OVHcloud has decided to enter a space that until recently seemed reserved for specialized laboratories such as Mistral AI, OpenAI, Anthropic, or Google DeepMind: the development of frontier AI models. This move doesn’t automatically make the French group a direct rival to Mistral because it hasn’t yet published a comparable model or verifiable performance metrics. But it does change the European conversation about sovereign AI.
Until now, Europe had focused much of its hopes on Mistral as the continent’s leading champion of generative models. OVHcloud presents a different thesis: to compete in AI, having talented researchers and powerful models isn’t enough; controlling data centers, servers, energy, networks, and cloud operations is also crucial. And here, the French group has an advantage many startups lack: already deployed infrastructure.
From Cloud Provider to Model Laboratory
OVHcloud’s shift began to take shape with the acquisition of Dragon LLM, a French company specializing in sovereign generative models and fine-tuning for regulated sectors. The deal also launched OVHcloud’s AI Lab, with the declared goal of training and fine-tuning sovereign LLMs that can be deployed both on cloud and local infrastructures.
The ambition has stepped up a gear. Octave Klaba, founder and CEO of OVHcloud, told Reuters that the company aims to train frontier models from scratch and become a second major European player in LLMs, alongside Mistral. Their approach involves a family of specialized models, not a single generalist, with plans to release them as open source once they reach a sufficient performance level.
Klaba also shared a relevant detail to explain why OVHcloud believes it can enter this race now: in his view, the costs of training advanced models have decreased. What previously might require around 1.5 billion euros could now be attempted with 150 or 200 million, thanks to improvements in chips, training techniques, and synthetic data usage.
The company has already completed pretraining a model using Jupiter, the European supercomputer installed in Germany, though performance results haven’t been made public yet. This nuance is important. OVHcloud has the intent, team, and infrastructure, but it still needs to prove that its models can compete in quality, cost, and adoption.
| Aspect | OVHcloud | Mistral AI |
|---|---|---|
| Main profile | European cloud provider with own infrastructure | AI Model laboratory |
| Initial advantage | Data centers, servers, cloud operations, and enterprise clients | Specialized talent, recognized models, and a strong AI brand |
| Known revenues | €1.085 billion in fiscal year 2025 | Aims for near €1 billion in 2026, per market estimates |
| Infrastructure | Over 500,000 servers and 46 data centers, according to OVHcloud | Depends on agreements with external infrastructure providers |
| Model strategy | Family of specialized models, with open source plans | Open and commercial models focused on enterprise and developers |
| Recent activities | Acquisition of Dragon LLM and creation of AI Lab | Funding rounds and business expansion |
| Main challenge | Proving model quality and developer ecosystem | Scaling infrastructure, revenues, and differentiation against US giants |
AI is no longer decided solely by the best model
The most interesting insight isn’t whether OVHcloud can “beat” Mistral. Such a comparison oversimplifies the market. What matters is that generative AI is entering a second phase, where infrastructure matters just as much as algorithms.
The early years of the generative race were dominated by models, benchmarks, context windows, conversational assistants, and funding. Now, the physical aspects come into focus: who has GPUs, who has memory, energy, cooling, the capacity to host sensitive workloads, and who can offer predictable costs to businesses and governments.
In this arena, OVHcloud has its own arguments. The company positions itself as the leading European cloud provider, with 1.6 million clients, 46 data centers, over 500,000 servers, 3,000 employees, and €1.085 billion in revenue for fiscal 2025. It also reports an adjusted EBITDA of €438 million, indicating that it’s not a startup in pure spending mode but an established cloud business with operational revenues.
This scale doesn’t guarantee a good AI model, but it does shift cost structures. An established data center operator can train, deploy, and commercialize models differently. It can integrate AI into existing cloud services, offer sovereign deployments, better control regulatory compliance, and sell to clients already trusting its infrastructure.
Additionally, the geopolitical context favors European providers. The suspension of access to advanced models from Anthropic for certain foreign users and concerns over US platform dependence have heightened interest among companies, governments, and regulated sectors for European alternatives—open models and controlled deployments.
Sovereignty, cloud, and open models
OVHcloud isn’t new to the discourse on digital sovereignty. Its cloud proposition has long relied on European control, predictable pricing, technological independence, and deployments for regulated sectors. The acquisition of Dragon LLM aligns with this stance, as the acquired firm focused on specialized models for sensitive areas with local deployment capabilities.
The challenge is turning this positioning into a tangible product. In enterprise AI, claiming “sovereign” isn’t enough. Companies need useful models, stable APIs, documentation, deployment tools, security assessments, support, clear costs, integration with their own data, and the ability to operate in hybrid environments. They also need transparency about training data, inference locations, and compliance with the AI Act, GDPR, and sector-specific regulations.
Choosing to focus on specialized models may be a smart move. Competing head-to-head with the largest generalist models requires enormous compute and talent. In contrast, a family of tailored models can be more practical for banking, defense, industry, public administration, healthcare, or professional services—especially if deployed on European infrastructure and under clear contractual control.
For Mistral, OVHcloud’s entry isn’t necessarily bad. Europe needs more than one strong player to have real influence in AI. Relying on a single ecosystem would be fragile. Internal competition can improve models, accelerate integrations, attract clients, and demonstrate that European AI isn’t solely dependent on a highly valued laboratory.
The competition will be intense. Mistral already has a brand, developer community, and models in production. OVHcloud has infrastructure, clients, and operational experience but must prove its technical credibility in a market that moves fast and doesn’t forgive easily. Announcing a model without solid results isn’t enough; a sovereign model that underperforms isn’t either.
OVHcloud’s move conveys a clear idea: future AI winners may not only be those designing the best models but those controlling the entire stack. Data, compute, energy, network, security, deployment, support, and costs will all be part of the product. AI is becoming an infrastructure-intensive industry, not just a software race.
For Europe, this could be an opportunity. The continent lacks the hyper-scalers of the US and the industrial capacity of China but has cloud providers, data centers, public research, institutional demand, and regulatory pressure favoring controllable solutions. OVHcloud is trying to capitalize on this intersection.
The key question remains whether they will succeed in time. The window to build a European alternative in AI remains open but won’t last forever. Clients want sovereignty, but also performance, cost-effectiveness, and ease of use. OVHcloud has shown the will to participate. Now it must demonstrate that its physical advantage translates into meaningful intelligence.
Frequently Asked Questions
Will OVHcloud compete directly with Mistral AI?
OVHcloud aims to train frontier models and become another relevant European AI player. Since it hasn’t yet published comparable results, the real competition will depend on the performance, availability, and adoption of its models.
What does OVHcloud bring that a startup doesn’t?
It offers already deployed cloud infrastructure: data centers, servers, operations, enterprise clients, and experience in sovereign services. This foundation can reduce dependency on third parties for training and deploying AI.
What is Dragon LLM?
Dragon LLM is a French company specializing in sovereign generative models and fine-tuning for regulated sectors. OVHcloud acquired it to strengthen its AI Lab and develop LLM-based services.
Why is infrastructure so important in AI?
Because training and deploying advanced models require GPUs, energy, memory, cooling, networks, and data centers. Model quality matters, but reliable and cost-effective access to compute resources has become a strategic advantage.
via: LinkedIn
