Within the Phase-II upgrade of the LHC, the readout electronics of the ATLAS LAr Calorimeters is prepared for high luminosity operation expecting a pileup of up to 200 simultaneous pp interactions. This environment with high pileup increases the difficulty of energy reconstruction. The energy computation is performed in real-time using dedicated electronic boards based on FPGAs. To cope with the high pileup, new machine learning approaches are explored. The main challenge is to develop neural networks that are efficient for the energy reconstruction while being of reduced size to fit into FPGAs and the stringent latency requirement of the ATLAS trigger system. In this presentation, I will discuss the usage of recursive neural networks (RNNs) for the energy reconstruction and their implementation on INTEL FPGAs.