Photonic compute and memory is the category of architectures that process or store information using light rather than — or alongside — electrons. It is the hardware substrate for the investment theses explored in Photonic Compute Market and Optical Memory Market, and covers two distinct device classes:
Photonic compute uses optical interference, diffraction, or nonlinear effects to perform operations — most commonly matrix–vector multiplication — at the speed of light and without charge transport. Implementations range from Mach–Zehnder interferometer meshes (integrated silicon-photonic) and SLM-based free-space matmul, to D²NNs and Photonic Tensor Cores. The canonical figure of merit is operations-per-joule at the system boundary, which must account for optical-to-electrical-to-optical (O/E/O) conversion losses that can erase the in-fabric efficiency gain. Optical Neural Networks are the primary inference target.