ScaleFlux and NVIDIA to Unveil New AI Memory Architectures at FMS 2026

The August 5 keynote marks the centerpiece of a significant technical presence for the Milpitas-based company, which has secured seven speaking slots across the three-day event. The sessions aim to address the widening gap between memory capacity and the demands of modern AI inference, a bottleneck that currently limits GPU utilization and drives up infrastructure costs.

Beyond the keynote, ScaleFlux experts are set to dissect the technical shift of flash storage into a primary memory tier. Presentations will cover critical industry friction points, including SSD controller design for heterogeneous workloads, the evolution of CXL memory for reliability, and strategies to bypass the high cost of High-Bandwidth Memory (HBM). By revisiting the classic "Five-Minute Rule" in the context of modern AI, the company intends to demonstrate how specific engineering approaches—such as managing write amplification without traditional data placement—can fundamentally change the economics of data centers.

These sessions reflect a broader industry consensus that traditional boundaries between memory and storage are no longer sustainable for the next generation of AI. Attendees at booth #519 will have the opportunity to examine the practical architectures behind these shifts, focusing on solutions that favor scalability and improved security in an era of increasingly intelligent storage devices.