SpiNNaker2 is a digital neuromorphic chip designed for scalable machine learning, featuring event-based and asynchronous processing. It allows the composition of large-scale systems with thousands of chips, leveraging low-power analog and digital technologies to reduce energy consumption and latency. The system is composed of asynchronously operating processing elements (PEs) interconnected by an efficient network-on-chip, enabling efficient communication and computation. SpiNNaker2 supports a wide range of machine learning applications, including artificial neural networks (ANNs), spiking neural networks (SNNs), and generalized event-based neural networks. It offers native support for hybrid models beyond pure DNNs or SNNs, including symbolic models. The system's event-based nature enables autonomous and asynchronous execution between cores, with dynamic adaptation of clock frequency and supply voltage to save energy. SpiNNaker2 has been manufactured in large numbers and assembled into the world's largest brain-like supercomputer with over 5 million processing elements, available to researchers globally. The system supports event-based and asynchronous algorithms, offering an alternative to GPU-centric models. SpiNNaker2's architecture enables efficient communication between PEs, chips, and boards, with packet types allowing for efficient data transfer. The system supports various machine learning applications, including deep rewiring for sparse-to-sparse training, event-based backpropagation, and biologically-inspired learning. SpiNNaker2's event-based processing allows for energy-efficient training and inference of ANNs, with applications such as deep rewiring achieving high accuracy with low memory constraints. The system also supports spiking neural networks, with event-based backpropagation and e-prop for efficient learning. SpiNNaker2's architecture enables large-scale simulation of event-based and asynchronous machine learning systems, offering a path to more energy-efficient computation. The system's flexibility and scalability make it a promising platform for future machine learning research.SpiNNaker2 is a digital neuromorphic chip designed for scalable machine learning, featuring event-based and asynchronous processing. It allows the composition of large-scale systems with thousands of chips, leveraging low-power analog and digital technologies to reduce energy consumption and latency. The system is composed of asynchronously operating processing elements (PEs) interconnected by an efficient network-on-chip, enabling efficient communication and computation. SpiNNaker2 supports a wide range of machine learning applications, including artificial neural networks (ANNs), spiking neural networks (SNNs), and generalized event-based neural networks. It offers native support for hybrid models beyond pure DNNs or SNNs, including symbolic models. The system's event-based nature enables autonomous and asynchronous execution between cores, with dynamic adaptation of clock frequency and supply voltage to save energy. SpiNNaker2 has been manufactured in large numbers and assembled into the world's largest brain-like supercomputer with over 5 million processing elements, available to researchers globally. The system supports event-based and asynchronous algorithms, offering an alternative to GPU-centric models. SpiNNaker2's architecture enables efficient communication between PEs, chips, and boards, with packet types allowing for efficient data transfer. The system supports various machine learning applications, including deep rewiring for sparse-to-sparse training, event-based backpropagation, and biologically-inspired learning. SpiNNaker2's event-based processing allows for energy-efficient training and inference of ANNs, with applications such as deep rewiring achieving high accuracy with low memory constraints. The system also supports spiking neural networks, with event-based backpropagation and e-prop for efficient learning. SpiNNaker2's architecture enables large-scale simulation of event-based and asynchronous machine learning systems, offering a path to more energy-efficient computation. The system's flexibility and scalability make it a promising platform for future machine learning research.