastructure, adhering to the same regional regulations. The first such model will be GLM-5.2 from the Beijing company Z.ai, also known as Zhipu. This model features a one-million-token context window and costs $1.40 per million input tokens. Z.ai has released GLM-5.2 with open weights and no usage restrictions.

      Mistral's argument is that the infrastructure layer is more critical than any individual model, as expressed by The New Stack. Customers have a single platform to run open models without needing to restart each time they switch.

      Matan Grinberg, CEO of Factory, shared the customer perspective, stating that Mistral “enables us to run open models under strict regional controls and service commitments.” Here, “sovereignty” refers to jurisdiction over where the weights are processed, which differs from the origin of those weights.

      This perspective leads into an ongoing debate in Europe that has lasted for two years. Mistral has warned that American providers could potentially disconnect European customers, a situation that has been referred to as the sovereignty moment. The counterargument concerns the foundational layer, with some suggesting that renting GPU capacity creates a false sense of sovereignty while the chips remain American, as noted in contributions from earlier this year