Super Micro Computer Inc. used its Open Storage Summit on 13 August to spotlight how rapidly expanding “agentic” artificial-intelligence workloads are reshaping data-centre storage architectures. Executives from AMD, DataDirect Networks and Western Digital said real-time inference now requires high-speed, low-latency data fabrics that span multiple sites and clouds, turning virtually all data into ‘hot’ data ready for retrieval. Kevin Kang, senior manager at AMD, said the chipmaker is integrating its EPYC CPUs, GPUs and SmartNICs to move data directly to GPU memory and cut latency across AI pipelines. Balaji Venkateshwaran of DDN added that replacing AWS object storage with the company’s Infinia system cut latency enough to accelerate an AI application by a factor of 22, underscoring the payoff from targeted storage upgrades. Panelists warned that power consumption is becoming a gating factor as GPUs draw about ten times more electricity than CPUs, leaving less budget for storage unless efficiency improves. Liquid cooling, solid-state drive advances and closer hardware-software co-design were cited as priorities for keeping GPUs fed with data while containing energy use. The speakers said partnerships across vendors are essential to deliver end-to-end systems that can scale with surging AI demand.