The increase in virtual educational spaces has meant that teachers develop digital skills required in the 21st century. However, few studies explore teacher training in virtual environments linked to the concept of universalization of computational thinking to improve their performance in virtual teaching environments. The objective of this work is to analyze the impact of teacher training in universalization of computational thinking to enhance educational innovation in virtual teaching environments and the use of artificial intelligence (AI). Supported by the Google Scholar, Scopus and Springerlink databases, a systematic review of 75 scientific publications was carried out, using quantitative and qualitative scientific methods, as well as the Prism Model, which allowed organizing quality and selection criteria. The results show that the definition of computational thinking changes as its understanding materializes and develops in new generations of teachers based on a contemporary vision of the concept applied to all citizens. The curricular integration of programming and robotics in teacher training improves their digital literacy. At the same time, the educational practices resulting from the dual dialectic: experiment-conceptualize and act-reflect, supported by Kold's model, could favor educational innovation and the use of AI in educational environments within an ethical and social framework. The TRACK model is pointed out as a strategy to implement computational thinking in teacher training, considering three types of knowledge (namely: technological, pedagogical and the contents of the discipline) necessary for STEAM education, as well as improving teacher performance. in virtual teaching environments.