ORICO has announced the FlashMem FM9800, a new PCIe 5.0 solid-state drive featuring dedicated cache space for running AI workloads locally on PCs. The SSD combines conventional storage with a separate AI caching area, offering configurations with 85GB or 128GB of reserved cache capacity.
The company detailed the upcoming drive in an official announcement published on November 13, 2026. Unlike conventional consumer SSDs that primarily store operating systems, applications and files, the FM9800 is designed to coordinate flash storage with system RAM and GPU memory during supported AI inference workloads.
The FlashMem FM9800 uses an M.2 2280 form factor, a PCIe 5.0 interface and 3D TLC NAND flash memory. ORICO plans to offer two capacity options, each divided into separate areas called Storage OS and AI Cache.

The 1TB model provides 768GB for conventional storage and 85GB for AI caching. The larger 2TB model increases those allocations to 1,638GB and 128GB, respectively.
ORICO has not specified how the remaining nominal capacity is allocated. The company has also confirmed that the drive includes a graphene thermal pad designed to dissipate heat from its controller and NAND components.
The main difference between the FM9800 and a conventional SSD lies in how its storage is managed. The Storage OS area handles everyday files, applications and operating-system data, while the AI Cache area accommodates information associated with local AI inference.
This includes selected AI model weights, context information and intermediate data generated while running models. ORICO says the two storage areas use separate wear-management strategies and read/write scheduling mechanisms to accommodate their different workloads.
The architecture is intended to work alongside AI storage middleware and the Herdsman local inference engine. Together, these components are designed to coordinate data placement across GPU VRAM, system RAM and the SSD’s dedicated cache.
For supported applications, information that does not need to remain continuously in faster memory could instead be held in flash storage. This may help systems manage larger AI workloads when available RAM or VRAM capacity becomes a constraint.
However, the additional AI cache should not be confused with an actual memory upgrade. NAND flash remains substantially slower than system RAM and dedicated GPU memory, so the FM9800 cannot directly replace either.
Its potential usefulness depends on software compatibility, the AI model being executed, memory requirements and how efficiently data can move between different storage and memory layers.
These distinctions matter for PC enthusiasts evaluating local AI hardware. Available GPU and VRAM monitoring tools can help identify memory usage during local AI workloads, but additional flash capacity does not automatically eliminate memory bottlenecks.
The FM9800 also arrives as PC memory configurations and changing RAM demand remain relevant considerations for consumers building or upgrading their systems. ORICO’s approach offers a different type of storage expansion rather than a substitute for installing more physical memory.
For gaming PCs, the FM9800’s PCIe 5.0 interface makes it relevant as a next-generation storage product, but ORICO has not demonstrated improvements in game loading times, frame rates or other gaming workloads. Its dedicated cache architecture is specifically intended for compatible local AI applications.
Performance claims also remain unverified. ORICO has not published final sequential read and write specifications, detailed endurance ratings or independent benchmark results demonstrating the benefits of its AI caching system.
The manufacturer says its official product specifications and testing documentation will provide additional performance, endurance and warranty information.
Pricing and a confirmed retail release date have not been announced. Until independent testing and software compatibility details become available, the FlashMem FM9800 remains an upcoming PCIe 5.0 SSD with a specialized AI caching design rather than a proven alternative to conventional high-performance consumer drives.
