Huawei’ s OceanStor M900 Context Memory Storage is aimed at solving a key cause of inefficiency when performing AI inference in hyperscale data centres
A shared memory space Huawei’ s answer to these problems is its new OceanStor M900 Context Memory Storage, which David Wang, Deputy Chairman of the Board and Rotating Chairman at Huawei, announced at HUAWEI CONNECT 2026.
Intended for AI inference in hyperscale data centres, it provides them with a fully shared memory space that offers PB-scale capacity and TB / s-level performance.
Global pooling and sharing with tiered storage is made possible by a highspeed UnifiedBus interconnect network and the solution boosts the available KV cache capacity per NPU from gigabytes to terabytes. The latter means that more context can be reused, shared and stored, significantly boosting the KV cache ratio. Huawei states that the OceanStor M900 is the industry’ s first architecture to integrate the CPU, network controller unit, and NAND controller unit. Through using native KV semantics to enable
“ IN THE AI ERA, THE VALUE OF DATA HAS UNDERGONE A QUALITATIVE LEAP, WHICH MEANS THE ROLE OF STORAGE MUST ALSO CHANGE ACCORDINGLY”
Dr Liang Ye Strategy and Transaction Partner Ernst & Young
a one-hop connection between the SuperPoD’ s NPU to SSDs, the company says that it has achieved a 90 % reduction in access latency and that this architecture doubles the inference cluster’ s token throughput and halves the time to first token( TTFT) during typical AI programming scenarios.
huawei. com 19