Thermal modelling of the winding-hotspot is a necessary prerequisite for using numerous promising concepts which help to ensure the most economic utilization and the highest reliability of power transformers. Examples are Realtime Thermal Rating (RTR), dynamic availability management as well as HI-Rating (Health Index), and based on this condition triggered maintenance. The state of the art of thermal modelling of oil immersed power transformers was researched as part of the project 'dynamic availability management', and evaluated in respect of the practicability for application in the above scenarios, by implementing the selected models in MATLAB®. For this reason, and representing the first group of thermal models, the ones proposed in the latest release of the standards IEEE C57-91:2011 and IEC 60076-7:2008 have been evaluated. The second group are differential equations, derived from thermoelectric analogy models. In both groups, there has been a huge number of proposed improvements of the basic models by considering additional influences, like nonlinear thermal resistance, solar radiation, wind speed, unbalanced loading conditions, harmonic distortion and moisture in the paper-oil-isolation-system. The third group describes methods that are based on artificial intelligence. Their application is divided into two scenarios. On the one hand they can be used to support the IEC 60076-7 and thermoelectric analogy model to implement an additional error prediction model and improve the obtained results. On the other hand, they can be used as stand-alone model for temperature prediction. To compare the mentioned models, different error metrics are defined and tested with the implemented models. The results of these model calculations allow for a more optimal utilization of the existing infrastructure. Furthermore, by incorporating the thermal models for precisely planned short- and long-term overloading, an additional fault tolerance can be achieved without putting the valuable and time-consuming to replace infrastructure under unacceptable risk. This way, outage time and costs can be reduced.