Artificial Intelligence (AI) is playing an increasingly important role in our daily life. The functionality and performance originate from large neural network architectures. The inference and training thereof are a computational challenge. Accelerators performing the calculation of the neural network interconnects are applied to enhance the performance and power-efficiency. Parallel analog computing schemes offer the potential of additional enhancements by mitigating memory access and a better mapping of the network in hardware. Electrical and integrated optic approaches will be discussed. What is required to make photonic solutions a compelling alternative?