2CRSi lands a €60M+ order for 83 NVIDIA B300 servers from Exaion — sovereign inference, delivered by December

2CRSi said on November 3, 2026, that it has received a binding order worth more than €60 million from Exaion for 83 NVIDIA HGX B300 AI servers, along with associated storage and maintenance services. The order builds on a strategic partnership between the two companies that has been in place since 2021, and it gives 2CRSi a defined, dated delivery program: every one of the 83 servers is due before the year ends.

The size of the order matters because it is binding, not a memorandum or a letter of intent. 2CRSi describes the deal as a binding order from an existing partner, for named hardware, with a delivery schedule attached. A summary of the same release confirms the same figures: more than €60 million, 83 servers, delivery by the end of 2026.

The delivery schedule

The 83 servers will be deployed by the end of 2026 in three tranches: the first 17 are to be delivered this week, 33 by the end of November, and the final 33 before the end of December. The arithmetic is the point — 17 plus 33 plus 33 is 83 — and the pace is aggressive for latest-generation AI hardware, with more than 80 servers to land in under two months.

Alain Wilmouth, chairman and chief executive of 2CRSi, said the order reflects renewed confidence from customers and suppliers in the company’s solutions, and that delivering more than 80 servers by the end of the year demonstrates an ability to rapidly deliver latest-generation AI infrastructure designed and assembled in France. That framing positions speed and local assembly as part of the offer, not just the hardware itself.

From training to inference

For Exaion, the servers are a step change in what its infrastructure is for. The company currently operates four data centers in France and Canada and aims to reach 300 MW of capacity, with 20 MW expected to be operational in 2027. The new order, 2CRSi said, will support Exaion as it expands its offering from model training to private on-premises inference, supporting data control and digital sovereignty.

That distinction is the commercial logic of the deal. Training builds a model; inference runs it, repeatedly, on live data. Moving inference on-premises, under the customer’s own control, is the pitch for organizations that want the benefits of AI without sending strategic data to a shared cloud. Fatih Balyeli, chairman and chief executive of Exaion, said private on-premises inference is a key priority for companies that want to take full advantage of artificial intelligence while keeping control of their strategic data, and that digital sovereignty depends on deploying high-performing, secure, and fully controlled computing infrastructure.

A partnership since 2021

The two companies have been building their collaboration since 2021, combining complementary expertise. 2CRSi contributes server design and manufacturing, including immersion cooling technologies. Exaion brings expertise in digital infrastructure, cloud, and advanced computing services. The announcement presents the new order as an extension of that work, aimed at high-performance computing infrastructure that combines computing power, energy efficiency, and a low environmental footprint.

The companies’ backgrounds explain the fit. Exaion was co-founded in 2020 by Fatih Balyeli and Laurent Bernou-Mazars, as a spin-off of EDF Group’s EDF Pulse Incubation entrepreneurial program, and its Montreal-based subsidiary, Exaion Inc., serves North American customers. 2CRSi was founded in 2005 in Strasbourg, designs and manufactures high-performance servers for AI, high-performance computing, and data storage, and was listed on Euronext Paris in June 2018 before transferring to Euronext Growth in November 2022.

What the announcement does not say

The release is precise about the order and the schedule, and silent on much else. It gives no per-server price, no revenue-recognition timing, and no performance figures for the servers in Exaion’s data centers. It names no end customers who will run workloads on the new hardware, and the “over €60 million” figure is a floor, not an exact contract value. Delivery is a deployment schedule, not proof that the servers are yet operating at full capacity.

What it does establish is concrete: a named buyer, a named product, a binding order above a stated value, and a dated delivery plan that ends in December 2026. For a European AI infrastructure market where sovereignty claims often arrive without hardware attached, 83 servers with a delivery date is the substance. Whether Exaion’s inference business grows into the 300 MW it aims for will be decided by the customers those servers serve — and by whether the delivery schedule holds.