As the space industry is undergoing an evolution, the current approaches toward design, development, and qualification of Electric Propulsion (EP) systems largely based on empirical “trial-and-error” methodologies are falling short of addressing the emerging needs and keeping abreast of the rapid changes in market trends. Furthermore, with the proliferation of Artificial Intelligence (AI) within the space industry toward next-generation autonomous satellites and spacecrafts, the conventional EP monitoring and control strategies become inadequate and need to give way to approaches compatible with satellite-level autonomy requirements. A digital twin (DT) – a technology capable of providing an accurate dynamically adapting virtual representation of a physical asset – is a game-changing concept that catalyzes the transcendence of the EP industry past its pressing challenges today. In this paper, we aim to: (i) define the DT concept, highlighting how it surpasses traditional modelling, (ii) enumerate the DT’s breakthrough promises for the EP industry, and (iii) specify the challenges to realize practical and scalable EP DTs. Additionally, we report on the technical progress achieved and/or planned at Imperial Plasma Propulsion Laboratory to fill the foundational gaps in three building block elements of DTs, namely, (i) a cost-effective kinetic model to generate extensive high-fidelity databases for machine learning (ML), (ii) ML-enabled models for prediction and analysis of performance and operational behavior, and (iii) a DT architecture that integrates the numerical models in terms of a computing infrastructure and provides data pipelines and interfaces for the DT’s data exchanges with the real world, its dynamic updating, and uncertainty quantification.