How they use Nebius AI Cloud

Jua trains and refines its Earth Physics Transformer weather model on Nebius AI Cloud. Its own engineering article describes distributed training and visits to the Nebius data center. Nebius’s customer story also describes dedicated endpoints that deliver forecasts. Jua uses those forecasts in products for energy traders and utilities.

In April 2025, Jua engineer Jordan Daubinet described running GPU-provider evaluation pilots that included Nebius, assessing performance, support and usability for training infrastructure. In May 2026, Nebius reported that Jua moved large-scale EPT-2 training to its cloud after reliability challenges with three other providers. The post describes NVIDIA GPUs and InfiniBand networking supporting model development and Earth physics simulation. Nebius reports that EPT-2 delivers hourly global forecasts at six times higher temporal and spatial resolution than comparable AI models. Its customer story also reports forecasting performance that exceeds leading AI and traditional weather models.

Nebius says Jua expected training runs to last months when it set out to simulate Earth’s physics. Its customer story describes thousands of GPUs running on Nebius AI Cloud. The story estimates that Jua’s forecasts have an annual profit and loss impact of $1.5 million per gigawatt in European energy markets.