Light-based processing replaces electrons to execute matrix multiplications at higher speeds. Photonic chips reduce heat and power consumption compared to traditional GPUs. This shift targets the energy bottleneck of massive neural networks. While promising, current integration with digital memory remains a hurdle for AI hardware developers seeking scalable, low-latency inference systems.