FMS 2026: AIC Showcased Context Memory and AI Inference Storage Platforms

AIC, a player in enterprise storage and server solutions, exhibited at FMS: the Future of Memory and Storage 2026 from August 4–6 at the Santa Clara Convention Center.This year AIC presents the storage platforms behind AI’s newest infrastructure tier — context memory, the flash layer that holds precomputed model context for GPU clusters to retrieve rather than regenerate.

AI inference is redrawing the memory hierarchy. KV cache — the precomputed context behind every model response — has outgrown GPU memory, and operators increasingly serve it from shared NVMe flash instead of recomputing it on GPUs. Nvidia CMX defines this new tier of Ethernet-attached shared flash between GPU memory and capacity storage. At FMS 2026, AIC puts that tier on the show floor as deployable hardware.

AIC’s FMS 2026 lineup spans next-generation PCIe Gen6 platforms, dual-active high-availability architectures, DPU-accelerated JBOFs for NVMe-over-Fabrics, and CXL-ready all-flash servers — the building blocks of scalable, efficient AI data pipelines from ingest to inference. Together they reflect AIC’s focus on the full AI storage stack, from petabyte-scale capacity to the tiers closest to the GPU.

Featured onsite

• F2032-G6 — AIC’s 2U dual-active flash storage system for context memory. Built on PCIe Gen6 with a dual-port backplane and DPU-accelerated NVMe-over-Fabrics, so a controller failure becomes a path failover rather than an array rebuild. On display at AIC’s booth with an Nvidia CMX board for the context memory tier
• F2026-01-G5 — 2U PCIe Gen5 JBOF with dual-port NVMe throughout and support for multiple DPUs, serving disaggregated flash to GPU clusters over NVMe-over-Fabrics
• HA2026-HC — dual-active, high-availability storage platform for mission-critical workloads that cannot absorb a rebuild window
• SB201-SU — 2U all-flash NVMe server with dual Intel Xeon 6 processors and an E3, CXL-ready drive cage, for AI data ingest, preparation, and serving
• J4078-02-04X — 4U high-density SAS JBOD providing the capacity tier for backup, archive, and bulk data lakes behind the flash layer

AIC also displayed many other systems built for the next generation of storage architecture — engineered to reduce bottlenecks and meet the needs of increasingly demanding AI workloads.

AIC Across the Show Floor
AIC platforms also featured at booths across FMS 2026. AIC’s F2032-G6 will be on display at Micron’s Booth #107, populated with Micron 6600 ION 245TB SSDs. ScaleFlux has shown the AIC F2026-01-G5 at Booth #519. AIC systems also featured at Silicon Motion’s Booth #315.

H3 Platform, a leader in PCIe and composable infrastructure solutions, co-exhibited inside AIC Booth #419, showing a 3U GPU-accelerated memory and storage system developed in collaboration with AIC. The system pairs H3’s software-defined PCIe fabric orchestration with high-density NVMe to bring data closer to the GPU.

The FMS 2026 lineup reflects AIC’s ongoing collaborations with leaders in the storage, memory and AI ecosystem such as Nvidia, Micron, Sandisk, Solidigm, Kioxia, ScaleFlux, MangoBoost, Silicon Motion, Graid Technology, and many others. This close collaboration ensures AIC will continue to build systems ready to meet the demands of tomorrow’s AI workloads.

“We are always excited to attend FMS,” said Michael Liang, president and CEO, AIC. “AI storage is AIC’s key focus, and this show is a key opportunity for us to present our latest generation of systems, ready to meet the needs of increasingly intensive AI workloads. We have served this industry for over 30 years, and we look forward to continued growth alongside it.”

“In inference clusters, KV cache is the working set: every generated token depends on context that is expensive to recompute and cheap to retrieve,” said CT Sun, CTO, AIC. “This is the problem our systems are built to solve: keep that context on shared flash close to the GPUs, and keep it available through any failure. Get this tier right and inference keeps scaling. That is what we design for, and it is what makes these platforms ready for the next generation of AI workloads.”