The artificial intelligence landscape is undergoing a seismic shift. While the US and China dominate the conversation with models like GPT-4 and DeepSeek, a European player traditionally known for its cloud infrastructure, OVHcloud, has taken a bold step: developing its own frontier AI models. This move aims not only to compete technically but also to offer a sovereign alternative for European companies wary of technological dependence on foreign powers.

From Cloud Provider to Model Creator

OVHcloud, one of Europe's largest cloud service providers, has announced plans to train from scratch a family of large language models (LLMs) with the goal of releasing them as open source once they reach certain performance thresholds. According to statements by its CEO, Octave Klaba, to Reuters, the company has already completed pre-training of one of its models on Jupiter, Europe's most powerful supercomputer and the first at exascale scale, located in Germany.

This step puts OVHcloud in direct competition with Mistral AI, the Parisian startup that has become the European standard-bearer against US giants. Klaba argues that the economics of model training have changed dramatically: the cost of a project that previously required around $1.15 billion (€1 billion) has now dropped to less than $230 million (€200 million), thanks to improvements in chips, training methods, and the use of synthetic data.

The Geopolitical Context: Data Sovereignty and Access Continuity

This announcement comes at a critical time. European governments and companies are reassessing AI infrastructure not only for performance but also for data governance and access continuity. Tensions escalated recently when Anthropic, a US company, suspended access to its Fable 5 and Mythos 5 models for foreign citizens, following US export control directives. Such incidents reinforce the need for local alternatives.

For CIOs and CTOs, the decision to adopt an AI model is no longer purely technical; it involves assessing jurisdictional and dependency risks. As Sanchit Vir Gogia, chief analyst at Greyhound Research, notes, “Sovereignty doesn’t eliminate the kill switch. It changes who holds it.” In this context, OVHcloud's bet could offer a way to reduce dependence on US and Chinese providers, but it does not completely eliminate geopolitical risks.

Beyond Training: The Real Economic Challenge

Although the reduction in training costs is significant, experts warn that it is only the first step. Neil Shah, vice president of research at Counterpoint Research, points out that the €200 million figure likely covers only the initial training phase. Once trained, models require ongoing investment in fine-tuning, post-training, sovereign infrastructure, storage, security, distribution, and enterprise support. “A model is considered an asset that loses value if not continuously trained and kept up to date with data,” Shah explains.

Moreover, OVHcloud will need to achieve sufficient scale for serving models to be economically viable against established competitors like Google and Anthropic. Charlie Dai, principal analyst at Forrester, agrees that lower training costs provide a credible starting point, but enterprise competitiveness will depend on sustained capabilities such as inference efficiency, data pipelines, evaluation frameworks, and ecosystem breadth.

Sovereign Infrastructure, But Not Entirely

A critical point is that pre-training was performed on Jupiter, a European public supercomputer located in Germany, but it runs on US technology. This shows that European AI sovereignty remains partial. As Gogia notes, “$200 million today allows serious training, but it doesn’t buy a solid enterprise AI business.” The absence of published benchmarks and technical details makes OVHcloud's plan, for now, more a statement of intent than a demonstrated capability.

For companies, this means that the decision to adopt OVHcloud models will require evidence that the models can be maintained in production, governed effectively, audited when necessary, and abandoned without major disruptions. Technological lock-in is also a factor: although customers could switch cloud providers, moving AI workloads built around specific models and governance tools would be much harder.

Implications for IT Professionals and Business Strategy

This move by OVHcloud has direct implications for IT departments and AI adoption strategies. On one hand, it offers a potential alternative for companies seeking to comply with data sovereignty regulations like GDPR. On the other, it introduces new variables in risk assessment. CIOs will need to consider not only model performance and cost but also provider stability, continuous update capability, and interoperability with other systems.

In this regard, OVHcloud's experience in cloud infrastructure could be an advantage, but also a distraction. As seen in other areas, such as adoption of enterprise cloud solutions, integration with existing platforms is key. Additionally, security and data governance cannot be neglected, as addressed in penetration testing and ethical hacking.

OVHcloud's bet could also catalyze a broader European AI ecosystem, similar to what is happening with initiatives like AI agents for HR that automate processes without waiting for instructions. However, success will depend on OVHcloud's ability to demonstrate value beyond initial training and on the political will of European governments to support these initiatives.

Original source: ComputerWorld. Analysis and adaptation by ForgeNEX.