Photonics for artificial intelligence and neuromorphic computing has seen significant advancements due to the development of photonic integration platforms and optoelectronic components. Photonic integrated circuits enable ultrafast artificial neural networks, offering a new class of information processing machines. These systems have the potential to address the growing demand for machine learning and AI in areas such as medical diagnosis, telecommunications, and high-performance computing. Neuromorphic photonics provides subnanosecond latencies, complementing AI applications. This review discusses recent advances in integrated photonic neuromorphic systems, current and future challenges, and the technologies needed to overcome them. Conventional computers are not efficient for distributed, parallel, and adaptive models used in AI. Neuromorphic engineering aims to match hardware to algorithms for faster and more energy-efficient processing. Neuromorphic hardware is also applied to problems outside of ML, such as robot control and neuroscientific hypothesis testing. Current neuromorphic electronics use shared digital communication buses, trading bandwidth for interconnectivity. Optical interconnects could negate this trade-off, accelerating ML and neuromorphic computing. Optics is well-suited for matrix multiplication and interconnects. Optical approaches to neural networks were pioneered by Psaltis and others. Silicon photonics is a crucial advance, enabling high-quality passive components and high-speed active optoelectronics. A scalable silicon photonic neural network was first proposed in 2014 and demonstrated in 2017. On-chip silicon electronics provide a route to overcome component sensitivity, and progress in on-chip optoelectronics enables cascadability and nonlinearity. Photonic circuits are well-suited for high-performance neural networks due to interconnectivity and linear operations. Connections between neurons are described by scalar synaptic weights, allowing matrix-vector operations. Optical signals can be multiplied by transmission through tunable waveguide elements and added through wavelength-division multiplexing. Neural networks require long-range connections, and photonic waveguides offer lower attenuation and less heat generation compared to metal wires. Optics has long been recognized for matrix multiplication and interconnects. Optical approaches to neural networks were pioneered decades ago. Today, silicon photonics is a crucial advance, enabling high-quality passive components and high-speed active optoelectronics. A scalable silicon photonic neural network was first proposed in 2014 and demonstrated in 2017. On-chip silicon electronics provide a route to overcome component sensitivity, and progress in on-chip optoelectronics enables cascadability and nonlinearity. Photonic circuits are well-suited for high-performance neural networks due to interconnectivity and linear operations. Connections between neurons are described by scalar synaptic weights, allowing matrix-vector operations. Optical signals can be multiplied by transmission through tunable waveguide elements and added through wavelength-division multiplexing. Neural networks require long-range connections, and photonic waveguides offer lower attenuation and less heat generation compared toPhotonics for artificial intelligence and neuromorphic computing has seen significant advancements due to the development of photonic integration platforms and optoelectronic components. Photonic integrated circuits enable ultrafast artificial neural networks, offering a new class of information processing machines. These systems have the potential to address the growing demand for machine learning and AI in areas such as medical diagnosis, telecommunications, and high-performance computing. Neuromorphic photonics provides subnanosecond latencies, complementing AI applications. This review discusses recent advances in integrated photonic neuromorphic systems, current and future challenges, and the technologies needed to overcome them. Conventional computers are not efficient for distributed, parallel, and adaptive models used in AI. Neuromorphic engineering aims to match hardware to algorithms for faster and more energy-efficient processing. Neuromorphic hardware is also applied to problems outside of ML, such as robot control and neuroscientific hypothesis testing. Current neuromorphic electronics use shared digital communication buses, trading bandwidth for interconnectivity. Optical interconnects could negate this trade-off, accelerating ML and neuromorphic computing. Optics is well-suited for matrix multiplication and interconnects. Optical approaches to neural networks were pioneered by Psaltis and others. Silicon photonics is a crucial advance, enabling high-quality passive components and high-speed active optoelectronics. A scalable silicon photonic neural network was first proposed in 2014 and demonstrated in 2017. On-chip silicon electronics provide a route to overcome component sensitivity, and progress in on-chip optoelectronics enables cascadability and nonlinearity. Photonic circuits are well-suited for high-performance neural networks due to interconnectivity and linear operations. Connections between neurons are described by scalar synaptic weights, allowing matrix-vector operations. Optical signals can be multiplied by transmission through tunable waveguide elements and added through wavelength-division multiplexing. Neural networks require long-range connections, and photonic waveguides offer lower attenuation and less heat generation compared to metal wires. Optics has long been recognized for matrix multiplication and interconnects. Optical approaches to neural networks were pioneered decades ago. Today, silicon photonics is a crucial advance, enabling high-quality passive components and high-speed active optoelectronics. A scalable silicon photonic neural network was first proposed in 2014 and demonstrated in 2017. On-chip silicon electronics provide a route to overcome component sensitivity, and progress in on-chip optoelectronics enables cascadability and nonlinearity. Photonic circuits are well-suited for high-performance neural networks due to interconnectivity and linear operations. Connections between neurons are described by scalar synaptic weights, allowing matrix-vector operations. Optical signals can be multiplied by transmission through tunable waveguide elements and added through wavelength-division multiplexing. Neural networks require long-range connections, and photonic waveguides offer lower attenuation and less heat generation compared to