OVHcloud pushes into frontier models

France’s OVHcloud is moving from cloud infrastructure into frontier model development, aiming to train a family of models from scratch and open-source them once performance targets are met. CEO Octave Klaba said advances in chips, training methods, and synthetic data have lowered the cost of a project that may once have required about $1.15 billion (€1 billion) to less than $230 million (€200 million).

One OVHcloud model has completed pre-training on Jupiter, the Germany-based EuroHPC supercomputer described as Europe’s fastest and its first exascale system, but detailed benchmarks have not been released. Analysts said the lower training cost likely covers only the initial run, while long-term competitiveness will depend on fine-tuning, post-training, infrastructure, storage, security, distribution, enterprise support, and efficient model serving.

The push comes as European organizations weigh data governance and access continuity alongside model performance. Analysts said a European-owned model could reduce reliance on US and Chinese providers, but sovereignty questions remain because Jupiter uses American silicon and AI workloads can create new forms of lock-in once applications are built around a provider’s models and governance tools.