OVHcloud, Europe’s largest cloud provider, plans to launch its first AI model in early 2027, chief executive Octave Klaba told Reuters on October 20 as the French company reported its annual results. The model is a mixture-of-experts design that has been in training since June on Jupiter, Europe’s fastest supercomputer, and it would make OVHcloud a model maker in its own right rather than only a host for other labs’ models.
The announcement is a plan, not a shipped model: no name, parameter count or benchmark figures were given. What Klaba did give Reuters was the capacity picture behind it, and a timetable that starts with a pilot of the company’s OVHai Workspace before the model itself arrives.
What Klaba Told Reuters About The OVHcloud Model And Its Training Run
According to Reuters’ reporting of the results-day interview, the model uses a mixture-of-experts architecture — the design family in which only part of the network activates for each token, trading total size for cheaper inference, and the same broad approach used by several of the open-weight releases tracked in the open-weight versus closed models debate. Training began in June on Jupiter, the German-based system ranked as Europe’s fastest supercomputer, which gives the project a European-sovereignty framing from the hardware up: a French cloud company, training on a European machine, for a model it will sell as a European alternative.
Klaba also set out OVHcloud’s capacity position. Reuters reported 250 MW connected, 140 MW operating and 110 MW available, with about half of the company’s capacity available — the headroom the model programme and its AI hosting business will draw on. An OVHai Workspace pilot is imminent, ahead of the model’s early-2027 launch window. The results themselves gave the plan its backdrop: in separate Reuters coverage of the same results, OVHcloud met its 2026 targets and posted its first positive cash flow since its IPO. New model launches are arriving from every direction this month — compare the shape of this announcement with MiniMax’s M3.1 Flash launch, a shipped model with weights and prices attached, and with the checklist in how to read an AI model launch: OVHcloud’s news today is the training run and the intent, not the artefact.
Why OVHcloud Is Moving From Hosting Models To Making One
OVHcloud has built its AI business on infrastructure: GPU capacity, data centres across Europe and a sovereignty pitch to governments and regulated industries that cannot put data on US hyperscalers. Owning a model changes its position in that stack. A house model lets OVHcloud sell a complete, European-controlled service — compute, hosting and the model itself — instead of renting capacity to customers who bring models from Mistral, Meta or US labs. It also puts the company in direct conversation with the sovereign-AI strategy that has centred on Mistral, examined in why Europe is betting on Mistral: Europe’s policy goal is a domestic frontier capability, and OVHcloud is arguing that the cloud layer should produce one too, not merely host it.
Why It Matters
The OVHcloud model matters less for what it is today — an unnamed mixture-of-experts model mid-training — than for who is building it. Frontier models have so far come from dedicated labs; a cloud provider entering the field says the economics have shifted enough that owning a model is now a cloud strategy, because the model is what locks customers onto the capacity underneath it. With 110 MW available and half its capacity free, OVHcloud has the physical headroom most labs are queuing for. The risk is equally clear: training since June on Europe’s fastest machine is a serious spend for a company that has only just posted its first positive cash flow since listing, and an early-2027 launch will land in a market where Mistral, DeepSeek and the US labs will all have moved again. Whether Jupiter’s run produces a competitive model is unproven — but Europe’s sovereign-AI map now has a second French name on it.
Background
OVHcloud was founded by Octave Klaba in 1999, grew from a Roubaix hosting company into Europe’s largest cloud provider, and listed in Paris in 2021. It has spent the years since expanding data-centre capacity and building an AI offer around GPU infrastructure and data-sovereignty guarantees for European customers, while France’s Mistral AI became the continent’s flagship model lab. Jupiter, the exascale system at Forschungszentrum Jülich in Germany, entered service as Europe’s fastest supercomputer and is the training ground Klaba has chosen for the company’s first model.
0 Comments