OVHCloud is preparing an open-source LLM for late November

OVHCloud plans to release a large language model as open source by the end of November. The company highlights an energy capacity of 250 MW in its own data centers and a reduction in development costs, while targeting specific use cases such as cybersecurity, coding, and autonomous agents.

OVHCloud relies on 250 MW in its data centers

OVHCloud has 250 megawatts of electrical power in data centers that it owns. The company already offers infrastructure capable of running open models. Octave Klaba recalls being criticized during the IPO for his investments in this infrastructure and asserts that it is now worth a lot of money.

Targeted use cases: cybersecurity, coding, and autonomous agents

The development of OVHCloud's models focuses on three main use cases: cybersecurity, both defensive and offensive, coding, and agentic tasks, meaning AIs capable of operating relatively autonomously in areas such as finance. No specific technical data about the model has been disclosed so far.

Publication expected by the end of November and training underway

The release of the large open-source language model is scheduled for the end of November. The training of the first version is nearing completion. Octave Klaba indicates that this model will be different from others and will be significantly advanced in terms of architecture. He believes it is essential for OVHCloud to have its own model to avoid dependency on a third-party provider that could cut access after exploiting the data.

Cost context: announced reduction and new GPUs

Octave Klaba claims that the development costs of large language models have decreased. He attributes this reduction to hardware and software innovations, particularly in model architecture. He also mentions the arrival of new, more powerful GPUs, which are not much more expensive.