Memory management is a decades-old area with a large amount of practical impact. However, there is a frequent misconception that all important research problems in this area are either already solved or are unsolvable. In this talk, I will debunk this myth and make the case that there are three seismic shifts that make this research area more relevant than ever and enable fundamentally new research approaches and problems: 1) The advent of AI for Code and its applicability to memory management problems, 2) New memory management problems that arise in the context of compilation for ML accelerators, and 3) The emergence of RISC-V as an open hardware ecosystem that allows customization of the hardware and enables academic work on hardware-software co-design of memory managers.