In the last decade, Artificial Intelligence (AI) algorithms are becoming very influential in several sectors thanks to their flexibility and adaptability. Among the several fields of the AI, the most attractive and with the greatest results is Machine Learning (ML). It aims to extract and elaborating characteristics starting from a huge amount of data. Space missions could strongly benefit of these algorithms, allowing the elaboration and processing of data directly where they are acquired. Thanks to the technological advancements in designing low-power hardware accelerators, several enterprises (both private and public) have used Components Off-The-Shelf (COTS) hardware accelerators to run their on-the-edge applications. This has allowed to mitigate the server Graphical Processing Unit (GPU) high power consumption and overheating while maintaining good performance. However, these devices cannot be used in the harsh space environment, due to their low radiation tolerance, that is a fundamental requirement for safely working in space. GPU@SAT aims to revolutionise the current Artificial Intelligence execution and signal processing analysis, exploiting a general-purpose GPUlike IP-core fitted on a radiation-hardened FPGA. The board will be adopted in the framework of a European Space Agency (ESA) project under the Advanced Research in TElecommunications Systems (ARTES) programme. AIKO, the company leader of this activity, is developing AI algorithms for Fault Detection Isolation and Recovery systems. The adoption of such algorithms, executed via dedicated and space-qualified systems, directly on-board of satellites, can strongly increase the life of the mission, while reducing the design costs.