Multiscale finite elements provide a solution to numerical homogenization for nonlinear problems. The effect of the microstructure, including a nonlinear behavior, for various loadings is transferred to the homogenized medium through the Representative Volume Element (RVE) technique. The classical nested approach, so-called finite element square (FEM2) (Drosopoulos and Stavroulakis in Non-linear mechanics for composite, heterogeneous structures. CRC Press, 2022; Yvonnet in Computational homogenization of heterogeneous materials with finite elements, Springer, Berlin, 2019; Drosopoulos et al. Comput Mater Sci 95:522–535, 2014; Urbański in The unified, finite element formulation of homogenization of structural members with a periodic microstructure. Cracow University of Technology, 2005), where the upper level homogenized model is linked to a lower level RVE model is very expensive. A data-driven alternative based on interpolation of the constitutive law from a number of calculated examples produced off-line from the RVE and classical or neural network metamodels have been proposed (Drosopoulos and Stavroulakis in Non-linear mechanics for composite, heterogeneous structures. CRC Press, 2022; Drosopoulos and Stavroulakis ACM in J Comput Cult Herit 14:1–19, 2021; Drosopoulos et al. in ASCE J Eng Mech 144:04,018,072, 2018) and can be called FE-ANN. Furthermore, principles of physics informed neural network (PINN) training (Raisi et al. in J Comput Phys 378:686–707, 2019), can be incorporated at both scales: the RVE results provide sensitivity analysis information, while at the upper homogenized level, the PINN may provide more flexibility. The flexibility of a mixed formulation and the usage of an ensemble of physics informed neural networks for the plane nonlinear elasticity problems with a polynomial hyperelastic material law (Dastjerdi et al. in Eng Anal Boundary Elem 143:219–236, 2022) is shown. The technique is a modification of an example taken from (Mouratidou in Ensemble of physics-informed neural networks for solving plane elasticity problems with examples (Mouratidou et al. in Acta Mech 235:6703–6722, 2024). The proposed method follows recent developments in the field, like the coupling of finite elements with artificial neural networks (FE-ANN) (Drosopoulos and Stavroulakis in Non-linear mechanics for composite, heterogeneous structures. CRC Press, 2022; Raisi et al. in J Comput Phys 378:686–707, 2019). The historical developments and the links between mentioned tools are presented with proposals for further development in the spirit of LATIN and data-driven tools (Kirchdoerfer and Ortiz in Comput Methods Appl Mech Eng 304:81–101, 2016). .