OVHcloud plans to train frontier AI models - the most advanced, large-scale systems built from scratch using vast data and computing power - its CEO said on Wednesday, positioning the firm as a potential European challenger to Mistral.
The move marks a shift for OVHcloud, Europe's largest cloud provider, as governments and companies seek alternatives to U.S. and Chinese AI systems - a search made more pressing by the recent abrupt switch-off of Anthropic's top-tier models.
"It became quite clear to us that if we don't master this technology, we can't guarantee our future," OVHcloud CEO Octave Klaba told Reuters at the VivaTech conference.
Klaba said the economics of developing such cutting-edge models have shifted, citing advances in chips, training techniques and synthetic data. A project that might once have cost about 1 billion euros ($1.2 billion) could now be attempted for 150 million to 200 million euros, he said.
He described the industry as entering a "second wave," with new entrants building on groundwork laid by firms such as OpenAI, Anthropic, and Mistral. He added OVHcloud would not use client data to train its models.
The company plans to launch a family of models rather than a single system. "We can clearly see that the major players release multiple models, because each model is built for something specific," Klaba said. "There's no one model that does all the magic alone."
He pointed to DragonLLM, a recently acquired startup, adding that pre-training has been completed on a model using Jupiter, Europe's fastest supercomputer, but cautioned that OVHcloud was not yet ready to make detailed performance claims.
OVHcloud intends to open-source its models once they reach sufficient performance. "We'll see when we're good enough to open source them, but that is indeed the goal," he said.
