The AI technology landscape is rapidly evolving, raising many questions about the next major advancements and how to prepare for them. Amid extensive discussion and speculation, this talk presents a software perspective that identifies two clear trends: (1) converging computation patterns and (2) increasing memory-to-compute ratios. These trends suggest the AI stack will be greatly simplified by new system–hardware interfaces and will move toward a more scalable, memory-centric architecture. Based on these observations, we will examine current challenges in AI development and deployment, how emerging technologies might address them, and how we can prepare for or lead these changes.

Dr. Changho Hwang is a senior researcher at Microsoft, specializing in scalable AI systems design and optimization. His work spans system design for emerging AI algorithms, efficient GPU communication, and software for next-generation AI hardware. He is also interested in using AI techniques to design and optimize computer systems. He has published papers in top-tier conferences including NSDI, ISCA, MLSys, and ICML. Before joining Microsoft, he earned a Ph.D. in Electrical Engineering from KAIST, where he was advised by Prof. KyoungSoo Park.