

Why AI Needs a New Kind of Computer
Modern AI can write code and generate images, yet much of its hardware still follows a design rooted in 1945. This episode explores the von Neumann bottleneck: the costly shuttling of information between memory and processors. As AI’s appetite for electricity grows, researchers are rethinking that architecture—bringing computation closer to stored data and borrowing ideas from the brain. Follow the journey through memristors that store values and perform calculations, brain-inspired chips that communicate through electrical spikes, IBM’s NorthPole architecture, and connections that carry data using light. Then enter the world of biological computing, where living neurons interact with silicon and learn to play a simplified version of Pong. Along the way, we explore the engineering hurdles, the promise of more efficient AI on everyday devices, and the demands of keeping a computer’s cells alive. If a machine contains living tissue that can learn and adapt, where do we draw the line between computer and organism?


















