OVHcloud published its 2026 financial results on Tuesday, October 20, a year after founder Octave Klaba returned to run the company. Revenue rose 7.2% to €1.15 billion, adjusted EBITDA reached €480 million (42% of revenue, up on the year) and free cash flow turned positive. In an interview with La Tribune, Klaba says OVHcloud is about to launch its own AI model: "We are going to bring to market an LLM very ahead of its time."
The year in numbers
Klaba came back after the removal of Benjamin Revcolevschi, who had succeeded Michel Paulin, promising an internal overhaul to restore strong growth, lift a battered share price and put the Roubaix-based provider in the AI race. His plan is called "Step ahead 2026-2030."
- Growth: fourth-quarter revenue rose 10.5% at constant scope, and monthly revenue crossed €100 million by the end of the fiscal year.
- Profit: the €480 million of adjusted EBITDA works out to about 42% of the €1.15 billion in revenue.
- Klaba's verdict: "Twelve months after my return, we have reached all the objectives we announced," he says, adding that he is "ready to accelerate."
The pace is still a different league from the American hyperscalers, whose latest quarters showed growth of 37% for Amazon Web Services, 42% for Microsoft Azure and 82% for Google Cloud. OVHcloud's 7.2% is a fraction of that, which is the gap an AI strategy is meant to close.
The LLM
Klaba is positioning OVHcloud at the center of the AI shift by training models of its own rather than only hosting others'. According to Reuters coverage of the announcement at VivaTech, OVHcloud plans a family of frontier models, aiming to become Europe's second LLM player behind Mistral. Klaba put the cost of a project that once required about €1 billion at €150 to 200 million today, crediting better chips, training techniques and synthetic data. Pre-training of a first model was reported finished on the Jupiter supercomputer with DragonLLM, a recently acquired startup. OVHcloud says it will not train on customer data, and intends to open-source the models once they are good enough, with no date set.
The interview also covers Klaba's views on Mistral, technological sovereignty and the global AI race. It fits the European debate this blog has followed: France's economy minister saying "If European AI Is Just Mistral, We're Screwed", and Mistral's own pivot to being a hosting layer as well as a model lab, alongside its Large 4 launch.