“The world needs new AI hardware architectures that are memory-centric, allowing compute to be scaled independently to improve power efficiency and cost.”