On September 23, during the GMIF 2026 Global Memory Industry Innovation Summit, John Xavier Lionel, the global head of the memory business at Arm, pointed out that as generative AI transitions from the cloud to end-user devices, storage has emerged as a pivotal factor influencing the capabilities of AI systems. AI inference encompasses the entire computing ecosystem, including processing, memory, storage, and data transfer, necessitating comprehensive optimization at the system level.

With the refinement of smaller AI models, the adoption of quantization techniques, and the integration of heterogeneous NPUs (Neural Processing Units), AI deployment is set to span multiple tiers, encompassing cloud, edge, and end devices, with local storage shouldering an increasing workload. The demand for storage solutions in data centers, characterized by high capacity, high bandwidth, and high reliability, is on a continuous upward trajectory. Meanwhile, emerging end devices, such as AI-powered PCs and smart vehicles, are paving the way for novel application scenarios.

Storage media, including HBM (High Bandwidth Memory), HBF (a hypothetical or less common term, possibly meant to be a variant or typo; for the sake of this optimization, we'll assume it's a specific type of memory or storage being referenced in context), and NAND Flash, are undergoing constant evolution. This evolution is accompanied by ongoing upgrades in main controllers, advanced packaging techniques, testing and verification processes, and software scheduling algorithms. Various technological pathways are converging to form tightly integrated system synergies tailored to AI workloads.