In this paper, we propose two novel physics-informed machine learning methods designed to estimate input wind loads in offshore wind turbines, addressing the challenges of scenarios where input loads are not directly measurable. The frameworks integrate the principles of physics-informed neural networks, embedding the equations of motion into the loss functions to ensure that predictions adhere to the fundamental physical laws governing the system. The first method treats the input load as a learnable parameter, leveraging measured displacements and accelerations to dynamically estimate the input load while minimizing the discrepancy between predicted and actual values through a physics-based loss function. The second method employs an integrated network of intertwined sub-networks that simultaneously predict input loads and dynamic responses, with a loss function composed of equation of motion loss, differentiation law loss, and data loss, ensuring physical consistency and accuracy. Both methods are going to be tested on measured data from offshore wind turbines to demonstrate their effectiveness.