Huawei has introduced a 122 TB SSD for AI infrastructure.

      Huawei has released the OceanDisk 1800 storage device with a capacity of 122.88 TB. The company uses Die-on-Board technology to increase storage density in environments with limited access to advanced 3D NAND.

      Huawei has introduced the OceanDisk 1800 SSD line for data centers and AI workloads. It includes models with 61.44 TB and 122.88 TB capacities. A 245 TB version is expected in the future. The drives are designed for infrastructure where data needs to be stored close to computations and quickly transferred to AI systems.

      The main detail here is not the record capacity, but the design of the storage device. Huawei employs Die-on-Board technology. In a conventional scheme, NAND chips are first placed in a separate package and then mounted on the SSD circuit board. In the OceanDisk 1800, they are placed directly on the board. This allows for a denser use of the internal space of the storage device.

      For Huawei, this approach makes practical sense. The company has been under U.S. restrictions since 2019 and does not have free access to the most advanced memory manufacturing and packaging technologies. Therefore, capacity must be increased not only through more advanced NAND chips but also through layout, packaging, and storage architecture.

      According to TrendForce, Die-on-Board can increase placement density by about 33%. This does not completely remove technological limitations, but it gives Huawei the opportunity to produce high-capacity drives on an accessible component base.

      The OceanDisk 1800 is not designed to serve as a standalone SSD in a server but to operate within dense storage systems. According to Blocks & Files, a Smart Disk Enclosure based on these drives provides up to 1.47 PB in a 2U chassis. In the OceanStor Pacific 9926 configuration, 36 SSDs of 122.88 TB each provide 4.42 PB of physical capacity in the same 2U. With data compression at a ratio of 2.5 to 1, the usable capacity can reach 11 PB.

      For AI data centers, this is an important metric. The higher the storage density, the fewer racks, supporting infrastructure, and cabling are required. Data can be kept closer to computations, and the storage system can be scaled without unnecessary expansion of the footprint.

      Large models, multimodal data, RAG scenarios, training, and inference require not only GPUs and HBM. They need fast and dense storage that does not create additional gaps between computations and data. If data is delivered slowly or stored too far from the computational boundary, expensive accelerators do not operate at full efficiency.