OVHcloud expands beyond hosting with Olala models and agent workspace
OVHcloud unveiled 17 new AI products at its Paris summit, expanding from cloud hosting into model development and agent-based services. Following its acquisition of Dragon LLM, the company launched Olala, a model family covering base, instruction, translation, summarization, and mini variants. The smallest has 7 billion MoE parameters, uses just 1 billion parameters at inference, and can run on a standard CPU.
The roadmap includes version 1.1 with a context window of 128,000 tokens by the end of the year, followed by version 2.0 with 100 to 300 billion parameters in early 2027. OVHcloud wants advanced DevOps, agentic AI, and cybersecurity capabilities, using 27 years of its own operational logs for reinforcement learning rather than client data. Dedicated research labs will cover models, voice, and document processing.
A new digital workplace, described by Octave Klaba as an “AI OS,” combines email, calendars, file storage, text editing, and videoconferencing through a prompt-first interface. Its free alpha includes 5 GB of storage per user, with data hosted in Europe. Other launches span agent deployment, model routing, managed search, and automated phone reception.
OVHcloud aims to become Europe’s default sovereign technology choice. Klaba said partnership discussions with Mistral AI ended without an agreement after 18 months of talks, leaving OVHcloud to build expertise through acquisitions and internal development.