We are experiencing a massive shift in how we write software, from human-driven to data-driven. Data-driven software, or software 2.0, will allow us to properly realize the true benefits of intelligent robots through the adoption of deep learning, reinforcement learning and continual learning. However, to fully realize these benefits in robotics we need a ground up reimagining of computer hardware and systems to make them data & AI centric, and deliver supercomputer level performance in the power envelope of the edge.