Topic: Trends, Challenges and Opportunities for AI-enabled High-Performance Computing

Artificial intelligence and high‑performance computing are entering a new era defined not by individual workloads, but by end‑to‑end workflows that tightly integrate simulation, data, learning, and inference. For system architects, this shift fundamentally changes the optimization target: AI is no longer an external accelerator, but a latency‑sensitive component of the application critical path. As a result, communication, data movement, and coordination increasingly dominate system efficiency alongside raw compute capability. The next generation of systems will be defined by tight coupling and macroheterogeneity, integrating purpose‑built compute, memory, and network partitions into a coherent execution environment. Realizing this model introduces new challenges in orchestration, predictability, and performance evaluation, where workflows and not devices become the unit of optimization. AI rack‑scale architectures can play a decisive role in this paradigm by delivering dense, high‑bandwidth AI capability as a first‑class architectural element, provided they retain the flexibility to adapt to rapidly evolving AI models and emerging agentic execution patterns. Grounding these architectures in open, interoperable standards is essential for enabling portability, extensibility, and a sustainable ecosystem. This keynote outlines the architectural trends, challenges, and opportunities shaping AI‑enabled HPC systems for the coming decade.