OVHcloud enters the model race: its first AI model lands early 2027, trained on Europe's Jupiter supercomputer

• Who: OVHcloud, the French cloud operator, via CEO Octave Klaba.
• What: the company expects to launch its first AI model in early 2027 — its own model, after a first pilot of its OVHai Workspace platform.
• Design: the model will use a mixture-of-experts architecture, trained on Jupiter, Europe’s fastest supercomputer.
• Capacity: 250MW connected, 140MW in operation, 110MW free to onboard new clients without new construction.
• Workspace: OVHai Workspace — a productivity platform tuned on OVH’s internal data running workplace apps like email — pilots shortly.
• When: announced October 20, 2026, in statements to Reuters following OVHcloud’s annual earnings report.

Europe's cloud giant wants to build its own brain

OVHcloud expects to launch its first AI model in early 2027, chief executive Octave Klaba said on Tuesday, in remarks to Reuters reported following the company's annual earnings report. The French cloud operator is best known for selling raw compute and hosting to European companies wary of the American hyperscalers — now it wants to compete where the value concentrates: the models themselves.

The plan is not starting from zero. Klaba had already told Reuters in June that OVH had begun training its own AI models on Jupiter, Europe's fastest supercomputer. What is new is the timeline and the commercial intent: a launch date of early 2027, and a positioning as a potential European challenger to Mistral, France's flagship foundation-model company.

A mixture-of-experts bet, not a me-too model

Klaba said the model will use a mixture-of-experts design — routing tasks to specialised sub-models rather than firing every parameter on every query, the architecture of choice when you want frontier-class quality without frontier-class inference bills. For a cloud operator, that design is also a commercial argument: the model is meant to run efficiently on OVH's own infrastructure, not on rented Nvidia capacity.

Before the model arrives, OVH will pilot OVHai Workspace, described as a productivity platform tuned on OVH's own internal data running workplace apps such as email. It is the pragmatic first step — an internal-data productivity layer that proves the stack works, and a product that gives enterprise customers a familiar on-ramp to the model when it ships.

Capacity to spare — and a swipe at Europe's AI policy

Klaba's other headline number is about capacity, not models. OVHcloud has 250 megawatts of datacenter capacity connected, of which 140MW is in operation, leaving 110MW ready to take on new business — about half its cloud available for new clients. The implication is aimed squarely at rivals: OVH can onboard AI demand without spending on construction, while competitors must invest to serve regional demand.

He used the earnings spotlight for a sharper point as well. Klaba reiterated his criticism of Europe's state-backed computing programmes, arguing that their financing — half EU, half member state — naturally pushes projects toward national initiatives. The result, he said, is "a race of national data centre projects, none backed by a customer base to match." Last month he called national strategies for the EU's planned AI gigafactories "doomed to failure," dismissing a proposed €200 million public commitment to buy AI services over five years as too small to underpin the private investment required.

Whether the market buys the full stack — capacity, workspace, model — remains to be seen. But the shape of the ambition is now explicit: a European cloud operator with sovereign infrastructure training sovereign models, and a CEO arguing that customer-backed capacity beats state-backed blueprints.