en.Wedoany.com Reported - French cloud service provider OVHcloud plans to enter cutting-edge AI model development from the infrastructure layer, aiming to build Europe's own AI systems as an alternative to US and Chinese models. The company's CEO, Octave Klaba, told Reuters that OVHcloud will train a series of models from scratch and release them as open source once predetermined performance targets are met.
This plan will put OVHcloud in direct competition with Paris-based model developer Mistral AI, which was previously seen as Europe's main force challenging US AI labs. Klaba stated that improvements in chips, training methods, and synthetic data have significantly reduced model development costs. A project that originally required approximately $1.15 billion (€1 billion) is now expected to be controlled within $230 million (€200 million).
Reuters reported that OVHcloud has confirmed one of the models has completed pre-training on the EuroHPC supercomputer Jupiter in Germany. The supercomputer is described as Europe's fastest and the first exascale system, but OVHcloud has not disclosed specific performance benchmarks. This move comes amid growing concerns among European governments and enterprises about data governance and access continuity for AI infrastructure. These concerns were further heightened this month when Anthropic stated that US export control directives had suspended providing certain model access to personnel outside the US.
Neil Shah, Vice President and Partner of Research at Counterpoint Research, noted that the $230 million cost estimate mainly covers the initial training run. After training is complete, models still require ongoing investment to maintain value, or they will become depreciating assets. OVHcloud also needs to invest in fine-tuning, post-training, sovereign infrastructure, storage, security, distribution, and enterprise support, and must achieve sufficient scale to compete on pricing with established AI providers like Google and Anthropic.
Charlie Dai, Principal Analyst at Forrester, believes that efficiency improvements have lowered the barrier to entry, and this budget range is sufficient to develop credible cutting-edge models. However, enterprise competitiveness depends on ongoing capabilities beyond training, such as inference efficiency, data pipelines, evaluation frameworks, and ecosystem coverage.
Sanchit Vir Gogia, Chief Analyst at Greyhound Research, emphasized that OVHcloud's plan is currently only an expression of intent rather than a demonstrated capability. He noted that the pre-training was completed on Jupiter, a supercomputer located in Germany, which is a public European supercomputer running on US chips, reflecting the partial nature of European AI sovereignty. Gogia stated that enterprise customers need evidence that models can be supported in production environments, effectively governed, auditable when necessary, and exited without disruption. While a European-owned model can reduce dependence on US and Chinese providers, it cannot eliminate jurisdictional risks. He further pointed out that OVHcloud's entry into model development may change the lock-in risks enterprises need to assess: customers may be able to migrate cloud infrastructure in the future, but AI workloads built around models and governance tools will be more difficult to migrate.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
