OVH plans to control the entire cloud and AI value chain.
With its “Shaping the Future” program, the French operator has high ambitions for digital sovereignty. Mastering its AI models is part of that. It will complement the current value chain of IaaS, PaaS, servers, storage, and connectivity.
Server Design
OVHcloud designs and assembles its own servers internally (particularly at sites in France and Canada
Data Center Expertise
The company has developed its own water-cooling technology. It builds and operates its 46 data centers spread across four continents.
Circular Economy
It uses a reverse supply chain that allows it to reuse a significant number of its servers' components and spare parts
Proprietary fiber-optic network
The company deploys and manages its own global fiber-optic network (with a capacity of over 100 Tbit/s).
Cloud Infrastructure and Storage
It operates its own dedicated bare-metal servers, provides SecNumCloud private cloud infrastructure (Hosted Private Cloud) and, more recently, has launched the SNC Cloud Platform, in addition to a public cloud offering based on OpenStack, as well as a bare-metal cloud solution for high-performance computing.
Adding sovereign AI to this cloud value chain
To gain control over the end-to-end cloud value chain, OVH has announced the development of its own LLM (OVHai LLM) and has expressed its desire to break away from Mistral AI.
An open-source version (open weight) is scheduled to be released in late November 2026. This model is not intended to replicate existing LLMs. Instead, it is based on an architecture designed to optimize information processing, token processing, and data compression. The model is based on the acquisition of Dragon LLM last March.
Mastering its AI model is a question of survival for OVH.
According to OVH’s founder, “mastering one’s own AI model is a matter of survival to avoid dependence on U.S. software infrastructure, which could eventually take over the ERPs and collaborative tools markets. This explains the company's decision to break away with Mistra AI’s models.”
OVH’s models will be trained on its proprietary data in three key sectors: cybersecurity, code assistance, and agent-based AI for support functions.
Therefore, we could consider that OVH is moving more toward providing specific SLMs (Small Language Models) dedicated to specific business domains rather than a true general-purpose LLM.
The Value of OVHai LLM
Unlike Mistral AI, which develops very broad-spectrum general-purpose models (“Frontier models”), OVHai LLM is based on a hybrid and sovereign AI architecture designed to move beyond traditional Transformer architectures, which are particularly energy-intensive and resources-intensive. The solution will offer models that are natively specialized for regulated sectors (banking, insurance).
An OVHAI-based SLM (Small Language Model) can run locally on standard servers or even a simple PC, without requiring massive and expensive graphics chips, allowing operators to move away from non-sovereign resources, such as Nvidia.
Integrated into OVHcloud’s software stack, OVHai’s SLM model will run on a 100% European, proprietary infrastructure. It will be immune to extraterritorial laws (such as the U.S. Cloud Act) and will benefit from the highest security certifications, such as SecNumCloud (ANSSI) and HDS (health data).
Practical uses for AI made in OVHcloud
Setting up a sovereign “AI Lab” for fine-tuning
This lab will provide customers with managed tools to fine-tune small, highly specialized language models using their own business data. This service will specifically target regulated sectors (banking, insurance, healthcare, legal) that require highly accurate AI models but whose data cannot leave their infrastructure.
Set for a global launch in early 2027, the core business model will use a Mixture-of-Experts architecture. This method connects and intelligently routes queries to small, highly specialized sub-models (similar to Dragon LLM’s historical vertical approach) rather than relying on a single massive model.
Integration with PaaS AI Endpoints services
Developers will be able to use these specialized models (such as the Dragon-3B model, which is already available on Hugging Face) as “Tokens as a Service” via a simple serverless API.
Integration with the OVHai Workspace SaaS platform
OVHcloud will integrate Dragon LLM’s document comprehension and agent-based AI capabilities into its SaaS applications.
It has launched the pilot phase of OVHai Workspace, a sovereign office and collaboration suite (email, video conferencing, file storage) that is entirely driven and supported by its proprietary AI.
While most cloud providers are currently adopting collaborative and/or co-developed cloud and AI strategies to gain market share, OVH stands out today for its commitment to taking control of its entire cloud and AI value chain (hardware, software, models, data centers, connectivity).
As such, it will soon be one of the few providers able to claim complete digital autonomy. The question remains whether this strategy will ultimately prove successful.
