ScaleFlux Controller Chips Target AI Memory Bottleneck

ScaleFlux CEO Hao Zhong argues that GPU performance has grown over 1,000 times in eight years, but the data pipeline feeding those processors has not kept pace, leaving expensive compute waiting on memory and storage. This year, ScaleFlux launched controller chips for PCIe Gen 6 SSDs and Compute Express Link memory to address the second and third stages of the AI data pipeline. As workloads shift toward inference and agentic AI, Zhong says external DRAM and SSD capacity and bandwidth have become the central bottleneck in AI infrastructure planning.