To avoid a significant increase in operational expenditure future generations of communication satellites, especially those deployed as part of a large constellation, will need to incorporate a much greater level of operational autonomy. Such an advancement couldalso: help realise an increase in the complexity and hence capabilities of such constellations; move from failure-triggered replenishment and reconfiguration toward predictive maintenance; as well as potentially extend satellite lifetimes. This system-level studytherefore seeks to identify the requirements for, and the topology of, a new satellite platform architecture able to realise such anadvancement through the exploitation of Artificial Intelligence and Machine Learning technologies. The activity will firstly consider the operational functions best realised onboard a future platform and those most efficiently implemented within the ground/control segment. The study would go on to define a top-level reference platform architecture that provides the identified functions as well as specify the component building blocks, including their functionality and key interfaces, (which could be standardised to maximise advantages, e.g. interoperability, reduction of complexity and cost, increase in reliability and maintainability, and to promote competition).