What we know
Phison has revealed a computational-storage SSD that includes a built-in Neural Processing Unit (NPU) and 64GB of memory. The device also features Mixed Media pSLC and QLC storage technologies, combining fast performance with high capacity.
Why it matters
By integrating an NPU and memory into the SSD, Phison’s design allows AI workloads to be processed closer to where data is stored, potentially reducing latency and data transfer bottlenecks. Running AI inference tasks directly on the storage device could improve efficiency and speed for applications that rely on machine learning models. This innovation reflects a growing trend toward computational storage to enhance performance in data-intensive AI environments.
What is still unknown
Details on the SSD’s accuracy, full list of supported AI frameworks and languages, and its commercial availability remain undisclosed. No independent tests of performance or energy savings have been reported. The timeline for broader release and pricing information are also not available.

