In the framework of EU Horizon project AI4PEX, the Meteorological Research Unit of SMHI is developing a novel AI/ML architecture that will allow extraction of mesoscale cloud patterns from the advanced satellite sensor imageries. In the long run, this architecture will also establish connection between clouds and prevailing meteorology. These developments have important implications. Clouds make our planet habitable by regulating the solar radiation and precipitation in the form of rain and snow. Clouds are also tightly connected to other essential climate variables such as surface temperatures. Therefore, understanding mesoscale patterns of clouds, their frequency and persistency lies at the heart of understanding their overall role in the Earth’s radiation budget. Since the climate models still cannot fully resolve some of the mesoscale cloud patterns, especially over the oceans where these clouds exert strong cooling effect, it is important to inform climate models about the extent and variability of these clouds. This will eventually lead to better representation of these clouds in climate models and less uncertainties in their future projections. In addition to the climate applications, the development of this AI architecture will also help to improve the representation of clouds in the weather forecast models, thus leading to better representation of cloud-radiation-precipitation interactions. In practice, we will train and process 5 years of geostationary satellite data from Meteosat Second Generation at 3 km spatial resolution and 15 minutes temporal resolution amounting to nearly 20 Tb of data. This is a very challenging task that requires access to GPUs and computing power to leverage the power of AI for climate applications.